Académique Documents
Professionnel Documents
Culture Documents
cover
HACMP System
Administration I: Planning and
Implementation
(Course Code AU54)
Student Notebook
ERC 6.0
Trademarks
IBM is a registered trademark of International Business Machines Corporation.
The following are trademarks of International Business Machines Corporation in the United
States, or other countries, or both:
AIX AIX5L DB2
DB2 Universal Database eServer Enterprise Storage Server
HACMP iSeries NetView
Notes POWER4+ POWER5
pSeries Redbooks RS/6000
SP Tivoli TotalStorage
WebSphere
Think is a trademark or registered trademark of Lenovo in the United States, other
countries, or both.
Java and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the
United States, other countries, or both.
Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft
Corporation in the United States, other countries, or both.
UNIX is a registered trademark of The Open Group in the United States and other
countries.
Other company, product and service names may be trademarks or service marks of others.
The information contained in this document has not been submitted to any formal IBM test and is distributed on an as is basis without
any warranty either express or implied. The use of this information or the implementation of any of these techniques is a customer
responsibility and depends on the customers ability to evaluate and integrate them into the customers operational environment. While
each item may have been reviewed by IBM for accuracy in a specific situation, there is no guarantee that the same or similar results will
result elsewhere. Customers attempting to adapt these techniques to their own environments do so at their own risk.
Copyright International Business Machines Corporation 1998, 2005. All rights reserved.
This document may not be reproduced in whole or in part without the prior written permission of IBM.
Note to U.S. Government Users Documentation related to restricted rights Use, duplication or disclosure is subject to restrictions
set forth in GSA ADP Schedule Contract with IBM Corp.
V3.1.0.1
Student Notebook
TOC Contents
Trademarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Agenda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix
TMK Trademarks
The reader should recognize that the following terms, which appear in the content of this
training document, are official trademarks of IBM or other companies:
IBM is a registered trademark of International Business Machines Corporation.
The following are trademarks of International Business Machines Corporation in the United
States, or other countries, or both:
AIX AIX 5L DB2
DB2 Universal Database eServer Enterprise Storage Server
HACMP iSeries NetView
Notes POWER4+ POWER5
pSeries Redbooks RS/6000
SP Tivoli TotalStorage
WebSphere
Think is a trademark or registered trademark of Lenovo in the United States, other
countries, or both.
Java and all Java-based trademarks are trademarks of Sun Microsystems, Inc. in the
United States, other countries, or both.
Microsoft, Windows, Windows NT, and the Windows logo are trademarks of Microsoft
Corporation in the United States, other countries, or both.
UNIX is a registered trademark of The Open Group in the United States and other
countries.
Other company, product and service names may be trademarks or service marks of others.
Duration: 5 days
Purpose
This course is designed to prepare students to install and configure a
highly available cluster using HACMP for AIX 5L.
Audience
The audience for this course is students who are experienced AIX
system administrators with TCP/IP networking and AIX LVM
experience who are responsible for the planning and installation of an
HACMP 5.3 cluster on an IBM Server pSeries server running AIX 5L
V5.2 or later.
Prerequisites
Students should ideally be qualified as IBM Certified Specialists -
pSeries AIX System Support or pSeries AIX System Administration
and in addition have TCP/ IP, LVM storage and disk hardware
implementation skills. These skills are addressed in the following
courses (or can be obtained through equivalent education and
experience):
AU16: AIX 5L System Administration II: Problem Determination
AP05: AIX 5L Essentials: AIX 5L TCP/IP Communications (CBT)
AU20: AIX Ver. 4 System Administration IV: Storage Management
AU07: AIX V4 Configuring TCP/IP and Accessing the Internet
Objectives
After completing this course, you should be able to:
Explain what high availability is.
Outline the capabilities of HACMP for AIX 5L.
Design and plan a highly available cluster.
Install and configure HACMP for AIX in the following modes of
operation:
- Single resource group on a primary node with standby node
- Two resource groups in a mutual takeover configuration
Configure resource group startup, fallover, and fallback policies
Perform basic system administration tasks for HACMP.
Curriculum relationship
This course should be taken before AU57
AU57: HACMP System Administration II: Administration and
Problem Determination
AU20: AIX Ver. 4 System Administration IV: Storage Management
pref Agenda
Day 1
Welcome
Unit 1 - Introduction to HACMP for AIX 5L
Exercise 1
Unit 2 - Shared Storage Considerations for High-Availability
Exercise 2
Day 2
Unit 3 - Networking Considerations for High-Availability
Unit 4 - Planning for Applications and Resource Groups
Unit 5 - HACMP Installation
Exercise 3
Exercise 4
Exercise 5
Day 3
Unit 6 - Initial Cluster Configuration
Exercise 6
Unit 7 - Basic HACMP Administration
Day 4
Exercise 7
Unit 8 - Events
Exercise 8
Unit 9 - Integrating NFS Into HACMP
Exercise 9
Day 5
Unit 10 - Problem Determination and Recovery
Exercise 10
Unit 11 - HACMP Migration
Exercise 11
Text highlighting
The following text highlighting conventions are used throughout this book:
Bold Identifies file names, file paths, directories, user names and
principals.
Italics Identifies links to Web sites, publication titles, is used where the
word or phrase is meant to stand out from the surrounding text,
and identifies parameters whose actual names or values are to
be supplied by the user.
Monospace Identifies attributes, variables, file listings, SMIT menus, code
examples of text similar to what you might see displayed,
examples of portions of program code similar to what you might
write as a programmer, and messages from the system.
Monospace bold Identifies commands, daemons, menu paths, and what the user
would enter in examples of commands and SMIT menus.
Welcome to:
AU54 - HACMP System Administration I:
Planning and Implementation
Figure Intro-1. AU54 - HACMP System Administration I: Planning and Implementation AU546.0
Notes:
Uempty
Course Objectives
Notes:
Course Agenda
Day 1
Welcome
Unit 1 - Introduction to HACMP for AIX 5L
Exercise 1
Unit 2 - Shared Storage Considerations for High-
Availability
Exercise 2
Notes:
Uempty
Course Agenda
Day 2
Unit 3 - Networking Considerations for High-
Availability
Unit 4 - Planning for Applications and Resource
Groups
Unit 5 - HACMP Installation
Exercise 3
Exercise 4
Exercise 5
Notes:
Course Agenda
Day 3
Unit 6 - Initial Cluster Configuration
Exercise 6
Unit 7 - Basic HACMP Administration
Notes:
Uempty
Course Agenda
Day 4
Exercise 7
Unit 8 - Events
Exercise 8
Unit 9 - Integrating NFS Into HACMP
Exercise 9
Notes:
Course Agenda
Day 5
Unit 10 - Problem Determination and Recovery
Exercise 10
Unit 11 - HACMP Migration
Exercise 11
Notes:
Uempty
Lab Exercises
Points to note :
Work as a team and split the workload.
Manuals are available online.
HACMP software has been loaded and may have
already been installed.
TCP/IP and LVM have not been configured.
Each lab must be completed successfully before
continuing on to the next lab, as each lab is a
prerequisite for the next one.
Any questions, ask your instructor.
Notes:
Notes:
Uempty
Course Overview Summary
Notes:
References
SC23-4864-06 HACMP for AIX 5L, Version 5.3: Concepts and
Facilities Guide
www.ibm.com/servers/eserver/pseries/library/hacmp_docs.html
HACMP for AIX 5L Manuals
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Explain what high availability is and why it is needed.
Outline the various options for implementing high
availability.
List the key considerations when designing and
implementing a high availability cluster.
Outline the features and benefits of HACMP for AIX 5L.
Describe the components of an HACMP for AIX 5L cluster.
Explain how HACMP for AIX 5L operates in typical cases.
Notes:
Objectives
In this unit we introduce the concept of high availability, examine why you might want to
implement a high availability solution, and compare high availability with some
alternative availability technologies.
HACMP terminology
In this course we will use the following terminology:
- HACMP will mean any version and release of the HACMP product.
- HACMP x will mean version x and any release of that version
- HACMP x.y will mean a specific version and release
- HACMP/ES will refer to HACMP 4 with enhanced scalability
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
So, What is High Availability
High Availability characteristics...
The masking or elimination of both planned and unplanned
downtime
The elimination of single points of failure (SPOFs)
Fault resilience, but NOT fault tolerance
Workload Fallover
WAN
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Other Unplanned
Downtme 14%
Planned Downtime
85%
Notes:
Downtime
You may be surprised to know that hardware component failure represents an
extremely small proportion of overall system downtime. By far, the largest single
contributor to system downtime is planned downtime.
Planned downtime
When you shut down a computer for the weekend, thats planned downtime. When you
stop the application to take a level 0 backup, thats planned downtime. When you
schedule a period of maintenance over the weekend, thats planned downtime.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Continuous
Availability
Continuous High
Operations Availability
Notes:
Uempty
Eliminating Single Points of Failure
Cluster Object Eliminated as a single point of failure by . . .
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Fault
Tolerant
High
Availability
Cluster
Stand-alone
Enhanced
Storage
Stand-alone
Notes:
Uempty
Stand-alone System
The stand-alone system
limited availability benefits:
9Journaled Filesystem
9Dynamic CPU Deallocation
9Service Processor
9Redundant Power
9Redundant Cooling
9ECC Memory
9Hot Swap Adapters
9Dynamic Kernel
9Disk mirroring
9Dynamic LPAR
Notes:
Drawback
Ultimately the stand-alone server suffers from one enormous availability drawback, that
is, it is a stand-alone system with no means of surviving failure of a critical system
component, for example, the operating system.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Node/Operating System
Network Adapter
Network
Application
Site Failures
Copyright IBM Corporation 2005
Notes:
Uempty
High-Availability Clusters (HACMP Base)
High-Availability Clustering
9Journaled Filesystem
9Dynamic CPU Deallocation
9Service Processor
9Redundant Power
9Redundant Cooling
9ECC Memory
9Hot Swap Adapters
9Dynamic Kernel
9Redundant Data Paths
9Data Mirroring
9Hot Swap Storage
9Redundant Power for Storage Arrays
9Redundant Cooling for Storage Arrays
9Hot Spare Storage
9Dual Disk Adapters
9Redundant nodes (operating system)
9Redundant Network Adapters
9Redundant Networks
9Application Monitoring
9Site Failure (limited)
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Drawback
The base product HACMP 5.2 and later only partially solves the site SPOF in the case
where data does not have to be replicated. This can be done with LVM mirroring using
SAN technology.
Uempty
Fault-Tolerant Computing
Fault-tolerant solutions should not fail:
Notes:
Fault-Tolerant
A fault-tolerant solution is specifically designed not to fail. Fault-tolerant solutions are
extremely expensive when compared with high availability solutions, and the few
manufacturers that produce fault-tolerant systems offer a limited range of solutions
(which means poor scalability). Price/performance is an issue with fault tolerance, as is
the proprietary nature of the hardware and associated operating system. Application
availability is also a consideration.
That said, fault-tolerant systems are designed to NOT fail. If you cannot suffer any
downtime, no matter how small, then a fault-tolerant solution may be for you.
Drawback
Even this solution does not account for site failures or software failures
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Solutions
Journaled Filesystem Redundant Data Paths Redundant Servers Lock Step CPUs
Dynamic CPU Data Mirroring Redundant Networks Hardened Operating
Deallocation Hot Swap Storage Redundant Network System
Service Processor Redundant Power for Adapters Redundant Memory
Availability Redundant Power Storage Arrays Heartbeat Monitoring Continuous Restart
benefits Redundant Cooling Redundant Cooling for Failure Detection
ECC Memory Storage Arrays Failure Diagnosis
Hot Swap Adapters Hot Spare Storage Automated Fallover
Dynamic Kernel Automated
Reintegration
Depends, but
Downtime Couple of days Couple of hours In theory, none!
typically 3 mins
Notes:
Uempty
So, What About Site Failure?
Limited distance (LVM mirroring and/or SAN): HACMP for AIX 5L
ESS PPRC
GLVM
HAGEO
Toronto Brussels
Data Replication
Copyright IBM Corporation 2005
Notes:
Limited distance
The base product HACMP 5.3 (and V5.2) allows you to create sites as long as you can
use LVM mirroring for redundancy. Using SAN technology you can get limited distance
support for site failures.
Extended distance
The HACMP/XD (Extended Distance) priced feature provides three distinct software
solutions for disaster recovery. These solutions enable an HACMP cluster to operate
over extended distances at two sites.
a. HACMP/XD for ESS PPRC increases data availability for IBM TotalStorage
Enterprise Storage Server (ESS) volumes that use Peer-to-Peer Remote Copy
(PPRC) to copy data to a remote site for disaster recovery purposes. HACMP/XD for
ESS PPRC takes advantage of the PPRC fallover/fallback functions and HACMP
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
HACMP XD
Uempty
Why Might I Need High Availability?
24X7 operations
60% of all large companies now operate round the clock (7x24)
Lose of Revenue $M
Downtime losses: 200
Per hour $50K (46%), $1M (8%), $330K (industry average) 150
Peak losses: 130,000 $US per minute (telephone network) 100
Loss of customer loyalty
50
Loss of customer confidence
0
No Disaster recovery:
50% of affected companies will never reopen
90% of affected companies are out of business in less than two
years
$
Copyright IBM Corporation 2005
Notes:
24X7 Operations
More and more companies need applications to be available all the time.
Downtime losses
These figures were extracted from a report produced by Tandem Corporation in 1995, so
they are a little out of date, however, the currency of the figures is not important since the
principles remain true today and in the future. Downtime costs money.
Many companies invest in a high-availability solution not only to protect against the
immediate revenue impact of failure, but also to avoid the embarrassment of bad press,
loss of customer confidence, and loyalty. In todays Internet-based economy, your nearest
competitor is only a URL away.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
No Disaster recovery
Figures for company failures following site disasters are taken from IBM Business
Recovery Services. Where a company relies on access to its data in order to do business,
and loses that data as a result of a site disaster (and has no means of recovery), the
chances of business failure are exceptionally high. It should be noted, however, that high
availability is NOT the same as disaster recovery. Disaster recovery deals with site failure,
and the partial or complete loss of computing resources in one location.
II
Uempty
Benefits of High-Availability Solutions
+ =
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
People
Systems
Management
High
availability Data
Networking Continuous
availability
Continuous
operation
Hardware
Environment
Software
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Lets Review
1. Which of the following is a characteristic of high availability?
a. High availability always requires specially designed hardware
components.
b. High availability solutions always require manual intervention to
ensure recovery following failover.
c. High availability solutions never require customization.
d. High availability solutions offer excellent price performance when
compared with Fault Tolerant solutions.
2. True or False?
High availability solutions never fail.
3. True or False?
A thorough design and detailed planning is required for all high availability
solutions.
4. True or False?
Hardware failures are the most common cause of downtime.
5. A proposed cluster with a two year life (for planning purposes) has a
vulnerability which is likely to occur twice per year at a cost of
$10,000 per occurrence. It costs $25,000 in additional hardware
costs to eliminate the vulnerability. Should the vulnerability be
eliminated?
a. Yes
b. No
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
In this topic, we take a look at the basic concepts and features of HACMP.
Uempty
IBM's HA Solution for AIX
HACMP for AIX 5L Characteristics
High Availability Cluster Multiprocessing
Based on cluster technology
Provides two environments:
Serial (High Availability): the process of ensuring an application is
available for use through the use of serially accessible shared data
and duplicated resources
Parallel (Cluster Multiprocessing): concurrent access to shared
data
Notes:
HACMP characteristics
IBMs HACMP product is a mature and robust technology for building a high-availability
solution. A high-availability solution based upon HACMP provides automated failure
detection, diagnosis, recovery and reintegration. With an appropriate application, HACMP
can also work in a concurrent access or parallel processing environment, thus offering
excellent horizontal scalability.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
clstrmgr clstrmgr
urce
Reso group
Shared Storage
Node Node
Fallover
Cluster is comprised of physical components (topology) and logical
components (resource groups and resources).
Notes:
Fundamental concepts
HACMP is based on the fundamental concepts of cluster, resource group, and cluster
manager (clstrmgr).
Cluster
A cluster is comprised basically of nodes, networks and network adapters. These objects
are referred to as Topology objects.
Resource group
A resource group is typically comprised of an application, network address, and volume
group using shared disks. These objects are referred to as Resource objects.
Uempty clstrmgr
The cluster managers together are the software components that communicate with each
other to control on which node a resource group is activated or where the resource group is
moved on a fallover based on parameters set up by the administrator. The clstrmgr runs on
all the nodes of the cluster
Here is a simple diagram of a two-node cluster, using shared disk, and providing fallover for
a single application.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
N tw
Ne
IP ork
on or
N
tw
e
-IP k
Communication
Interface
n
atio
m unic
Com Device
e No
Nod de
ter
C lus
Notes:
Topology Components
A clusters topology is the cluster, nodes (pSeries servers), networks (connections
between the nodes), the communication interfaces (for example, Ethernet or token-ring
network adapters) and the communication devices (/dev/tty for RS232 for example).
Nodes
In the context of HACMP, the term node means any IBM EServer pSeries system
which is a member of a high-availability cluster running HACMP.
Networks
Networks consist of IP and Non-IP networks. The non-IP networks ensure cluster
monitoring can be done if there is a total loss of IP communication. Non-IP networks are
strongly recommended to be configured in an HACMP.
Uempty Networks can also be logical or physical. Logical networks have been used with the IBM
SP environments when different frames were in different subnets but needed to be
treated as if they were in the same network for HACMP purposes.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Nodes
IBM Eserver pSeries High-end IBM
IBM
IBM
Midrange IBM
IBM
pSeries 670
IBM
pSeries
pSeries 590 pSeries 595
pSeries 660
pSeries
pSeries 650
IBM
pSeries 655
pSeries
server
pSeries 620
Models 6F0, 6F1
server
pSeries 630 IBM
IBM
pSeries 550
server
pSeries 570
IBM
pSeries 630
serv er
IBM
Model 6C4
pSeries 640
pSeries
Notes:
Supported nodes
As you can see, the range of systems that supports HACMP is, well - everything. The
only requirement is that the system should have at least four adapter slots spare (two
for network adapters and two for disk adapters). Any other adapters (for example,
graphics adapters) occupy additional slots. The internal Ethernet adapter fitted to most
entry level pSeries 600 servers cannot be included in the calculations. It should be
noted that even with four adapter slots free, there is still be a single point of failure as
the cluster is only able to accommodate a single TCP/IP local area network between the
nodes.
HACMP 5 works with pSeries and iSeries servers in a no-single-point-of-failure server
configuration. HACMP for AIX 5L supports the pSeries, Cluster 1600, RS/6000, and
iSeries models which are designed for server applications and which meet the minimum
requirements for internal memory, internal disk, and I/O slots. As of July 2005, the
Uempty following pSeries, iSeries, and RS/6000 models and their corresponding upgrades are
supported in HACMP 5.2 and later:
- PCI desktop systems: Models 140, 150, 170, 240, 260, and 270
- PCI deskside systems: Models E20, E30, F30, F40, F50, F80, 6F0, and 6F1
(pSeries 620). Note: Heartbeat network is only supported with serial port 3 or 4 on
the Model 6Fx.
- .. Entry systems: Models 25S, 250, and 25T
- .. Compact server systems: Models C10 and C20
- .. Desktop systems: Models 370, 380, 390, 397, and 39H
- .. Deskside systems: Models 570, 57F, 580, 58F, 58H, 590, 59H, 591, 595,
7028-6E1 (pSeries
- 610), 7029-6E3 (pSeries 615), 7025-6F1 (pSeries 620), and 7028-6E4 (pSeries
630)
- Rack systems: Models 98B, 98E, 98F, 990, 99E, 99F, 99J, 99K, B80, R10, R20,
R21, R24,
- R50, R5U, S70, S7A, H10, H50, H70, H80, M80, 7028-6C1 (pSeries 610),
7029-6C3 (pSeries 615), 7028-6C4 (pSeries 630), 6H0, 6H1, 6M1 (pSeries 660),
7039-651 (pSeries 655), and 7038-6M2 (pSeries 650), including models with PCI-X
Expansion Drawer (7311-D10 and 7311-D20)
- High-end servers: Models 7040-681 (pSeries 690), 7040-671 (pSeries 670),
including models with POWER4+ processors, and PCI-X Planar 10 slot I/O drawer
(7040-61D feature 6571)
POWER5 servers: pSeries models 9110-510, 9111-520, 9113-550, 9117-570,
9119-590, and 9119-595. iSeries models running AIX 9406-520, 9406-550, 9406-570,
9406-590, and 9406-595 with required APARs (see the Sales manual at
www.ibm.com/common/ssi)
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
LPAR support
There is also support for dynamically adding LPAR resources in AIX 5L V5.2 or later
LPAR environments to take advantage of Capacity Upgrade of Demand (CUoD).
HACMP 5.2 (and later) supports Virtual SCSI (VSCSI) and Virtual LAN (VLAN) on
POWER5 (IBM EServer p5 and IBM EServer i5). See
http://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/FLASH10390 for
more details.
Uempty
Supported Networking Environments
Ethernet Etherchannel
Token Ring PC
PC Server
Server Server
Traffic Flow Server
Server
FDDI
Server Server
Server Non -IP
Traffic Flow Server
RS232/422
TMSSA
heartbeat over disk
Notes:
Supported networks
HACMP 5.3 (and V5.2) supports client users on a LAN using TCP/IP. HACMP monitors
and performs IP address switching for the following TCP/IP-based communications
adapters on cluster nodes:
- Ethernet
- EtherChannel
- Token ring
- FDDI
- SP Switches
- ATM
- ATM LAN Emulation
HACMP also supports non-IP networks such as RS232/442, Target Mode SCSI
(TMSCSI), Target Mode SSA (TMSSA), and Enhanced Concurrent Mode.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
It is highly recommended to have both IP and non-IP networks defined to HACMP. For a
list of specific adapters you can consult the Sales Manual
Uempty
HACMP's Resource Components
Resources
Ap
pl
ou e
i ca
Gr lum
p
tio
Vo
n
Se
le
rv
Se
Fi tem
e
Ad rvic
r
s
dr e I Sy
es P
s
Resource Group
Nodes
Run Policies
Resources
Notes:
Resource Group
A resource group is a collection of resources treated as a unit along with what nodes
they can potentially be activated on and what policies the cluster manager should use to
decide which node to choose during startup, fallover, and fallback. A cluster may have
more than one resource group (usually one for each application), thus allowing for very
flexible configurations. Resource groups will be covered in more detail in Unit 4.
Resources
Resources are logical components that can be put into a resource group. Because they
are logical components, they can be moved without human intervention.
The resources shown in the visual are a typical set of resources used in resource
groups such as:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Solution Components
Not Just HACMP
FAStT
RS/6000
RS/6000
IBM
SAN
ESS RS/6000
Fibre
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
AIX contributions
AIX offers many availability benefits, for example, logical volume manager mirroring of all
data types, the journaled filesystem, and AIXs capability to execute an online JFS backup.
AIX has a dynamic, multithreaded kernel which allows configuration changes to the
operating system while the computer is running.
How often have you had to reboot an AIX system in order to configure a new hardware
device or extend a filesystem?
Uempty
Supported Shared Storage Environments
SSA Adapter
RS/6000
RS/6000
FAStT
SSA SAN
IBM
Maximum 25m
Host
System
SCSI
Controller
ESS Fibre
RS/6000
SCSI
Most HACMP clusters require shared storage. Disk technologies
which support multihost connections include: SCSI, SSA and FC
(with or without RAID).
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Local
Network
Notes:
Uempty
So What is HACMP Really?
An application which:
Controls where resource groups run
Monitors and reacts to events
Does fallover
Does reintegration
Provides tools for cluster wide configuration and
synchronization
Cluster Manager Subsystem
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The clstat command uses clinfo to display status via ASCII, Xwindow, or Web
browser interfaces.
- In HACMP 5, clcomdES allows the cluster managers to communicate in a secure
manner without using rsh and rhost files.
Uempty
Additional Features of HACMP
OLPW Configuration
smit via web assistant
ClstrmgrES verification
CSPOC Auto tests
DARE SNMP
Application
Tivoli
Monitoring
Netview
Notes:
Additional features
HACMP also has additional software to provide facilities for administration, testing,
remote monitoring, and verification:
- Application monitoring can be used to monitor the clusters applications and restart
them should they fail. Multiple monitors can be defined for an application including
monitoring the startup (only available in HACMP/ES and HACMP 5).
- Configuration changes can be made to the cluster while the cluster is running. This
facility is known as Dynamic Automatic Reconfiguration Event (or DARE for short).
Such dynamic changes in cluster configuration can also be emulated to test the
outcome.
- C-SPOC is a series of SMIT menus that allow AIX related cluster tasks to be
propagated across all nodes in the cluster. It includes an RG_Move facility which
allows a resource group to be placed offline or on another node without stopping the
cluster manager (RG_Move only available in HACMP 4.5 and later).
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-45
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Some Assembly Required
Customized
Pre-event scripts
Notes:
Customization required
HACMP is shipped with event scripts (Korn Shell scripts) which handle most failure
scenarios. If you have a requirement to customize some special fallover behavior, then this
is achieved through pre- and post-event scripts.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-47
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Lets Review
1. Which of the following are examples of topology components in
HACMP (select all that apply)?
a. Node
b. Network
c. Service IP label
d. Hard disk drive
2. True or False?
All clusters require shared disk for storage of HACMP log files.
3. True or False?
All nodes in an HACMP cluster must have roughly equivalent performance
characteristics
4. True or False?
HACMP requires rhost files.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-49
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
In this topic, we take a closer look at what HACMP actually does.
Uempty
Just What Does HACMP Do?
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-51
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
How the cluster responds to a failure depends on what has failed, what
the resource group's fallover policy is, and if there are any resource
group dependencies
The cluster's configuration is determined by the application's
requirements
Typically another equivalent component takes over duties of failed
component (for example, another node takes over from a failed node)
Notes:
Uempty
What Happens When a Problem Is Fixed?
?
How the cluster responds to the recovery of a failed component depends
on what has recovered, what the resource group's fallback policy is, and
what resource group dependencies there are
The cluster's configuration is determined by the application's
requirements
Cluster administrator may need to indicate/confirm that the fixed
component is approved for use
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-53
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
AA
A
Halifax Vancouver
Halifax Vancouver (no change)
One node is primary
A AA
Halifax Vancouver
Halifax Vancouver
Notes:
Standby
Standby configurations are configurations where one (or more) nodes have no
workload.
Uempty Drawbacks
- One node is not used (this is ideal for availability but not from a utilization
perspective).
- A second outage on fallback
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-55
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
A Vancouver returns
Halifax fails
Halifax Vancouver
A minimize downtime A
Halifax Vancouver Halifax Vancouver
A
Halifax Vancouver
Notes:
Minimize downtime
A resource group can be configured to not fall back to the primary node (or any other
higher priority node) when it recovers. This avoids the second outage which results
when the fallback occurs.
The cluster administrator can request that HACMP move the resource group back to
the higher priority node at an appropriate time or it can simply be left on its current node
indefinitely (an approach which calls into question the terms primary and secondary, but
which is actually quite a reasonable approach in many situations).
Uempty
Takeover: Two Sided (Mutual)
A B
Halifax fails Vancouver fails
Halifax Vancouver
AB
B A very common
Halifax Vancouver
Halifax Vancouver
Notes:
Takeover
Takeover configurations imply that there is workload on all nodes which may or may not
be under the control of HACMP but that a node can takeover the work of another node
in the cluster.
Mutual takeover
An extension of the primary node with a secondary node configuration is to have two
resource groups, one failing from right to left and the other failing from left to right. This
is referred to as mutual takeover.
Mutual takeover configurations are very popular configurations for HACMP since they
support two highly available applications at a cost which is not that much more than
would be required to run the two applications in separate stand-alone configurations.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-57
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Additional costs
Note that there are at least a few additional costs:
- Each cluster node probably needs to be somewhat larger than the stand-alone
nodes as they must each be capable of running both applications, possibly in a
slightly degraded mode, should one of the nodes fail.
- There may be additional software licenses required for the applications when they
run on their respective backup nodes (this is a potentially significant cost item which
is often forgotten in the early cluster planning stages).
- HACMP for AIX license fees.
- This is not intended to be an all inclusive list of additional costs.
Uempty
Concurrent: Multiple Active Nodes
Halifax, Regina and
Vancouver are all
running Application A,
each using a separate A A A
Service IP Address Regina Vancouver
Halifax
Notes:
Concurrent mode
HACMP also supports resource groups in which the application is active on multiple
nodes simultaneously. In such a resource group, all nodes run a copy of the application
and share simultaneous access to the disk. This style of cluster is often referred to as a
concurrent access cluster or concurrent access environment.
Service labels
Since the application is active on multiple nodes, each node has its own service IP
label. The client systems must be configured to randomly (or otherwise) select which
service IP address to communicate with, and be prepared to switch to another service
IP address should the one that theyre dealing with stop functioning (presumably,
because the node with the service IP address has failed). It is also possible to configure
an IP multiplexer between the clients and the cluster which redistributes the client
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-59
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
sessions to the cluster nodes, although care must be taken to ensure that the IP
multiplexer does not itself become a single point of failure.
How to choose
Whether this mode of operation can be used for your application is a function of the
application, not of HACMP. At present, Oracle RAC (Real Application Clusters) is the only
commercial application tested to work in a concurrent access environment.
Uempty
Fundamental HACMP Concepts
Topology: networking components
Resources: the entities which are being made highly available
Resource group: a collection of resources which HACMP
controls as a single unit
A given resource may only appear in at most one resource
group
Resource group policies:
Startup policy: which node the resource group is activated on
Fallover policy: determines target when there is a failure
Fallback policy: determines target when fallback occurs
Customization is the process of augmenting HACMP, typically
via implementing scripts which HACMP invokes at appropriate
times
Notes:
Terminology
A clear understanding of the above concepts and terms is important as they appear
over and over again both in the remainder of the course and throughout the HACMP
documentation, log files and smit screens.
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-61
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Points to Ponder
Resource groups:
Must be serviced by at least two nodes
Can have different policies
Can be migrated (manually or automatically) to rebalance loads
Clusters:
Must have at least one IP network and one non-IP network
Need not have any shared storage
Can have any combination of supported nodes *
Cluster may be split across two sites
May or may not require replicating data (HACMP/XD).
Applications:
Can be restarted via monitoring
Must be restartable via scripts
Notes:
Importance of planning
Planning, designing, configuring, testing and operating a successful HACMP cluster
requires considerable attention to detail. In fact, a careful methodical approach to all the
phases of the clusters lifecycle is probably the most important factor in determining the
ultimate success of the cluster.
Methodical approach
A careful methodical approach takes into account the relevant points above, and many
other issues which are discussed this week or which are discussed in the HACMP
documentation.
Uempty
HACMP 5.3 Limits
Cluster limits:
32 nodes in a cluster
64 resource groups per cluster
256 IP addresses known to HACMP (for example, service and
boot IP labels)
RSCT limit:
48 heartbeat rings
Notes:
Cluster Limits
HACMP 5.3 supports clusters with up to:
- 64 resource groups
- 256 interfaces
- 32 AIX5L/HACMP images (pSeries servers, SP nodes, RS/6000 systems, or
LPARs).
RSCT limit
HACMP uses the Topology Services component of RSCT for monitoring networks and
network interfaces. Topology Services organizes all the interfaces in the topology into
different heartbeat rings. The current version of RSCT Topology services has a limit of
48 heartbeat rings, which is usually sufficient to monitor networks and network
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-63
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
interfaces. Roughly speaking, the number of heartbeat rings is (usually) very close to
the number of network adapters on the node with the most adapters.
These limits do not tend to be a major concern in most clusters. Refer to the HACMP
documentation for additional information if you are planning a cluster which might
approach some of these limits.
Uempty
Things HACMP Does Not Do
TSM
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-65
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Zero downtime
An example of zero down time may be the intensive care room. Also HACMP is not
designed to handle many failures at once
Security issues
One security issue that is now addressed is the need to eliminate .rhost files. Also
there is better encryption possible with inter node communications but this may not be
enough for some security environments.
Unstable environments
The prime cause of problems with HACMP is poor design, planning, implementation,
and administration. If you have an unstable environment, with poorly trained
Uempty administrators, easy access to the root password, and a lack of change control,
HACMP is not the solution for you.
With HACMP, the only thing more expensive than employing a professional to plan,
design, install, configure, customize, and administer the cluster is employing an
amateur.
Other characteristics of poorly managed systems are:
- Lack of change control
- Failure to treat cluster as single entity
- Too many cooks
- Lack of documented operational procedures
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-67
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
A A
B B
Notes:
Goals
During this week you will design, plan, configure, customize, and administer a two-node
high-availability cluster running HACMP 5.3 on an IBM EServer pSeries server.
You will learn how to build a standby environment for one application as well as a mutual
takeover environment for two applications. In the mutual takeover environment each
system will eventually be running its own highly available application, and providing fallover
backup for the other system.
Uempty
Overview of the Implementation Process
Plan and configure AIX
Eliminate single points of failure
Storage (adapters, LVM volume group, filesystem)
Networks (IP interfaces, /etc/hosts, non-IP networks and
devices)
Applications start and stop scripts
Install the HACMP filesets and reboot
Configure the HACMP environment
Topology
Cluster, node names, HACMP IP and non-IP networks
Resources and Resource groups:
Identify name, nodes, policies
Resources: Application Server, service label, VG, filesystem
Synchronize then start HACMP
Notes:
Implementation process
Look at AIX environment
- For storage, plan for adapters and LVM components required for application
- For networks, plan and for communication interfaces, devices, name resolution via
/etc/hosts and service address for the application
- For application build start and stop script and test outside of the control of HACMP
Install the HACMP for AIX 5L software and reboot
Configure the topology and resource groups (and resources)
Synchronize, start and test
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-69
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
hints
Node A IP Label IP Address Netmask Node A IP Label IP Address Netmask
Service database 192.168.9.3 255.255.255.0 Service webserv 192.168.9.5 255.255.255.0
Boot nodeaboot 192.168.9.4 255.255.255.0 Boot nodebboot 192.168.9.6 255.255.255.0
Standby nodeastand 192.168.254.3 255.255.255.0 Standby nodebstand 192.168.254.3 255.255.255.0
Public Network
Diagram Label
Device
= a_tmssa
= /dev/tmssa1
tmssa network
Label
Device
= b_tmssa
= /dev/tmssa2
Be methodical
Resource Group httprg contains Resource Group databaserg contains
Volume Group = httpvg Volume Group = dbvg
hdisk2,hdisk8 hdisk3, hdisk4, hdisk5, hdisk6, hdisk7
Major # = 50 Major # = 51
JFS Log = httplvlog JFS Log = dblvlog
Logical Volume = httplv Logical Volume = dblv1, dblv2
FS Mount Point = /http FS Mount Point = /db, /dbdata
Notes:
Hints
Create a cluster diagram -- a picture is worth 10 thousand words (due to inflation a
thousand is not enough!).
Use the Online Planning Worksheets. They can be used without installing HACMP and
can be used to generate AND save HACMP configurations.
Try to reduce Single Points of Failure (SPOFs).
Always include a non IP network.
Mirror across power and buses.
Document test plan. HACMP also provides test scripts called auto test
Be methodical.
Execute the test plan prior to placing the cluster into production!
Uempty
Sources of HACMP Information
HACMP manuals come with the product READ THEM!
Sales Manual: www.ibm.com/common/ssi
/usr/es/sbin/cluster/release_notes
IBM courses:
HACMP Administration I: Planning and Implementation (AU54)
HACMP Administration II: Administration and Problem Determination
(AU57)**
HACMP Administration III: Problem Determination and Recovery (AU59)*
HACMP Administration IV: Master Class (AU56)*
Implementing High Availability on eServer Cluster 1600 (AU58)
IBM Geographic High Availability HAGEO (AU52)
IBM Web Site:
http://www-03.ibm.com/servers/eserver/pseries/ha/
Non-IBM sources (not endorsed by IBM but probably worth a look):
http://www.matilda.com/hacmp/
http://groups.yahoo.com/group/hacmp/
Notes:
Manuals on CD
The HACMP 5.3 manuals are:
SC23-4867-05 HACMP for AIX, Version 5.3: Master Glossary
SC23-4864-06 HACMP for AIX, Version 5.3: Concepts and Facilities Guide
SC23-4861-06 HACMP for AIX, Version 5.3: Planning and Installation Guide
SC23-4862-06 HACMP for AIX, Version 5.3: Administration Guide
SC23-5177-00 HACMP for AIX, Version 5.3: Troubleshooting Guide
SC23-4865-06 HACMP for AIX, Version 5.3: Programming Client Applications
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-71
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Checkpoint
1. True or False?
Resource Groups may be moved from node to node
2. True or False?
HACMP XD is a complete solution for building
geographically distributed clusters.
3. Which of the following capabilities does HACMP not
provide (select all that apply)?:
a. Time synchronization.
b. Automatic recovery from node and network adapter failure.
c. System Administration tasks unique to each node. Backup and
restoration.
d. Fallover of just a single resource group.
4. True or False?
Resource Groups may be moved from node to node.
5. True or False?
All nodes in a resource group must have equivalent
performance characteristics.
Notes:
Uempty
Unit Summary
Notes:
Copyright IBM Corp. 1998, 2005 Unit 1. Introduction to HACMP for AIX 5L 1-73
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
References
SC23-4867-05 HACMP for AIX: HACMP Master Glossary
SC23-4864-06 HACMP for AIX: Concepts and Facilities Guide
SC23-4861-06 HACMP for AIX: Planning and Installation Guide
SC23-4862-06 HACMP for AIX: Administration Guide
SC23-5177-00 HACMP for AIX: Troubleshooting Guide
http://www-03.ibm.com/servers/storage
http://www.redbooks.ibm.com
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Discuss the shared storage concepts that apply within an
HACMP cluster
Describe the capabilities of various disk technologies as
they related to HACMP clusters
Describe the shared storage related facilities of AIX and
how to use them in an HACMP cluster
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 1 Objectives:
Fundamental Shared Storage Concepts
After completing this topic, you should be able to:
Explain the distinction between shared storage and private
storage
Describe how shared storage is used within an HACMP
cluster
Discuss the importance of controlled access to an HACMP
cluster's shared storage
Describe how access to shared storage is controlled in an
HACMP cluster
Notes:
Uempty
What Is Shared Storage?
SCSI
disks
Node SSA Node
1 2
disks
ESS
storage
rootvg
rootvg rootvg
rootvg
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Non-concurrent access
In a non-concurrent access environment, the disks are owned by only one node at a
time. If the owner node fails, the cluster node with the next highest priority in the
resource group node list acquires ownership of the shared disks as part of fallover
processing. This ensures that the data stored on the disks remains accessible to client
applications.
In a non-concurrent access environment, a highly available application runs on only one
node for possibly quite extended periods of time. Only one disk connection is active at a
time and the shared storage is not shared in any real time sense. Rather, it is storage
which can be associated automatically (without human intervention) with whichever
node the application is currently running on. Non-concurrent access mode is also
sometimes called serial access mode, since only one node has access to the shared
storage at a time.
We will focus on non-concurrent shared storage in this unit.
Concurrent access
In concurrent access environments, the shared disks are actively connected to more
than one node simultaneously. Therefore, when a node fails, disk takeover is not
required. In this case, access to the shared storage must be controlled by some locking
mechanism in the application.
Uempty
What Is Private Storage?
SCSI
disks
Node SSA Node
1 2
disks
ESS
storage
rootvg
rootvg rootvg
rootvg
Notes:
Private storage
Private storage is, of course, accessible to only a single cluster node. It might be
physically located within each systems box or externally in a rack or even in an ESS
system. The key point is that private storage is not physically accessible from more than
one cluster node.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Why
The shared storage is physically connected to each node that the application might run
on. In a non-concurrent access environment, the application actually runs on only one
node at a time and modification or even access to the data from any other node during
this time could be catastrophic (the data could be corrupted in ways which take days or
even weeks to notice).
Uempty
Who Owns the Storage?
Node A B Node
1 2
ODM ODM
ODM ODM
C D
Notes:
Introduction
There are two mechanisms to control ownership of shared storage. Although these two
mechanisms do not seem to have formal names, in this unit, we refer to them as the:
- Reserve/release-based shared storage protection mechanism and the
- RSCT-based shared storage protection mechanism
We use the term protection rather than access control both because it is a bit shorter
and because it reminds us that the purpose of the mechanism is to protect the shared
storage.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Reserve/Release-based Protection
varyonvg C D
Notes:
Disk reservation
Reserve/release-based shared storage protection relies on the disk technology
supporting a mechanism called disk reservation. Disks which support this mechanism
can be, in effect, told to refuse to accept almost all commands from any node other than
the one which issued the reservation. AIXs LVM automatically issues a reservation
request for each disk in a volume group when the volume group is varied online by the
varyonvg command. The varyonvg command fails for any disks which are currently
reserved by other nodes. If it fails for enough disks, which it almost certainly does since
if one disk is reserved by another node, the others presumably are also, then the varyon
of the volume group fails.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
must be some mechanism to ensure that any meta-data changes made to the volume
group on the active node will be updated in the ODM on the inactive nodes in the
cluster. For example, if you change the size of a logical volume on the active node, the
other nodes ODMs will still list the logical volume at the original size. When an inactive
node is made active and if the volume group were varied on without updating the ODM,
the information in the ODM on the node and the VGDA on the disks would disagree.
This will cause problems.
When using reserve/release-based shared storage protection, HACMP provides a
mechanism called lazy update to update the ODM on the takeover node at the time of
fallover.
Lazy update
Lazy update works by using the volume group timestamp in the ODM. When HACMP
needs to varyon a volume group, it compares the ODM timestamp to the timestamp in
the VGDA. If the timestamps disagree, lazy update does an exportvg/importvg to
recreate the ODM on the node. If the timestamps agree, no extra steps are required.
It is, of course, possible to update the ODM on inactive nodes when the change to the
meta-data is made. In this way, extra time at fallover is avoided. The ODM can be
updated manually or you can use Cluster Single Point of Control (C-SPOC) which can
automate this task. Lazy update and the various options for updating ODM information
on inactive nodes are discussed in detail in a later unit in this course.
Uempty
Reserve/Release
Voluntary Disk Takeover
Node httpvg Node
A B varyonvg
1 2
ODM
ODM ODM
ODM
dbvg C D
varyonvg
Node A B Node
1 2
ODM
ODM ODM
ODM
Node2:
varyoffvg httpvg
dbvg C D
varyonvg
Notes:
Voluntary takeover
With reserve/release-based shared storage protection, HACMP passes volume groups
between nodes by issuing a varyoffvg command on one node and a varyonvg
command on the other node. The coordination of these commands (ensuring that the
varyoffvg is performed before the varyonvg) is the responsibility of HACMP.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Reserve/Release
Involuntary Disk Takeover
Node Node
A B varyonvg
1 2
ODM
ODM ODM
ODM
varyonvg C D
Node A B Node
varyonvg
1 2
ODM ODM
ODM
varyonvg C D
Notes:
Implications
Note that if the right node had not really failed then it would lose access to the shared
disks (rather abruptly) when the left node varied them on. Although this may seem
rather rude and generally unpleasant, it is far preferable to the alternative that both
nodes access and update the data on the disks (each believing that it is the only node
accessing and updating the data). An involuntary takeover isnt possible unless all
paths used by HACMP to communicate between the two nodes have been severed.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
hdisk0
Node Node hdisk1
A B hdisk2
1 varyonvg 2 hdisk3
ODM ODM hdisk4
hdisk5
hdisk6
hdisk7
C D hdisk8
varyonvg
hdisk9
Notes:
Uempty group ultimately succeeds depends on whether or not the LVM is able to find enough of
the volume groups physical volumes (and other factors like whether or not quorum
checking is enabled on the volume group).
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Node Node
1 passive varyon A B active varyon 2
ODM
ODM ODM
Notes:
Introduction
HACMP 5.x supports the new style of shared storage protection which relies on AIXs
RSCT component to coordinate the ownership of shared storage when using
enhanced concurrent volume groups in non-concurrent mode.
How it works
HACMP takes advantage of new parameters on the varyonvg and varyoffvg
commands related to a pair of new concepts called active and passive volume group
varyon states. A volume group being managed by RSCT-based shared storage
protection is varied online in the passive state on all cluster nodes which might need
access to the volume groups data. The volume group is varied online in the active state
by the particular cluster node which needs access to the volume groups data now (in
other words, the node which is running the application has the volume group varied on
Uempty in the active state). The LVM on each node prohibits accesses or updates to the volume
groups data unless the node has the volume group varied on in the active state.
It is the responsibility of the RSCT component to ensure that each volume group is
varied online in the active state on not more than one node. Since this mechanism does
not rely on any disk reservation mechanism, it is compatible with all disk technologies
supported by HACMP.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
ODM ODM
1. A decision is made to move
active
varyon C D
passive
varyon
httpvg from the right node to
dbvg
the left
Node passive httpvg passive Node
A B
1 varyon varyon 2
ODM ODM
3. Left node obtains active
active passive varyon of httpvg
varyon C D varyon
dbvg
(varyonvg)
Notes:
Uempty
RSCT-based
Involuntary Fast Disk Takeover
Node passive httpvg active
Node
1 varyon
A B
varyon 2 1. Right node fails
ODM ODM 2. Left node realizes that right
active passive node has failed
varyon C D varyon
dbvg
Node httpvg
Node
1 active A B passive 2
varyon varyon
ODM ODM 3. Left node obtains active
mode varyon of httpvg
active passive
C D
varyon dbvg varyon
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Requirements
Fast disk takeover is only used if all of the requirements listed above have been met.
Since RSCT is independent of disk technology, all disks supported by HACMP can be
used in an enhanced concurrent mode volume group.
Uempty
Fast Disk Takeover Additional Details
Fast disk takeover is faster than reserve/release-based
disk takeover
Ghost disks do not occur when fast disk takeover is
enabled
Since fast disk takeover is implemented by RSCT, it is
independent of the disk technology supported by
HACMP
The gsclvmd subsystem which uses group services
provides the protection
The distinction between active varyon and passive
varyon is private to each node (that is, it isn't
recorded anywhere on the shared disks)
Notes:
Introduction
As with any technology, the implications of using fast disk takeover must be properly
understood if the full benefits are to be experienced.
Note: If RSCT is not running, it is possible (although it takes some work), to manually
varyon an enhanced concurrent volume group to active mode, while it is varied on in
active mode on another node. While this is possible, it is an unlikely occurrence. This
small risk can easily be avoided by never varying on your shared volume groups
manually.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 2 Objectives:
Shared Disk Technology
After completing this topic, you should be able to:
Discuss the capabilities of various disk technologies in an
HACMP environment
Discuss the installation considerations of a selected disk
technology when combined with HACMP
Explain the issue of PVID consistency within an HACMP
cluster
Notes:
Uempty
SCSI Technology and HACMP
HACMP-related issues with SCSI disk architecture:
SCSI buses require termination at each end
In HACMP environments the terminators have to be external to
ensure that the bus is still terminated properly after a failed system
unit has been removed
SCSI buses are ID-based. All devices must have a unique ID
number
The default for all SCSI adapters at initial power-on is ID 7\
SCSI adapters on shared SCSI busses must be configured to
not use ID 7 in order to ensure that there isn't an ID conflict
when some other SCSI adapter powers on
Maximum 25m
Host Host
System System
T T SCSI
SCSI
Controller
5 6 Controller
SCSI 4 SCSI 3 SCSI 2 SCSI 1
Module Module Module Module
Notes:
SCSI termination
In HACMP environments, SCSI terminators must be external so that the bus is still
terminated after a failed system unit has been removed.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
SCSI Continued
Different SCSI bus types have different maximum cable lengths for
the buses (maximum is 25 meters for differential SCSI)
Four node limit
Certain SCSI subsystems support hot swappable drives.
SCSI cables are not hot pluggable (power must be turned off on all
devices attached to the SCSI bus before a SCSI cable connection
is made or severed).
Clusters using shared SCSI disks often experience ghost disks.
For additional information see:
IBM course AU20,
AIX 5L System Administration IV: Storage Management.
http://www.ibm.com/redbooks
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
SSA Technology and HACMP
SSA uses a loop technology which offers multiple data paths to disk
Number and type of adapter restrictions on each loop. For example:
SSA loops can support eight adapters per loop
(Maximum of eight HACMP nodes sharing SSA disks)
Adapters used in RAID mode are limited to two per loop
Shared SSA disks never appear as ghost disks
For additional information see:
IBM course AU20,
AIX 5L System Administration IV: Storage Management
Redbook, Understanding SSA Subsystems in Your Environment,
SG24-5750-00.
http://www.storage.ibm.com/hardsoft/products/ssa/docs/index.html
SSA Adapter
Notes:
Introduction
Serial Storage Architecture (SSA) enables you to minimize single points of failure and
achieve high availability in an HACMP environment.
You can use IBM 7133 and 7131-405 SSA disk subsystems as shared external disk
storage devices to provide concurrent access in an HACMP cluster configuration.
SSA is hot pluggable. Consequently, if you include SSA disks in a volume group using
LVM mirroring, you can replace a failed disk drive without powering off the entire
system.
SSA disk subsystems are less prone to power supply problems because they have
redundant power supplies.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
SSA information
Manuals
The following SSA manuals are available:
SA33-3285 Advanced Serial RAID Plus adapter; Users Guide and
Maintenance Information
SA33-3287 Advanced Serial RAID Plus adapter; Installation Guide
SA33-3286 Advanced Serial RAID Plus adapter; Technical
Reference
SA33-3278 7133 Models D40 and T40; Serial Disk Systems;
Operator Guide
SA33-3281 7133 Models D40 and T40; Serial Disk Systems;
Hardware Technical Information
GY33-0192 7133 Models D40 and T40; Serial Disk Systems;
Service Guide
GA33-3279 7133 Models D40; Installation Guide
GA33-3280 7133 Models T40; Installation Guide
AIX SSA subsystem
There is online information available on the AIX device drivers and subsystem interface
within AIX. Refer to the AIX 5L information pages and search for SSA. This brings up
sections describing the SSA device drivers, their IOCTL interfaces and how they are
used within an AIX system.
The IBM Hursley SSA Online Customer Support page is at:
http://www.storage.ibm.com/hardsoft/products/ssa
This contains adapter and disk drive microcode download packages as well as
technical data and product descriptions for IBM SSA products developed at Hursley.
The top-level IBM disk storage systems page is at:
http://www.storage.ibm.com/disk/index.html
This contains links to the complete range of IBM disk storage systems.
IBM training
AIX System Administration IV: Storage Management - AU20 / Q1320
For details on this class, or other IBM training classes, go to:
http://www.ibm.com
select Services & Solutions
select Training
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
A1
4 1 A1
SSA A2 A2 SSA
4 1
Adapter B1
B2
5
16
B1
B2
Adapter
5
16
13 7133 8
13 7133 8
A1 A1
SSA A2 12 9 A2 SSA
Adapter B1
B2
12 9 B1
B2
Adapter
SSA -1
SSA -2
Notes:
Example
There are many scenarios for cabling SSA disks. The example shown above is just one
possibility, and it is actually only half of the solution. You would need a second 7133
drawer and you would use the B-loop for it (this avoids the 7133 itself being a (unlikely
but real) single point of failure). Once the 7133s are configured, you would then mirror
the data across the 7133s.
More information
Refer to the SSA documentation or attend the AU20 course for more information.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
SSA Adapters
The capabilities of SSA adapters have improved over time:
Only 6215, 6219, 6225 and 6230 adapters support Target Mode SSA and RAID5
Only the 6230 adapter with 6235 Fast Write Cache Option feature code supports
enabling the write cache with HACMP
Compatible adapters:
6214 + 6216 or 6217 + 6218 or 6219 + 6215 + 6225 + 6230
For more information and microcode updates:
http://www.storage.ibm.com/hardsoft/products/ssa/
Features and functionality of otherwise identical adapters and drives can vary
depending upon the level of microcode installed on the devices so be careful!
FC Adapter Name Adapters/loop RAID5 TMSSA
RAID/JBOD Cache
6214 SSA 4-port adapter (MCA) -/2 N N
6216 Enhanced SSA 4-port adapter (MCA) -/8 N N
6217 SSA 4-port RAID adapter (MCA) 1/1 N N
6218 SSA 4-port adapter (PCI) 1/1 N N
6219 SSA Multi-Initiator/RAID EL Adapter (MCA) 2/8 Not for HA Y
6215 SSA Multi-Initiator/RAID EL Adapter (PCI) 2/8 Not for HA Y
6225 Advanced SerialRAID Adapter (PCI) 2/8 Not for HA Y
6230 Advanced SerialRaid Adapter Plus (PCI) 2/8 Not for HA Y
6230 + 6235 Fast Write Cache Option 2/8 Yes for HA Y
Note: AIX 5L V5.2 does not support the MCA 6214, 6216, 6217 and 6219 SSA adapters.
Copyright IBM Corporation 2005
Notes:
Uempty
ESS Technology
The ESS is an example of a smart storage device. ESS provides
highly available storage centrally managed by the storage manager.
The inner workings of the storage device are masked from AIX.
Basic implementations are transparent to HACMP.
The optional HACMP XD add-on can be used to coordinate the
fallover of ESS PPRC based remote data mirrors.
Up to 32 Connection ports +
Host Adapters online upgrades
HA HA HA HA HA HA HA HA HA HA HA HA HA HA HA HA
8,16 or 32 GB Cache
Switch Nonvolatile backup of write data
Dedicated 4-way
in cache
SMP Systems
Cache NVS 64 Internal disk paths
All disks can communicate at
Disk Adapters Cluster1 NVS Cache Cluster2 the same time
DA DA Hot-swap disks with redundant
spares + online upgrades
SSA Loops RAID5 providing performance
and availability
Physical Disks
partitioned into
Logical Volumes
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
ESS models
The ESS is available in two series:
- IBM TotalStorage DS6000 series:
This is a rack mountable enclosure which holds 16 drives. Using expansion
enclosures, the DS6000 can provide up to 38.4 TBytes of storage.
- IBM TotalStorage DS8000 series:
This is a cabinet sized enclosure. Using expansion frames, the DS8100 can provide
up to 115 TBytes of storage and the DS8300 can provide up to 192 TBytes of
storage.
For more information:
http://www.ibm.com/servers/storage/disk/index.html
Uempty
ESS Continued
Advanced features of the Storage unit may be supported by HACMP.
Subsystem Device Driver (SDD) is supported by HACMP
(with appropriate PTFs)
For additional information refer to:
IBM course AU20,
AIX 5L System Administration IV: Storage Management
Implementing the Enterprise Storage Server in Your Environment,
SG24-5420-01
IBM TotalStorage Enterprise Storage Server Model 800, SG24-6424-00.
http://www.ibm.com/redbooks
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
http://www.ibm.com/redbooks
Notes:
Uempty
Fibre Channel Continued
An example of a redundant fabric fibre channel implementation:
Node Node
1 2
Fibre channel
RAID
storage servers
Notes:
Planning is important. Think about availability issues when designing storage
configurations. In this figure, the fiber switches have been set up so the loss of a single
switch does not cause data access to be lost from either cluster node.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Node 1
A B
ODM
C D
Notes:
AIX assigns a unique physical volume ID (PVID) to each disk that it sees. This is stored in
the ODM and linked to a logical construct called a hdisk. hdisks are numbered sequentially
as discovered by the configuration manager (cfgmgr).
Uempty
hdisk Inconsistency
# lspv # lspv
hdisk0 000206238a9e74d7 rootvg hdisk0 000206238a9e74d7 rootvg
hdisk1 00020624ef3fafcc None A hdisk1 000206238beef264 rootvg
hdisk2 00206983880a1580 None B hdisk2 00206983880a1ed7 None C
hdisk3 00206983880a1ed7 None C hdisk3 00206983880a31a7 None D
hdisk4 00206983880a31a7 None D hdisk4 00020624ef3fafcc None A
hdisk5 00206983880a1580 None B
Node Node
A B
1 2
ODM ODM
C D
Neither HACMP nor AIX are affected by having a physical disk known
by different hdisk numbers on different systems.
Notes:
Example
In this example, the node on the right uses the names hdisk1, hdisk2, hdisk3 and
hdisk4 for the physical disks A, B, C and D respectively.
The shared storage subsystem has not been cabled consistently (put the disk
controllers in the same slots and cable each controller in a slot to the same disks as the
corresponding controller in the other node(s)). In addition, the right-hand node has an
extra disk which was configured before the shared disks.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The result is that the right hand node uses the names hdisk2, hdisk3, hdisk4 and hdisk5
for physical disks C, D, A and B respectively.
This is likely to be a source of constant confusion for the humans who have to
administer this cluster. On the other hand, assuming that the humans do not get
confused, both HACMP and AIX deal with this configuration properly.
Error example
An example of where human confusion could lead to apparent HACMP and AIX
confusion is if disks A and C are intended to represent datavg and disks B and D are
intended to represent stuffvg. If stuffvg is imported on both nodes by designating hdisk2
on both nodes then the left node will use disks B and D for stuffvg and the right node will
use disks A and C for stuffvg!
Note that neither HACMP nor AIX are actually confused. Rather, confusion on the part
of the human has resulted in apparent confusion on the part of HACMP and/or AIX. Had
the human done the right thing (used hdisk2 or hdisk4 when importing stuffvg on the left
node and used hdisk3 or hdisk5 when importing stuffvg on the right node), then neither
HACMP nor AIX would have appeared to be confused.
Uempty
Removing hdisk Inconsistencies
# rmdev -d -l hdisk1 ; rmdev -d -l hdisk2
# rmdev -d -l hdisk3 ; rmdev -d -l hdisk4
# mkdev -c disk -t 160mb -s scsi -p scsi0 -w 6,1 -d
# cfgmgr
# lspv # rmdev -d -l hdisk2 ; rmdev -d -l hdisk3
hdisk0 000206238a9e74d7 rootvg # rmdev -d -l hdisk4 ; rmdev -d -l hdisk5
hdisk2 00020624ef3fafcc None A # cfgmgr
hdisk3 00206983880a1580 None B # lspv
hdisk0 000206238a9e74d7 rootvg
hdisk4 00206983880a1ed7 None C
hdisk1 000206238beef264 rootvg
hdisk5 00206983880a31a7 None D
hdisk2 00020624ef3fafcc None A
hdisk3 00206983880a1580 None B
"Fake" hdisk1 will exist in a defined state and will not hdisk4 00206983880a1ed7 None C
appear in lspv output (use lscfg to see hdisk1). hdisk5 00206983880a31a7 None D
Node Node
A B
1 2
ODM ODM
C D
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-45
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Example
In the example, the problem was corrected by
- Recabling the disks
- Removing the disks from the ODM on both nodes
- Creating a fake disk on Node1 (mkdev command) so the disk numbering matches
Node2
- Running cfgmgr on both nodes
Uempty
Support for OEM Disks
HACMP lets you use either IBM disks or OEM disks
Treat an unknown disk type the same way as a known type
/etc/cluster/disktype.lst
/etc/cluster/lunreset.lst
/etc/cluster/conraid.dat
Use custom disk processing methods
Identifying ghost disks
Determining whether a disk reserve is being held by another node
in the cluster
Breaking a disk reserve
Making a disk available for use by another node
Enhanced concurrent VGs
Additional considerations
Notes:
Introduction
HACMP lets you use either physical storage disks manufactured by IBM or by an
Original Equipment Manufacturer (OEM) as part of a highly available infrastructure.
Depending on the type of OEM disk, custom methods allow you (or an OEM disk
vendor) either
- To tell HACMP that an unknown disk should be treated the same way as a known
and supported disk type, or
- To specify the custom methods that provide the low-level disk processing functions
supported by HACMP for that particular disk type
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-47
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
following three files can be edited to perform this configuration. (There is no SMIT menu
to edit these files.)
/etc/cluster/disktype.lst
This file is referenced by HACMP during disk takeover.
You can use this file to tell HACMP that it can process a particular type of disk the same
way it processes a disk type that it supports. The file contains a series of lines of the
following form:
<PdDvLn field of the hdisk><tab><supported disk type>
To determine the value of the PdDvLn field for a particular hdisk, enter the following
command:
# lsdev -Cc disk -l <hdisk name> -F PdDvLn
The known and supported disk types are:
Uempty depending on whether vendor or vendor plus product match was desired. Note the
use of padding of Vendor ID to 8 characters.
A sample /etc/cluster/lunreset.lst file, which contains comments, is provided.
/etc/cluster/conraid.dat
This file is referenced by HACMP during varyon of a concurrent volume group.
You can use this file to tell HACMP that a particular disk is a RAID disk that can be used
in classical concurrent mode. The file contains a list of disk types, one disk type per line.
The value of the Disk Type field for a particular hdisk is returned by the following
command:
# lsdev -Cc disk -l <hdisk name> -F type
Note: This file only applies to classical concurrent volume groups. Thus this file has no
effect in AIX 5L V5.3, which does not support classical concurrent VGs.
HACMP does not include a sample conraid.dat file. The file is referenced by the
/usr/sbin/cluster/events/utils/cl_raid_vg script, which does include some
comments.
Additional considerations
The previously described files in /etc/cluster are not modified by HACMP after they
have been configured and are not removed if the product is uninstalled. This ensures
that customized modifications are unaffected by the changes in HACMP. By default, the
files initially contain comments explaining their format and usage.
Keep in mind that the entries in these files are classified by disk type, not by the number
of disks of the same type. If there are several disks of the same type attached to a
cluster, there should be only one file entry for that disk type.
Finally, unlike other configuration information, HACMP does not automatically
propagate these files across nodes in a cluster. It is your responsibility to ensure that
these files contain the appropriate content on all cluster nodes. You can use the
HACMP File Collections facility to propagate this information to all cluster nodes.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-49
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
More information
For detailed information about configuring OEM disks for use with HACMP, see:
SC23-4861-06 HACMP for AIX, Version 5.3: Planning and Installation Guide
Appendix D: OEM Disk, Volume Group, and Filesystems
Accommodation
Uempty
Lets Review Topic 2
1. Which of the following disk technologies are
supported by HACMP?
a. SCSI.
b. SSA.
c. FC.
d. All of the above.
2. True or False?
SSA disk subsystems can support RAID5 (cache-enabled) with HACMP.
3. True or False?
Compatibility must be checked when using different SSA adapters in the
same loop.
4. True or False?
No special considerations are required when using SAN based storage
units (DS8000, ESS, EMC HDS, and so forth).
5. True or False?
hdisk numbers must map to the same PVIDs across an entire HACMP
cluster.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-51
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-53
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 3 Objectives:
Shared Storage from the AIX Perspective
After completing this topic, you should be able to:
Discuss how LVM aids cluster availability
Describe the quorum issues associated with HACMP
Set up LVM for maximum availability
Configure a new shared volume group, filesystem, and
jfslog
Figure 2-32. Topic 3 Objectives: Shared Storage from the AIX Perspective AU546.0
Notes:
This topic discusses shared storage from the AIX perspective.
Uempty
Logical Volume Manager (LVM) Review
LVM is one of the major enhancements that AIX brings to traditional UNIX disk
management. LVM's capabilities are exploited by HACMP
Physical disk volumes are:
Organized into volume groups
Identified by a unique physical volume ID (PVID)
Divided into physical partitions which are mapped to logical partitions in
logical volumes
Applications (such as file systems and databases) use logical volumes
Physical Logical
Partitions Partitions
PVID
hdisk1
Volume
Group
Notes:
LVM review
The set of operating system commands, library subroutines and other tools that allow
the user to establish and control logical volume storage is called the logical volume
manager.
LVM controls disk resources by mapping data between a simple and flexible logical
view of storage space and the actual physical disks. The logical volume manager does
this by using a layer of device driver code that runs above the traditional physical device
drivers.
Logical volumes
This logical view of the disk storage, which is called a logical volume (LV), is provided to
applications and is independent of the underlying physical disk structure. The LV is
made up of logical partitions.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-55
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Physical volumes
Each individual disk drive is called a physical volume (PV). It has a physical volume ID
(PVID) associated with it and an AIX name, usually /dev/hdiskx (where x is a
unique integer on the system). Every physical volume in use belongs to a volume group
(VG) unless it is being used as a raw storage device or a readily available spare (often
called a hot spare). Each physical volume is divided into physical partitions (PPs) of a
fixed size for that physical volume. A logical partition is mapped to one or more physical
partitions.
Volume groups
Physical volumes and their associated logical volumes are grouped into volume group.
Operating system files are stored in the rootvg volume group. Application data are
usually stored in one or more additional volume groups.
Uempty
LVM Relationships
LVM manages the components of the disk subsystem. Applications talk to the
disks through LVM.
This example shows an application writing to a filesystem which has its LVs
mirrored in a volume group physically residing on separate hdisks.
LVM
Physical Logical
Partitions Partitions
Volume Group
write to
/filesystem
Mirrored
Logical
Volume
Application
Notes:
LVM relationships
An application writes to a file system. A file system provides the directory structure and
is used to map the application data to logical partitions of a logical volume. Because
there is a LVM, the application is isolated from the physical disks. The LVM can be
configured to map a logical partition to up to three physical partitions and have each
physical partition (copy) reside on a different disk.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-57
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
LVM Mirroring
LVM mirroring has some key advantages over other types of mirroring:
LVM
Physical Logical
Partitions
Volume Group
Partitions
write to
/filesystem
Mirrored
Logical
Volume
Application
Notes:
Introduction
Reliable storage is essential for a highly available cluster. LVM mirroring is one option to
achieve this. Other options are a hardware RAID disk array configured in RAID-5 mode
or some other solution which provides sufficient redundancy such as an external
storage subsystem like the ESS (DS6000/DS8000), EMC, and so forth.
LVM mirroring
Some of the features of LVM mirroring are:
- Data can be mirrored on three disks rather than having just two copies of data. This
provides higher availability in the case of multiple failures, but does require more
disks for the three copies.
- The disks used in the physical volumes could be of mixed attachment types.
Uempty - Instead of entire disks, individual logical volumes are mirrored. This provides
somewhat more flexibility in how the mirrors are organized. It also allows for an odd
number of disks to be used and provides protection for disk failures when more than
one disk is used.
- The disks can be configured so that mirrored pairs are in separate sites or in
different power domains. In this case, after a total power failure on one site,
operations can continue using the disks on the other site that still has power. No
information is displayed on the physical location of each disk when mirrored logical
volumes are being created, unlike when creating RAID 1 or RAID 0+1 arrays, so
allocating disks on different sites requires considerable care and attention.
- Mirrored pairs can be on different adapters.
- Read performance is good for short length operations as data can be read from
either of two disks, so the one with the shortest queue of commands can be used.
Write performance requires a write to two disks.
- Extra mirrored copies can be created and then split off for backup purposes.
- Data can be striped across several mirrored disks, an approach which avoids hot
spots caused by excessive activity on a few disks by distributing the I/O operations
across all the member disks.
- There are parameters such as Mirror Write Consistency, Scheduling Policy, and
Enable Write Verify which can help maximize performance and reliability.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-59
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
1 create shared volume group Name the VG something meaningful like shared_vg1
Notes:
Introduction
This visual describes a procedure for creating a shared and mirrored file system. There
is an easier-to-use method provided by an HACMP facility called C-SPOC which is
discussed later in the course. The C-SPOC method cannot be used until the HACMP
clusters topology and at least one resource group have been configured.
The procedure described in the visual permits the creation of shared file systems before
performing any HACMP related configuration (an approach favored by some cluster
configurators).
Detailed procedure
Here are the steps in somewhat more detail:
a. Use the smit mkvg fastpath to create the volume group.
Uempty b. Make sure that the volume group is created with the Activate volume group
AUTOMATICALLY at system restart parameter set to no (or use smit chvg to
set it to no). This gives HACMP control over when the volume group is brought
online. It is also necessary to prevent, for example, a backup node from attempting
to online the volume group at a point in time when it is already online on a primary
node.
c. Use the smit mklv fastpath to create a logical volume for the jfslog with the
parameters indicated in the figure above (make sure that you specify a type of jfslog
or AIX ignores the logical volume and creates a new one (which is not mirrored)
when you create file system below).
d. Use the logform command to initialize a logical volume for use as a JFS log device.
Note: The only intended use for the logform command is to initialize a JFS log
logical volume as a JFS log device. The SMIT interface for creating a JFS and the
crfs command allow only one JFS log device per volume group.
e. Use the smit mklv fastpath again to create a logical volume for the file system with
the parameters indicated in the figure above.
f. Use the smit crjfslv fastpath (not crjfs) to create a JFS file system in the now
existing logical volume.
g. Verify by mounting the file system and using the lsvg command. Notice that if
copies were set to 2, then the number for PPs should be twice the number for LPs
and that if you specified separate physical volumes then the values for PVs should
be 2 (the number of copies).
The procedure for creating a JFS2 file system is quite similar although there are a few
differences:
- The type of the JFS2 log logical volume should be jfs2log.
- The logform command requires an additional parameter to cause it to create a
JFS2 log:
# logform -V jfs2log <lvname>
- The type of the JFS2 file system logical volume should be jfs2.
- The fastpath for creating a JFS2 file system in an existing logical volume is
smit crjfs2lvstd.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-61
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
VGDA
ODM ODM
#1 #2 #3 #4 #5
mkvg unmount cfgmgr varyoffvg Start HACMP
chvg varyoffvg importvg
mklv (log) chvg
logform
mklv (data)
crfs
Notes:
Introduction
The steps to add a shared volume groups are:
0) Ensure consistent hdisks names
1) Create a new VG and its contents
2) Varyoff VG on Node1
3) Import VG on Node2 and set VG characteristics correctly
4) Varyoff VG on Node2
5) Start HACMP
Please note that the slide presents only a high-level view of the commands required to
perform these steps. More details are provided below.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-63
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
5. Start HACMP
a. Restart HACMP, which varies on the VG and mounts the filesystems and you can
then resume processing.
C-SPOC
Fortunately, there is an easier way.
These steps will be done automatically if the cluster is active and C-SPOC is used.
Otherwise, you can use the commands listed here in the notes.
Unfortunately, we are not looking at the easier way until we get to the C-SPOC unit.
Uempty
Quorum Checking
AIX performs quorum checking on volume groups in order to ensure that the
volume group remains consistent
The quorum rules are intended to ensure that structural changes to the volume
group (for example, adding or deleting a logical volume) are consistent across an
arbitrary number of varyon-varyoff cycles
Overriding the quorum rules can be VERY DANGEROUS!
100% VGDA's
varyonvg >50% VGDA's or if MISSINGPV_VARYON=TRUE
>50% VGDA's
Notes:
Quorum
Quorum is the check used by the LVM at the volume group level to resolve possible
data conflicts and to prevent data corruption. Quorum is a method by which >50% of
VGDAs must be available in a volume group before any LVM actions can continue.
Note: For a VG with 3 or more disks, there is one copy of the VGDA on each disk. For a
one disk VG, there are two copies of the VGDA. For a two disk VG, the first disk has two
copies and the second has one copy of the VGDA. The VGDA is identical for all disks in
the VG.
Quorum is especially important in a HA cluster. If LVM can varyon a volume group with
half or less of the disks, it might be possible for two nodes to varyon the same VG at the
same time, using different subsets of the disks in the VG. This is a very bad situation
which we will discuss in the next visual.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-65
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Normally LVM verifies quorum when the VG is varied on and continuously while the VG
is varied on.
Uempty
Disabling Quorum Checking
Cluster designers or implementers are often tempted to disable quorum
checking. Although often necessary/desirable, there are risks if quorum
is disabled or if a volume group varyon is forced:
It may be possible for each side of a two-node cluster to have different parts
of the same volume group vary'd online
It is possible that volume group structural changes are made to one part of
the volume group which are inconsistent with a different set of structural
changes which are made to another part of the volume group
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-67
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
half of all the mirrored logical volumes as this leads to a phenomenon known as data
divergence.
Trust me... you do not want to experience the results of data divergence!
Sometimes it may necessary to disable quorum in a cluster. In this case, take care that
you do not end up with data divergence. The primary strategy for avoiding data
divergence is to avoid partitioned clusters although careful design of the clusters
shared storage is also important.
Uempty
Eliminating Quorum Problems
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-69
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
- If you are mirrored across two disk cabinets, consider a quorum buster disk to
prevent loss if quorum if you lose access to one cabinet. This is discussed in the
next visual
Uempty
The Quorum Buster
In some conditions, loss of quorum may lead to an unplanned
system downtime. The quorum buster can help eliminate this
possibility.
sharedvg
Notes:
Introduction
In an environment where you have a shared volume group with mirrored logical
volume(s), you could have a situation where you lose half the disks. This would cause
quorum to fail even though all the data is still available through the mirrored copy.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-71
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
controller or a power supply with one of the two halves then the loss of that disk
controller or power supply results in the loss of the half and of the quorum buster which,
in turn, results in the loss of quorum and the volume group goes offline.
Uempty
HACMP Forced Varyon
HACMP 5.x provides a per resource group forced varyon:
Each resource group has a flag which can be set to cause HACMP to
perform a careful forced varyon of the resource group's VGs
If normal varyonvg fails and this flag is set:
HACMP verifies that at least one complete copy of each logical volume is
available
If verification succeeds, HACMP forces the volume group online
This is not a complete and perfect solution to quorum issues:
If the cluster is partitioned then the rest of the volume group might still
be online on a node in the other partition
HACMP 4.5 introduced forced varyon for all shared VGs:
Still available in HACMP 5.x
If the HACMP_MIRROR_VARYON environment variable is set to TRUE,
forced varyon is enabled for all shared VGs in the cluster
If set, HACMP_MIRROR_VARYON overrides the per resource group
forced varyon flag
Copyright IBM Corporation 2005
Notes:
varyonvg -f
AIX 5L provides the ability to varyon a volume group if a quorum of disks is not
available. This is called forced varyon. The varyonvg -f command allows a volume
group to be made active that does not currently have a quorum of available disks. All
disks that cannot be brought to an active state will be put in a removed state. At least
one disk must be available for use in the volume group.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-73
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Recommendations for Forced Varyon
Before enabling HACMP's forced varyon feature for a volume
group or the HACMP_MIRROR_VARYON variable for the entire
cluster, ensure that:
The affected volume groups are mirrored across disk enclosures
The affected volume groups are set to super-strict allocation
There are redundant heartbeat networks between all nodes
Administrative policies are in effect to prevent volume group structural
changes when the cluster is running degraded (that is, failed over or
with disks missing)
Notes:
More information
Refer to the HACMP for AIX Administration Guide Version 5.3 (chapter 14) and the
HACMP for AIX Planning and Installation Guide Version 5.3 (chapter 5) for more
information about forced varyon and quorum issues.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-75
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Introduction
Defining an enhanced concurrent volume group allows the LVM to use RSCT to
manage varyonvg and varyoffvg processing.
Concurrent access
In a concurrent access environment, all the nodes will varyon the volume group.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-77
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Active varyon
If using enhanced concurrent volume groups in a non-concurrence access
environment, only one node will varyon the VG in active mode, allowing full access.
Passive varyon
Other nodes will varyon the VG in passive mode. In passive mode, only very limited
operations are allowed on the volume group.
Uempty what makes fast disk takeover faster than traditional disk-reservation based volume
group takeover.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-79
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
lsvg vg_name
ON ACTIVE NODE
halifax # lsvg ecmvg
VOLUME GROUP: ecmvg VG IDENTIFIER: 0009314700004c00000000f
e2eaa2d6d
VG STATE: active PP SIZE: 8 MB
VG PERMISSION: read/write TOTAL PPs: 537 (4296 MB)
... ... ...
Concurrent: Enhanced-Capable Auto-Concurrent: Disabled
ON PASSIVE NODE:
toronto # lsvg ecmvg
VOLUME GROUP: ecmvg VG IDENTIFIER: 0009314700004c00000000f
e2eaa2d6d
VG STATE: active PP SIZE: 8 MB
VG PERMISSION: passive-only TOTAL PPs: 537 (4296 MB)
... ... ...
Concurrent: Enhanced-Capable Auto-Concurrent: Disabled
Notes:
Introduction
The VG PERMISSION field in the output of lsvg shows if a volume group is varied on in
active or passive mode.
Uempty
LVM and HACMP Considerations
Following these simple guidelines helps keep the configuration easier
to administer:
All LVM constructs must have unique names in the cluster.
For example, httplv, httploglv, httpfs and httpvg
Mirror or otherwise provide redundancy for critical logical volumes.
Don't forget the jfslog
If it isn't worth mirroring then consider deleting it now rather than
having to wait to lose the data when the wrong disk fails someday
Even data which is truly temporary is worth mirroring as it avoids
an application crash when the wrong disk fails
RAID-5 and ESS-based storage are alternative ways to provide
redundancy
The VG major device numbers should be the same
Mandatory for clusters exporting NFS filesystems, but it is a good
habit for any cluster
Shared data on internal disks is a bad idea
Focus on the elimination of single points of failure
Copyright IBM Corporation 2005
Notes:
Unique names
Since your LVM definitions are used on multiple nodes in the cluster, you must make
sure that the names created on one node are not in use on another node. The safest
way to do this is generally to explicitly create and name each entity (do not forget to
explicitly create, name and format (using logform) the jfslog logical volumes).
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-81
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
space could cause an outage if the wrong disk fails. In order to avoid the outage,
mirror the scratch space!
The mirrorvg command provides an easy way to mirror all the logical volumes on a
given volume group. This same functionality may also be accomplished manually if you
execute the mklvcopy command for each individual logical volume in a volume group.
Uempty
Support for OEM Volume Groups
OEM volume groups can be used with HACMP
HACMP 5.3 automatically detects and provides the
methods for Veritas volume groups (VxVM)
Configuring custom volume group processing methods
using SMIT
List volume groups of a specified type
List physical and logical disks in a volume group
Bring a volume group online and offline
Determine a volume group status
Verify volume groups configuration
Provide a location of log files and other debugging information.
View using the AIX 5L snap -e command.
Limitations and more information
Notes:
Introduction
You can configure OEM volume groups in AIX 5L and use HACMP as an IBM High
Availability solution to manage such volume groups.
Note: Different OEMs may use different terminology to refer to similar constructs. For
example, the Veritas Volume Manage (VxVM) term Disk Group is analogous to the AIX
LVM term Volume Group. We will use the term volume groups to refer to OEM and
Veritas volume groups.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-83
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
automatically. After you add Veritas volume groups to HACMP resource groups, you
can select the methods for the volume groups from the pick lists in HACMP SMIT
menus for OEM volume groups support.
Note: Veritas Foundation Suite is also referred to as Veritas Storage Foundation (VSF).
Additional considerations
The custom volume group processing methods that you specify for a particular OEM
volume group is added to the local node only. This information is not propagated to
other nodes; you must copy this custom volume group processing method to each node
manually. Alternatively, you can use the HACMP File Collections facility to make the
disk, volume, and file system methods available on all nodes.
Uempty
Support for OEM File Systems
OEM file systems can be used with HACMP
HACMP 5.3 automatically detects and provides the
methods for Veritas file systems (VxFS)
Configuring custom file systems processing methods using
SMIT
List file systems of a specified type
List volume groups hosting a specified file system type
Bring a file system online and offline
Determine a file systems status
Verify file system configuration
Provide a location of log files and other debugging information.
View using the AIX 5L snap -e command.
Limitations and more information
Notes:
Introduction
You can configure OEM file systems in AIX 5L and use HACMP as an IBM High
Availability solution to manage such file systems.
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-85
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Additional considerations
The custom file system processing methods that you specify for a particular OEM file
system is added to the local node only. This information is not propagated to other
nodes; you must copy this custom file system processing method to each node
manually. Alternatively, you can use the HACMP File Collections facility to make the
disk, volume, and filesystem methods available on all nodes.
Uempty
Checkpoint
1. True or False?
Lazy update keeps VGDA constructs in sync between cluster nodes
(reserve/release-based shared storage protection)
2. Which of the following commands will bring a volume group
online?
a. getvtg <vgname>
b. mountvg <vgname>
c. attachvg <vgname>
d. varyonvg <vgname>
3. True or False?
Quorum should always be disabled on shared volume groups.
4. True or False?
filesystem and logical volume attributes cannot be changed while the
cluster is operational.
5. True or False?
An enhanced concurrent volume group is required for the heartbeat over
disk feature.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 2. Shared Storage Considerations for High Availability 2-87
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Summary
Access to shared storage must be controlled
Nonconcurrent (serial) access
Reserve/release-based protection:
Slower and may result in ghost disks
RSCT-based protection (fast disk takeover):
Faster, no ghost disks, and some risk of partitioned cluster in the event of
communication failure
Careful planning is needed for both methods of shared storage protection to
prevent fallover due to communication failures
Concurrent access
Access must be managed by the parallel application
HACMP supports several disk technologies
Must be well understood to eliminate single points of failure
Shared storage should be protected with redundancy
LVM mirroring
LVM configuration options must be understood to ensure availability
LVM quorum checking and forced varyon must be understood to ensure
availability
Hardware RAID
Copyright IBM Corporation 2005
Notes:
References
SC23-4867-05 HACMP for AIX: HACMP Master Glossary
SC23-4864-06 HACMP for AIX: Concepts and Facilities Guide
SC23-4861-06 HACMP for AIX: Planning and Installation Guide
SC23-4862-06 HACMP for AIX: Administration Guide
SC23-5177-00 HACMP for AIX: Troubleshooting Guide
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Discuss how HACMP uses networks
Describe the HACMP networking terminology
Explain and set up IP Address Takeover (IPAT)
Configure an IP network for HACMP
Configure a non-IP network
Explain how client systems are likely to be affected by
failure recovery
Minimize the impact of failure recovery on client systems
Notes:
Unit objectives
This unit discusses networking in the context of HACMP.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 1 Objectives:
How HACMP Uses Networks
After completing this topic, you should be able to:
Explain how HACMP uses networks to:
Provide client access to the cluster
Detect failures
Diagnose failures
Communicate with other nodes in the cluster
Explain why a non-IP network is an essential part of any
HACMP cluster
Describe what a persistent node IP label is and what they
are typically used for
Provide an overview of IP Address Takeover
Notes:
Topic 1 objectives
This topic explores how HACMP uses networks. It includes an introduction to the
HACMP concept of IP Address Takeover (IPAT). A more detailed look at IPAT appears
in a later section.
Uempty
How Does HACMP Use Networks?
HACMP uses networks to:
1. Provide clients with highly available access to the cluster's
applications
2. Detect and diagnose node, network and network interface
card (NIC) failures
3. Communicate with other HACMP daemons on other nodes in
the cluster
1
2
RSCT RSCT
3
clcomd clcomd
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
administrators. Just being able to detect node, network and network interface card
(NIC) failures imposes a number of requirements on how the networks are designed.
Being able to distinguish between certain failures, for example the failure of a network
and the failure of a node, imposes yet more requirements on the network design.
Reliable Scalable Cluster Technology (RSCT) provides facilities for monitoring node
membership; network interface and communication interface health; and event
notification, synchronization and coordination via reliable messaging.
Uempty
Providing HA Client Access to the Cluster
Providing clients with highly available access to the cluster's
applications requires:
Multiple NICs per network per node
(Possibly) multiple networks per node
Careful network design and implementation all the way out to
the client's systems
Notes:
Network as SPOF
The network itself is, of course, a single point of failure since the failure of the network
will disrupt the users ability to communicate with the cluster. The probability of this
SPOF being an issue can be reduced by careful network design, an approach which is
often considered sufficient.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
What HACMP Detects and Diagnoses
Remember, HACMP only handles the following
failures directly:
Network Interface Card (NIC) failure
Node failure
Network failure
IP network
en0 en1 en0 en1
non-IP network
bondar hudson
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Other failures
HACMP uses AIX features to respond to other failures (for example, the loss of a
volume group can trigger a fallover), but HACMP is not directly involved in detecting
these other types of failures. HACMP really is only involved in NIC, node, and network
failures.
Uempty
Heartbeat Packets
HACMP sends heartbeat packets across networks
Heartbeat packets are sent and received by every NIC
This is sufficient to detect all NIC, node and network failures
Heartbeat packets are not acknowledged
bondar hudson
Notes:
Heartbeat packets
HACMPs primary monitoring mechanism is to send heartbeat packets. The cluster
sends heartbeat packets from every NIC and to every NIC and to and from non-IP
devices.
Heartbeating pattern
In a typical two-node cluster with two NICs on the network, the heartbeat packets are
sent in the pair-wise fashion shown above. The pattern gets more complicated when
the cluster gets larger as HACMP uses a pattern which is intended to satisfy three
requirements:
- That each NIC be used to send heartbeat packets (to verify that the NIC is capable
of sending packets).
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
- That heartbeat packets be sent to each NIC (to verify that the NIC is capable of
receiving heartbeat packets).
- That no more heartbeat packets are sent than are necessary to achieve the first two
requirements (to minimize the load on the network).
The details of how HACMP satisfies the third requirement are discussed in a later unit.
Detecting failures
Heartbeat packets are not acknowledged. Instead, each node knows what the
heartbeat pattern is and simply expects to receive appropriate heartbeat packets on
appropriate network interfaces. Noticing that the expected heartbeat packets have
stopped arriving is sufficient to detect failures.
Uempty
Failure Detection versus Failure Diagnosis
Failure Detection is realizing that something is wrong
For example, realizing that packets have stopped flowing
between bondar's en1 and hudson's en1
Failure Diagnosis is figuring out what is wrong
For example, figuring out that bondar's en1 NIC has failed
HACMP uses RSCT to do both detection and diagnosis
bondar hudson
Notes:
Diagnosis
The heartbeat patterns just discussed are sufficient to detect a failure in the sense of
realizing that something is wrong. They are not sufficient to diagnose a failure in the
sense of figuring out exactly what is broken.
For example, if the en1 interface on the bondar node fails as in the visual above, bondar
stops receiving heartbeat packets via its en1 interface, and hudson stops receiving
heartbeat packets via its en1 interface. Bondar and hudson both realize that something
has failed, but neither of them have enough information to determine what has failed.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Failure Diagnosis
When a failure is detected, HACMP (RSCT topology services)
uses specially crafted packet transmission patterns to
determine (that is, diagnose) the actual failure by ruling out
other alternatives
Example:
1. RSCT on bondar notices that heartbeat packets are no longer
arriving via en1 and notifies hudson (which has also noticed
that heartbeat packets are no longer arriving via its en1)
2. RSCT on both nodes send diagnostic packets between
various combinations of NICs (including out via one NIC and
back in via another NIC on the same node)
3. The nodes soon realize that all packets involving bondar's
en1 are vanishing but packets involving hudson's en1 are
being received
4. DIAGNOSIS: bondar's en1 has failed.
Notes:
Uempty
What If All Heartbeat Packets Stop?
A node might notice that heartbeat packets are no longer arriving
on any NIC.
In the configuration below, it's impossible for either node to
distinguish between failure of the network and failure of the other
node.
Each node concludes that the other node is down!
bondar hudson
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Partitioned cluster
Since each node is, in fact, still very alive, the result is that the applications are now
running simultaneously on both nodes. If the shared disks are also online to both nodes,
then the result could be a quite massive data corruption problem. This situation is called
a partitioned cluster. It is, clearly, a situation which must be avoided.
Note that essentially equivalent situations can occur in larger clusters. For example, a
five-node cluster might become split into a group of two nodes and a group of three
nodes. Each group concludes that the other group has failed entirely and takes what it
believes to be appropriate action. The result is almost certainly very unpleasant.
Uempty
All Clusters REQUIRE a Non-IP Network!
There must be more than one network to distinguish between:
Failure of the other node
Failure of a network
There must be a non-IP network to distinguish between:
Failure of the other node's IP subsystem
Total failure of the other node
Therefore, ALL CLUSTERS SHOULD HAVE A NON-IP
NETWORK!!!
non-IP network
bondar hudson
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
An Important Implementation Detail
HACMP must ensure that heartbeats are sent out via all NICs and
know which NIC is used.
If a node has multiple NICs on the same logical subnet then AIX
can rotate which NIC is used to send packets to the network.
Therefore, each NIC on each physical IP network on any given
node must have an IP address on a different logical subnet.
non-IP network
bondar hudson
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Node recovery
In contrast, a node is not considered to have recovered until the HACMP daemons have
been started up on the node. This allows the node to be rebooted and otherwise
exercised as part of the repair process without HACMP declaring failures and/or
performing reintegration while the repair action is going on.
The reintegration of a component might trigger quite significant actions. For example, if
a node is reintegrated which has a high priority within a resource group then, depending
on how the resource group is configured, the resource group might fallback.
Uempty
IP Address Takeover (IPAT)
Each highly available application is likely to require its own IP
address (called a service IP address)
This service IP address would usually be placed in the
application's resource group
HACMP would then be responsible for ensuring that the service
IP address was available on the node currently responsible for
the resource group
IP rvice
el
NF
Sy
Fil tem
lab
S
Se
e
s
ex
ro e
NF po
G lum
S
up
mo rts
Vo
un r ve r
ts io n Se
licat
App
Resou
rc
Grou e
p
Notes:
Service IP address
Most highly available applications work best, from the users perspective, if the
applications IP address never changes. This capability is provided by HACMP using a
feature called IP Address Takeover. An IP address is selected which is associated with
the application. This IP address is called a service IP address because it is used to
deliver a service to the use. It is placed in the applications resource group. HACMP
then ensures that the service IP address is kept available on whichever node the
resource group is currently on. The process of moving an IP address to another NIC or
to a NIC on another node is called IP address takeover (IPAT).
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
for which the client software can be configured to check multiple IP addresses when it is
looking for the server.
Also, IPAT is not supported for resource groups configured for concurrent access as the
application in such a resource group is active on all the nodes which are currently up.
Consequently, clients of a concurrent access resource group must be capable of finding
their server by checking multiple IP addresses.
Uempty
IPAT After a Node Failure
If the application's current node fails, HACMP moves the
application's resource group to the other node.
If IPAT is configured for the resource group then the application's
service IP address is associated with the application's new node.
192.168.25.12
192.168.25.12
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
192.168.25.12
192.168.25.12
Notes:
Users perspective
Since existing TCP/IP sessions generally recover cleanly from this sort of
failure/move-IP-address operation, users might not even notice the outage if they dont
happen to be interacting with the application at the time of the failure.
Uempty
Lets Review Topic 1
1. How does HACMP use networks (select all which apply)?
a. Provide client systems with highly available access to the
cluster's applications
b. Detect failures
c. Diagnose failures
d. Communicate between cluster nodes
e. Monitor network performance
2. Using information from RSCT, HACMP only directly handles
three types of failures: ______________, ______________,
______________.
3. True or False?
Heartbeat packets must be acknowledged or a failure is assumed to have
occurred.
4. True or False?
Clusters should include a non-IP network.
5. True or False?
Each NIC on each physical IP network on each node is required to have an
IP address on a different logical subnet.
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 2 Objectives:
HACMP Concepts and Configuration Rules
After completing this topic, you should be able to:
List the networking technologies supported by HACMP
Describe the purpose of public and private HACMP
networks
Describe the topology components and their naming rules
Define key networking related HACMP terms
Describe the basic HACMP network configuration rules
Figure 3-17. Topic 2 Objectives: HACMP Concepts and Configuration Rules AU546.0
Notes:
Topic 2 objectives
This section will explore HACMP networking concepts, terms and configuration rules in
more detail.
Uempty
HACMP Networking Support
Supported IP networking technologies:
Ethernet
All speeds
Not the IEEE 802.3 frame type which uses et0, et1 ...
FDDI
Token-Ring
ATM and ATM LAN Emulation
Etherchannel
SP Switch 1 and SP Switch 2
Supported non-IP network technologies:
Heartbeat over Disks (diskhb)
Requires Enhanced Concurrent Volume Group and
HACMP 5.x
RS232/RS422 (rs232)
Target Mode SSA (tmssa)
Target Mode SCSI (tmscsi)
Notes:
Supported IP networks
HACMP supports all of the popular IP networking technologies (and a few which are
possibly not quite as popular). Note that the IEEE 802.3 Ethernet frame type is not
supported.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Network Types
HACMP categorizes all networks:
IP networks:
Notes:
IP networks
As mentioned before, IP networks are used by HACMP for:
- HACMP heartbeat (failure detection and diagnosis)
- Communications between HACMP daemons on different nodes
- Client network traffic
IP network attribute
The default for this attribute is public. Oracle uses the private network attribute
setting to select networks for Oracle inter-node communications. This attribute is not
used by HACMP itself. See the HACMP for AIX: Planning and Installation Guide for
more information.
If the VIO server has only a single physical interface on a network then a failure of
that physical interface will be detected by HACMP. However, that failure will isolate
the node from the network.
Non-IP networks
HACMP uses non-IP networks for:
- Alternative non-IP path for HACMP heartbeat and messaging
- Differentiates between node/network failure
- Eliminates IP as a single point of failure
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Interface
Network
Interface
Interface
Network
Interface
Card
Card
Card
Card
ne non
tw -IP
Network
or
k
Communication Device
Serial Serial
Port non IP - rs232 Port
Notes:
Terminology
HACMP has quite a few special terms that are used repeatedly throughout the
documentation and the HACMP smit screens. Over the next few visuals we will discuss
some of the network related terminology in detail.
- node
An IBM Eserver pSeries system operating within an HACMP cluster.
- node name
The name of a node from HACMPs perspective.
- IP label
For TCP/IP networks, the name specified in the /etc/hosts file or by the Domain
Name Service for a specific IP address.
Uempty HACMP nodes will have multiple NICs, and thus multiple IP labels, but only one
hostname. Well look at the relationship between hostname, node name and IP
labels in the next visual.
In HACMP, IP labels are either service IP labels or non-service IP labels. Well
discuss this distinction in the next few visuals.
- IP network
A network which uses the TCP/IP family of protocols.
- non-IP network or serial network
A point-to-point network which does not rely on the TCP/IP family of protocols.
- communication interface
A network connection onto an IP network (slightly better definition coming shortly).
- communication device
A port or device connecting a node to a non-IP network (slightly better definition
coming shortly).
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
IP labels
# netstat -i
Name Mtu Network Address Ipkts Ierrs Opkts Oerrs Coll
lo0 16896 link#1 5338 0 5345 0 0
lo0 16896 127 localhost 5338 0 5345 0 0
lo0 16896 ::1 5338 0 5345 0 0
tr0 1500 link#2 0.4.ac.49.35.58 76884 0 61951 0 0
tr0 1500 192.168.1 vancouver-if1 76884 0 61951 0 0
tr1 1492 link#3 0.4.ac.48.22.f4 476 0 451 13 0
tr1 1492 192.168.2 vancouver-if2 476 0 451 13 0
tr2 1492 link#4 0.4.ac.4d.37.4e 5667 0 4500 0 0
tr2 1492 195.16.20 db-app-svc 5667 0 4500 0 0
Notes:
Hostname
Each node within an HACMP cluster is an IBM Eserver pSeries system. It almost
certainly has a hostname associated with it that was assigned when the machine was
first installed onto the network. For example, a hypothetical machine might have been
given the name gastown.
Uempty IP labels
Each IP address used by an HACMP cluster almost certainly has an IP label associated
with it. In non-HACMP systems, it is not unusual for the systems only IP label to be the
same as the systems hostname. This is rarely a good naming convention within an
HACMP cluster as there are just so many IP labels to deal with, and having to pick
which one gets a name that is the same as a nodes hostname is a pointless exercise.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Communication Device:
A communication device refers to one end of a point-to-point
non-IP network connection, such as /dev/tty1, /dev/hdisk1 or
/dev/tmssa1.
Communication Adapter:
A communication adapter is an X.25 adapter used to support a
Highly Available Communication Link.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty Note: In earlier versions of HACMP, the terms boot IP label and boot IP address were
used to refer to what is now being called non-service IP label / address. The older terms
still appear in a few places in the HACMP 5.x documentation.
Service Interface
A communications interface configured with a service IP label / address (either by alias
or by replacement).
Non-Service Interface
A communications interface not configured with a service IP label / address. Used as a
backup for a service IP label / address.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
General rules
The primary purpose of these rules is to ensure that cluster heartbeating can reliably
monitor NICs, networks and nodes.
There are two basic approaches:
- Heartbeating over IP interfaces
- Heartbeating over IP aliases
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
netmon
netmon, the network monitor portion of RSCT Topology Services, allows you to create a
configuration file that specifies additional network addresses to which ICMP ECHO
requests can be sent as an additional way to monitor interfaces. netmon is outside the
scope of this class. Please see the HACMP for AIX: Planning and Installation Guide for
information on using netmon.
Unmonitorable NICs
One final point: if no other mechanism has been configured into the cluster, HACMP
attempts to monitor an otherwise unmonitorable (is that a word?) NIC by checking to
see if packets are arriving and being sent via the interface. This approach is not
sufficiently robust to be relied upon - use heartbeating via IP aliases or netmon to get
the job done right.
Uempty
IP Network Configuration Examples
IP Address IP Address Valid configuration?
node1 node2 Assume subnet mask = 255.255.255.0
192.168.5.1 192.168.5.2 Both node2 interfaces are on same subnet; they
192.168.6.1 192.168.5.3 cannot be monitored, unless you use heartbeating
over IP alias or netmon.
192.168.5.1 192.168.5.2 OK, but NICs are a single point of failure. Normally
you should have at least two NICs per network on
each node.
192.168.5.1 192.168.5.2 OK
192.168.6.1 192.168.6.2
192.168.5.1 192.168.5.2 OK, but 2nd NIC on both nodes do not have a
192.168.6.1 192.168.7.1 common subnet with another node; they cannot
be monitored, unless you use heartbeating over IP
alias or netmon.
192.168.5.1 192.168.5.2 OK, but 3rd and 4th interfaces on node1 do not
192.168.6.1 192.168.6.2 have a common subnet with another node; they
192.168.7.1 cannot be monitored, unless you use heartbeating
192.168.8.1 over IP alias or netmon.
Notes:
Examples
The visual shows some IP network examples.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
net_rs232_04
Notes:
Non-IP networks
Non-IP networks are point-to-point. That is each connection between two nodes is
considered a network and a separate non-IP network label for it is created in HACMP.
For example, the visual shows four RS232 networks, in a ring configuration, connecting
four nodes to provide full cluster non-IP connectivity.
Uempty Rules
The rules for non-IP networks are considerably simpler than the rules for IP networks
although they are just as important.
The basic rule is that you MUST configure enough non-IP networks to provide a non-IP
communication path, possibly via intermediate nodes, between every pair of nodes in
the cluster. In other words, every node must have a non-IP network connection to at
least one other node. Additional communication paths, such as the ring or mesh
topologies discussed in the visual, provide more robustness.
In addition, there are some considerations based on the type of non-IP network you are
using.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-45
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
second that a disk or disk subsystem can support. However, if you choose to use a
disk that has significant I/O load, increase the value for the timeout parameter for the
disk heartbeat network.
- When SDD is installed and the enhanced concurrent volume group is associated
with an active vpath device, ensure that the disk heartbeating communication device
is defined to use the /dev/vpath device (rather than the associated /dev/hdisk
device).
- If a shared volume group is mirrored, at least one disk in each mirror should be used
for disk heartbeating.
This is particularly important if you plan to set the forced varyon option for a
resource group.
Uempty
Persistent Node IP Labels
An IP label associated with a particular node
Useful for administrative purposes:
Provides highly available IP address associated with a
particular node
Allows external monitoring tools (for example, Tivoli) and
administrative scripts to reach a particular node
Assigned, via IP aliasing, after node synchronization, to a
communications interface on the node
HACMP will strive to keep the persistent node IP label available on
that node -- never moved to another node.
Maximum of one persistent node IP label per network per node
Persistent node IP labels must adhere to subnet rules:
Persistent node IP labels must not be in any non-service
interface subnet
Notes:
Rationale
In earlier releases of HACMP, the only way to guarantee that a known IP address would
always be available on each node for administrative purposes was to configure a
separate network which was never used for IPAT. Such a configuration limits the
usefulness of the administrative network.
Persistent IP labels
Starting in the HACMP 4.5 (classic and ES) release, users may now configure
persistent node IP labels. These are IP aliases that are configured on a node and kept
available as long as at least one communication interface remains active on the
associated network.
Persistent IP labels can be used with IP address takeover (IPAT).
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-47
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
If HACMP is not up
If HACMP is not up on the node then the persistent node IP label is still aliased to a
communication interface although the failure of the underlying communication interface
will, of course, cause the persistent node IP label to become unavailable.
Uempty
Lets Review Topic 2
1. True or False?
Clusters must always be configured with a private IP network for
HACMP communication.
2. Which of the following are true statements about
communication interfaces (select all that apply)?
a. Has an IP address assigned to it using the AIX TCP/IP smit screens
b. Might have more than one IP address associated with it
c. Sometimes but not always used to communicate with clients
d. Always used to communicate with clients
3. True or False?
Persistent node IP labels are not supported for IPAT via IP replacement.
4. True or False?
There are no exceptions to the rule that, on each node, each NIC on the
same LAN must have an IP address in a different subnet
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-49
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-51
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 3 Objectives:
Implementing IP Address Takeover (IPAT)
After completing this topic, you should be able to:
Describe IPAT via IP aliasing and IPAT via IP replacement:
How to configure a network to support them
What happens when
There are no failed components
A communication interface fails
A communication interface recovers
A node fails
A node recovers
Know how to select which style of IPAT is appropriate in a
given context
Describe how the AIX boot sequences changes when IPAT
is configured in a cluster
Describe the importance of consistent IP addressing and
labeling conventions
Copyright IBM Corporation 2005
Notes:
Topic 3 objectives
This section explains how to configure both variants of IP Address Takeover (IPAT).
Uempty
Two Ways to Implement IPAT
IPAT via IP Aliasing:
HACMP adds the service IP address to an (AIX) interface IP
address using AIX's IP aliasing feature:
ifconfig en0 alias 192.168.1.2
IPAT via IP Replacement:
HACMP replaces an (AIX) interface IP addresses with the
service IP addresses:
ifconfig en0 192.168.1.2
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-53
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Which is better?
We will examine the advantages and disadvantages of each method in the next few
pages. Keep in mind that the question is not which is better but rather which is better
suited to a particular context.
Uempty
IPAT via IP Aliasing Configuration
Define IP address for each network interface in the AIX ODM
Each interface IP address must be in a different logical IP subnet*
Define these address in the /etc/host file and configure them in HACMP
topology as communication interfaces
* Refer to earlier discussion of heartbeating and failure diagnosis for explanation of why
Copyright IBM Corporation 2005
Notes:
Requirements
Before configuring an HACMP network to use IPAT via IP aliasing, the cluster
configurator should ensure that all of the following are true:
- The network is of a type that supports IPAT via IP aliasing:
Ethernet
token-ring
FDDI
SP switch 1 / SP switch 2
- No service IP labels on the network require hardware address takeover (HWAT)
- The non-service IP addresses on each node are all on separate IP subnets
- The service IP addresses are on separate IP subnets from all non-service IP
addresses
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-55
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
subnet IP labels
192.168.10/24 n1-if1, n2-if1
192.168.11/24 n1-if2, n2-if2
9.47.87/24 appA-svc
9.47.88/24 appB-svc
Planning considerations
A node on a network that uses IPAT via aliasing may be the primary node for multiple
resource groups on the same network, regardless of the number of actual boot
interfaces on the node. Still, users should plan their networks carefully in order to
balance the RG load across the cluster.
Uempty IPAT via IP aliasing is supported by HACMP/ES 4.5 and HACMP 5.x systems. It is not
supported by earlier versions of HACMP including HACMP 4.5 classic.
Unfortunately, HACMP/ES 4.5 does a rather poor job of distributing service IP labels
across NICs.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-57
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
9.47.87.22 (alias)
192.168.10.1 (ODM) 192.168.11.1 (ODM) 192.168.10.2 (ODM) 192.168.11.2 (ODM)
Notes:
Operation
HACMP uses AIXs IP aliasing capability to alias service IP labels included in resource
groups onto interfaces (NICs) on the node which runs the resource group. With aliasing,
the non-service IP address (stored in the ODM) is still present.
Note that one advantage of sorts of IPAT via IP aliasing is that the non-service IP
addresses do not need to be routable from the client/user systems.
Uempty
IPAT via IP Aliasing After an Interface Fails
If the communication interface being used for the service IP label
fails, HACMP aliases the service IP label onto one of the node's
remaining available (currently functional) non-service (ODM)
interfaces
The eventual recovery of the failed boot adapter makes it available
again for future use
9.47.87.22 (alias)
192.168.10.1 (ODM) 192.168.11.1 (ODM) 192.168.10.2 (ODM) 192.168.11.2 (ODM)
Notes:
Interface failure
If a communication interface fails, HACMP moves the service IP addresses to another
communication interface, which is still available, on the same network. If there are no
remaining available NICs on the node for the network, then HACMP initiates a fallover
for that resource group.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-59
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
9.47.87.22 (alias)
192.168.10.2 (ODM) 192.168.11.2 (ODM)
Notes:
Node failure
When a node which has an IPAT-enabled resource group fails, the resource group is
acquired by a surviving node which is listed in the resource groups configuration. The
service IP address is aliased onto an available (currently functional) communication
interface on the takeover node.
Uempty
IPAT via IP Aliasing: Distribution
preference for service IP label aliases
Network level attribute which controls the placement of service IP
labels onto physical NICs
Load balancing
VPN requirements
If there are insufficient interfaces available to satisfy the
preference, HACMP allocates service IP label aliases and
persistent IP labels to an existing active network interface card
Four choices
Anti-Collocation
Collocation
Collocation with Persistent Label
Anti-Collocation with Persistent Label
Figure 3-35. IPAT via IP Aliasing: Distribution preference for service IP label aliases AU546.0
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-61
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
- VPN requirements:
Allows you to configure the type of the distribution preference suitable for the VPN
firewall external connectivity requirements.
- HACMP will try to meet preferences, but will always keep service labels active:
The distribution preference is exercised as long as there are acceptable network
interfaces available. However, HACMP always keeps service IP labels active, even
if the preference cannot be satisfied.
Uempty
IPAT via IP Aliasing Summary
Configure each node's communication interfaces with non-
service IP addresses (each on a different subnet)
Assign service IP labels to resource groups as appropriate
Must be on separate subnet from non-service IP addresses.
There is a total limit of 256 IP addresses known to HACMP
and 64 resource groups. Within those overall limits:
There is no limit on the number of service IP addresses in a
resource group
There is no limit on the number of resource groups with
service IP labels
HACMP assigns service IP labels to communication interfaces
using IP aliases based on resource group rules and available
hardware
IPAT via IP aliasing requires that hardware address takeover
is not configured
IPAT via IP aliasing requires gratuitous ARP support
Notes:
Summary
The visual summarizes IPAT via IP aliasing. Some additional considerations are
discussed below.
Advantages
Probably the most significant advantage to IPAT via IP aliasing is that it supports
multiple service IP labels per network per resource group on the same communication
interface and allows a node to easily support quite a few resource groups. In other
words, IPAT allows you to share several service labels on one interface. Thus it can
require fewer physical interfaces than IPAT via replacement.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-63
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Disadvantages
Probably the most significant disadvantage is that IPAT via IP aliasing does not support
hardware address takeover (HWAT).
In addition, since you must have a subnet for each interface and a subnet for each
service IP label, IPAT via IP aliasing can require a lot of subnets.
Uempty
IPAT via IP Replacement Overview
AIX 5L boots with a non-service (ODM) IP address on each NIC
When HACMP is started, it replaces non-service IP labels with
service IP labels for the resource groups it brings online
Only one service IP label on a NIC at a time
If the NIC hosting a service IP label fails, HACMP will attempt to
replace the non-service IP label of another NIC with the service
IP label, in order to maintain the service IP label
Configuration rules:
Each service IP label must be in the same subnet as a non-
service label subnet
There must be at least as many NICs on each node as there
are service IP lables
All service IP labels must be in the same subnet
Advantages
Supports hardware address takeover (HWAT)
Requires fewer subnets
Disadvantages
Requires more NICs to support multiple service IP labels
Less flexible
Notes:
History
In the beginning, IPAT via IP replacement was the only form of IPAT available. IPAT via
IP aliasing became available when AIX became able to have multiple IP addresses
associated with a single NIC. Because IPAT via IP aliasing is more flexible and usually
requires less network interface cards, IPAT via IP replacement is not often used.
This visual gives a brief overview of IPAT via IP replacement. A detailed discussion can
be found in Appendix C.
Configuration rules
The visual summarizes the configuration rules. Notice that they are almost the opposite
to the rules for IPAT via IP aliasing.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-65
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Advantages
Probably the most significant advantage of IPAT via IP replacement is that it supports
hardware address takeover (HWAT). HWAT may be needed if your local clients or
routers do not support gratuitous ARP. This will be discussed in a few pages.
Another advantage is that it requires fewer subnets. If you are limited in the number of
subnets available for your cluster, this may be important.
Note: If reducing the number of subnets needed is important, another alternative may
be to use heartbeating via aliasing, see Heartbeating over IP aliases on page 41.
Disadvantages
Probably the most significant disadvantages are that IPAT via IP replacement limits the
number of service IP labels per subnet per resource group on one communications
interface to one and makes it rather expensive (and complex) to support lots of
resource groups in a small cluster. In other words, you need more network adapters to
support more applications.
Uempty
Changes to AIX Start Sequence
The startup sequence of AIX networking is changed when IPAT is
enabled.
/etc/inittab
/sbin/rc.boot
/etc/inittab cfgmgr
/sbin/rc.boot /etc/rc.net (modified for ipat)
cfgmgr exit 0
/etc/rc.net /etc/rc
cfgif mount all
/etc/rc /usr/sbin/cluster/etc/harc.net
mount all /etc/rc.net -boot
cfgif
/etc/rc.tcpip
daemons start < HACMP startup > clstmgr
event node_up
/etc/rc.nfs node_up_local
daemons start get_disk_vg_fs
exportfs acquire_service_addr
telinit -a
/etc/rc.tcpip
daemons start
/etc/rc.nfs
IPAT changes the init daemons start
sequence exportfs
Notes:
/etc/inittab changes
A node with a network configured for IPAT via IP replacement must not start inetd until
HACMP has had a chance to assign the appropriate IP addresses to the nodes NICs.
Consequently, the AIX start sequence is modified slightly if a node has a resource
group which uses either form of IPAT.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-67
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Changes to /etc/inittab
init:2:initdefault:
brc::sysinit:/sbin/rc.boot 3 >/dev/console 2>&1 # Phase 3 of system boot
. . .
srcmstr:23456789:respawn:/usr/sbin/srcmstr # System Resource Controller
harc:2:wait:/usr/es/sbin/cluster/etc/harc.net # HACMP for AIX network startup
rctcpip:a:wait:/etc/rc.tcpip > /dev/console 2>&1 # Start TCP/IP daemons
rcnfs:a:wait:/etc/rc.nfs > /dev/console 2>&1 # Start NFS Daemons
. . .
qdaemon:a:wait:/usr/bin/startsrc -sqdaemon
writesrv:a:wait:/usr/bin/startsrc -swritesrv
. . .
ctrmc:2:once:/usr/bin/startsrc -s ctrmc > /dev/console 2>&1
ha_star:h2:once:/etc/rc.ha_star >/dev/console 2>&1
dt:2:wait:/etc/rc.dt
cons:0123456789:respawn:/usr/sbin/getty /dev/console
xfs:0123456789:once:/usr/lpp/X11/bin/xfs
hacmp:2:once:/usr/es/sbin/cluster/etc/rc.init >/dev/console 2>&1
clinit:a:wait:/bin/touch /usr/es/sbin/cluster/.telinit # HACMP for AIX These must
be the last entries of run level a in inittab!
pst_clinit:a:wait:/bin/echo Created /usr/es/sbin/cluster/.telinit > /dev/console #
HACMP for AIX These must be the last entries of run level a in inittab!
Notes:
HACMP 5.1
HACMP 5.1 added the harc entry to the /etc/inittab file, which runs harc.net to
configure the network interfaces. Also, starting in HACMP 5.1, some of the other inittab
entries have been changed to run in run-level a. These are invoked by HACMP when it
is ready for the TCP/IP daemons to run. The final two lines use the touch command to
create a marker file when all of the run-level a items have been run. HACMP waits for
this marker file to exist so that it knows when the run-level a items have been
completed.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-69
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Adopt IP Address Numbering Conventions
HACMP cluster tend to have quite a few IP addresses associated
with them
If at all possible, adopt an IP address numbering convention
Requirements imposed by corporate IT policies or the network
administrators may make it impractical to follow any sort of
convention (do the best you can)
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-71
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
hudson-if1
Copyright IBM Corporation 2005
Notes:
Uempty
An IPAT via IP Aliasing Convention
Here's one possible IP label number convention for IPAT via IP aliasing
networks:
IP address is of the form AA.BB.CC.DD
AA.BB is assigned by network administrator
CC indicates which interface, service IP label on each node:
15,16 indicates non-service/interface IP labels
5 chosen for service labels
etc (as required) bondar-if1 192.168.15.29
bondar-if2 192.168.16.29
hudson-if1 192.168.15.31
DD indicates which node hudson-if2 192.168.16.31
29 indicates an IP address on bondar xweb-svc 192.168.5.92
yweb-svc 192.168.5.70
31 indicates an IP address on hudson
Be flexible. For example, this convention uses DD=29 for bondar and
DD=31 for hudson because the network administrator assigned bondar-if1
to be 192.168.15.29 and hudson-if1 to be 192.168.15.31. Fortunately, the
network administrator could be convinced to use .29 and .31 for the other
bondar and hudson interface IP addresses.
Notes:
Limitations
It may not be possible to adopt a particularly consistent IP address convention. Make
sure that, as a minimum, you adopt a consistent IP labeling convention as you are
frequently looking at the labels rather than the numbers, and any consistent convention
generally avoids mistakes.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-73
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
The /etc/hosts file
All of the cluster's IP labels must be defined in every cluster node's
/etc/hosts file:
127.0.0.1 loopback localhost 127.0.0.1 loopback localhost
# cluster explorers # cluster explorers
# netmask 255.255.255.0 # netmask 255.255.255.0
Notes:
/etc/hosts
Make sure that the /etc/hosts file on each cluster node contain all of the IP labels used
by the cluster (you do not want HACMP to be in a position where it must rely on an
external DNS server to do IP label to address mappings).
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-75
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
As a result, the /etc/hosts file of each cluster node must contain all HACMP-defined IP
labels for all cluster nodes.
Maintaining /etc/hosts
The easiest way to ensure that all of the /etc/hosts file contain all of the required
addresses is to get one /etc/hosts file setup correctly and then copy it to all of the other
nodes or use the filecollections facility of HACMP 5.x.
Uempty
Service IP Address Examples
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-77
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Configuration problems
The visual shows some common IP configuration errors to watch out for.
History
A common error in HACMP 4.x involves the /.rhosts file. HACMP 4.x relies upon the
ability to issue rsh commands between cluster nodes when cluster configuration
changes are being made. This results in a requirement that each cluster node trust
every other cluster node for root-owned rsh sessions. This requirement is met, often
grudgingly, by configuring appropriate /.rhosts files. These files are not needed except
when cluster configuration changes are being made so they can be removed or
renamed outside of maintenance windows.
HACMP 5.x takes a totally different approach to dealing with cluster configuration
changes which eliminates the need for /.rhosts files.
Uempty
Single IP Adapter Nodes
Single IP Adapter nodes may appear attractive as they appear to
reduce the cost of the cluster
The cost reduction is an illusion:
1. A node with only a single adapter on a network is a node with a
single point of failure - the single adapter.
2. Clusters with unnecessary single points of failure tend to suffer more
outages
3. Unnecessary outages cost (potentially quite serious) money
One of the fundamental cluster design goals is to reduce
unnecessary outages by avoiding single points of failure
HACMP requires at least two NICs per IP network for failure
diagnosis
Clusters with less than two NICs per IP network are not
supported*
* Certain Cluster 1600 SP Switch-based clusters are supported with only one
SP Switch adapter per network.
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-79
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-81
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 4 Objectives:
The Impact of IPAT on Clients
After completing this topic, you should be able to:
Explain how user systems are affected by IPAT related
operations
Describe the ARP cache issue
Explain how gratuitous ARP usually deals with the ARP
cache issue
Explain three ways to deal with the ARP cache issue if
gratuitous ARP does not provide a satisfactory resolution to
the ARP cache issue:
Configure clinfo on the client systems
Configure clinfo within the cluster
Configure Hardware Address Takeover within the cluster
Notes:
Topic 4 Objectives
This section looks at the impact of IPAT on client systems.
Uempty
How Are Users Affected?
IP address moves/swaps within a node result in a short outage
Long-term connection oriented sessions typically recover seamlessly
(TCP layer deals with packet retransmission)
Resource group fallovers to a new node result in a longer outage and sever
connection oriented services (long term connections must be reestablished,
short term connections retried)
In either case:
Short-lived TCP-based services like http and SQL queries experience
short server down outage
UDP-based services must deal with lost packets
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-83
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
or so seconds when moving an IP address within a node or it can take a few minutes or
more in the case of a fallover.
Uempty
What About the User's Computers?
An IPAT operation renders ARP cache entries on client systems
obsolete
Client systems must (somehow) update their ARP caches
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-85
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
ARP: ARP:
router (192.168.8.1) 00:04:ac:42:9c:e2 router (192.168.8.1) 00:04:ac:42:9c:e2
192.168.8.3 192.168.8.3
00:04:ac:27:18:09 00:04:ac:27:18:09
ARP: ARP:
xweb (192.168.5.1) 00:04:ac:62:72:49 192.168.8.1 xweb (192.168.5.1) ??? 192.168.8.1
client (192.168.8.3) 00:04:ac:27:18:09 00:04:ac:42:9c:e2 client (192.168.8.3) 00:04:ac:27:18:09 00:04:ac:42:9c:e2
192.168.5.99 192.168.8.99
00:04:ac:29:31:37 00:04:ac:29:31:37
xweb 192.168.5.1 (alias) 192.168.5.1 (alias) xweb
192.168.10.1 (ODM) 192.168.11.1 (ODM) 192.168.10.1 (ODM) 192.168.11.1 (ODM)
00:04:ac:62:72:49 00:04:ac:48:22:f4 00:04:ac:62:72:49 00:04:ac:48:22:f4
Notes:
Uempty
Gratuitous ARP
AIX 5L supports a feature called gratuitous ARP
AIX sends out a gratuitous (that is, unrequested) ARP update
whenever an IP address is set or changed on a NIC
Other systems on the local physical network are expected to
update their ARP caches when they receive the gratuitous ARP
packet
Remember: only systems on the cluster's local physical network
must respect the gratuitous ARP packet
So arp update problems have been minimized
Required if using IPAT via aliasing
Notes:
Gratuitous ARP
AIX 5L supports a feature called gratuitous ARP. Whenever an IP address associated
with a NIC changes, AIX broadcasts out a gratuitous (in other words, unsolicited) ARP
update. This gratuitous ARP packet is generally received and used by all systems on
the clusters local physical network to update their ARP cache entries.
The result is that all relevant ARP caches are updated almost immediately after the IP
address is assigned to the NIC.
The problem is that not all systems respond or even necessarily receive these
gratuitous ARP cache update packets. If a local system either does not receive or
ignores the gratuitous ARP cache packet then its ARP cache remains out-of-date.
Note that unless the network is VERY overloaded, local systems generally either
always or never act upon the gratuitous ARP update packet.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-87
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
What if Gratuitous ARP is Not Supported?
If the local network technology doesn't support gratuitous ARP or
there is a client system or router on the local physical network
which must communicate with the cluster and which does not
support gratuitous ARP packets:
clinfo can used on the client to receive updates of changes.
clinfo can be used on the servers to ping a list of clients, forcing an
update to their ARP caches.
HACMP can be configured to perform Hardware Address Takeover
(HWAT).
Suggestion:
Do not get involved with using either clinfo or HWAT to deal with
ARP cache issues until you've verified that there actually are ARP
issues which need to be dealt with.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-89
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
snmpd
clinfo
clstrmgr clinfo.rc
Notes:
Uempty
Option 2: clinfo From Within the Cluster
clinfo may also be used on the cluster's nodes to force an ARP cache
update.
In this option, clinfo runs on every cluster node
If clinfo is only run on one cluster node then that node become a
single point of failure!
clinfo flushes local ARP cache (on the cluster node) then pings a
defined list of clients listed in /usr/es/sbin/cluster/etc/clinfo.rc
Clients pick up the new IP address to hardware address relationship as a
result of the ping request
clinfo.rc
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-91
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
TOTAL_CLIENT_LIST="${PING_CLIENT_LIST}"
if [[ -s /etc/cluster/ping_client_list ]] ; then
#
# The file "/etc/ping_client_list" should contain only a line
# setting the variable "PING_CLIENT_LIST" in the form given
# in the example above. This allows the client list to be
# kept in a file that is not altered when maintenance is
# applied to clinfo.rc.
#
. /etc/cluster/ping_client_list
TOTAL_CLIENT_LIST="${TOTAL_CLIENT_LIST} ${PING_CLIENT_LIST}"
fi
#
# WARNING!!! For this shell script to work properly, ALL entries in
# the TOTAL_CLIENT_LIST must resolve properly to IP addresses or hostnames
# (must be found in /etc/hosts, DNS, or NIS). This is crucial.
. . .
Notes:
clinfo.rc
The clinfo.rc script must be edited manually on the cluster nodes which run clinfo.
There is no reason why clinfo cannot also be run on the client systems although these
changes are only required on the cluster nodes which are running clinfo.rc.
Remember: all the cluster nodes should be running clinfo if clinfo is being used
within the cluster to deal with ARP cache issues since you never know which cluster
nodes will survive whatever has gone wrong).
Edit the /usr/es/sbin/cluster/etc/clinfo.rc file on each server node. Add the IP
label or IP address of each system that accesses service IP addresses managed by
HACMP to the PING_CLIENT_LIST list. Then start the clinfo daemon (clinfo can be
started as part of starting HACMP on the cluster nodes).
Uempty /etc/cluster/ping_client_list
You can also provide the list of clients to be pinged in the file
/etc/cluster/ping_client_list. This is probably the best method as it ensures that the
list of clients to ping is not overlaid by future changes to clinfo.rc.
More details
This script is invoked by HACMP as follows:
clinfo.rc {join,fail,swap} interface_name
The next set of details likely do not make sense until we are further into the course.
When clinfo is notified that the cluster is stable after undergoing a failure recovery of
some sort or when clinfo first connects to clsmuxpd (the SNMP part of HACMP), it
receives a new map (description of the clusters state). It checks for changed states of
interfaces:
- If a new state is UP, clinfo calls clinfo.rc join interface_name.
- If a new state is DOWN, clinfo calls clinfo.rc fail interface_name.
- If clinfo receives a node_down_complete event, it calls clinfo.rc with the fail
parameter for each interface currently UP.
- If clinfo receives a fail_network_complete event, it calls clinfo.rc with the
fail parameter for all associated interfaces.
- If clinfo receives a swap_complete event, it calls clinfo.rc swap
interface_name.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-93
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
HWAT considerations
There are a few points which must be kept in mind when contemplating HWAT:
Uempty - The hardware address which is associated with the service IP address must be
unique within the physical network that the service IP address is configured for.
- HWAT is not supported by IPAT via IP aliasing because each NIC can have more
than one IP address but each NIC can only have one hardware address.
- HWAT is only supported for Ethernet, token ring and FDDI networks (MCA FDDI
network cards do not support HWAT). ATM networks do not support HWAT.
- HWAT increases the takeover time (usually by just a few seconds).
- HWAT is an optional capability which must be configured into the HACMP cluster
(we will see how to do that in detail in a later unit).
- Cluster nodes using HWAT on token ring networks must be configured to reboot
after a system crash as the token ring card will continue to intercept packets for its
hardware address until the node starts to reboot.
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-95
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Checkpoint
1. True or False?
Clients are required to exit and restart their application
after a fallover.
2. True or False?
All client systems are potentially directly affected by the
ARP cache issue.
3. True or False?
clinfo must not be run both on the cluster nodes and
on the client systems.
4. If clinfo is run by cluster nodes to address ARP cache
issues, you must add the list of clients to ping to either the
__________________________ or the
__________________________ file
Notes:
Uempty
Unit Summary (1 of 2)
HACMP uses networks to:
Provide highly available client access to applications in the cluster
Detect and diagnose NIC, node, and network failures using RSCT heartbeats
Communicate with HACMP daemons on other nodes
All HACMP clusters require a non-IP network
Differentiate between node, IP subsystem and network failures
Prevent cluster partitioning
HACMP networking terminology
Service IP label/address: HA address used by client to access application
Non-service IP label/address: Applied to NIC at boot time; stored in AIX ODM
Persistent IP label/address: Node bound HA address for admin access to a node
Communication interface: Association between a NIC and an IP label/address
Communication device: Device used in non-IP network
Communication adapter: X.25 adapter used in a HA communication link
IP Address Takeover (IPAT): Moves service IP address to working NIC after a
failure
IPAT via aliasing: Adds the service address to a NIC using IP aliasing
IPAT via replacement: Replaces the non-service address with the service address
Notes:
Copyright IBM Corp. 1998, 2005 Unit 3. Networking Considerations for High Availability 3-97
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Summary (2 of 2)
HACMP has very specific requirements for subnets
IPAT via aliasing
NICs on a node must be on different subnets; which must use the same subnet mask
There must be at least one subnet in common with all nodes
Service addresses must be on different subnet than any non-service address
A service address can be on same subnet with another service address
IPAT via replacement
NICs on a node must be on different subnets; which must use the same subnet mask
Each service address must be in same subnet as one of the non-service addresses on
the highest priority node
Multiple service addresses must be in the same subnet
Heartbeating over IP alias (any form of IPAT)
Service and non-service addresses can coexist on the same subnet, or be on separate
subnets
One subnet required for heartbeating; does not need to be routed
HACMP can update local clients ARP cache after IPAT
Gratuitous ARP (default)
clinfo on clients
clinfo on server nodes
Hardware address takeover (HWAT)
Notes:
References
SC23-4864-06 HACMP for AIX, Version 5.3: Concepts and Facilities
Guide
SC23-4861-06 HACMP for AIX, Version 5.3 Planning and Installation
Guide
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
List and explain the requirements for an application to be
supported in an HACMP environment
Describe the HACMP start and stop scripts
Describe the resource group behavior policies supported by
HACMP
Enter the configuration information into the Planning
Worksheets
Notes:
Uempty
How to Define an Application to HACMP
Two steps to define an Application to HACMP
Step 1 Create Resource
Application Server: Defines start and stop scripts
Step 2 Create Resource Group:
Resource Group
Shared
Disk
List of Nodes
Policies: Where to run
Resources
Application Server
Service Address
Volume Group
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Application Considerations
Automation
No intervention
Dependencies
Using names unique to one node
Other applications
Interference
Conflicts with HACMP
Robustness
Application can withstand problems
Implementation
Other aspects to plan for
Monitoring using HACMP
Notes:
Introduction
Many applications can be put under the control of HACMP but there are some
considerations that should be taken into account.
Automation
One key requirement for an application to function successfully under HACMP is that
the application be able to start and stop without any manual intervention. Since the
cluster daemons call the start and stop scripts, there is no option for interaction.
Additionally, upon an HACMP fallover, the recovery process calls the start script to bring
the application online on a standby node. This allows for a fully automated recovery
Other requirements for start and stop scripts will be covered on the next visual.
Uempty Dependencies
Dependencies to be careful of when coding the scripts include:
i. Referring to a locally attached device.
ii. Hard coding such as /dev/tty0 which may not be the same on another node.
iii. Using a hostname which is not the same on other nodes.
iv. Software licensing: Software can be licensed to a particular CPU ID. If this is the
case with your application, it is important to realize that a fallover of the software
will not successfully restart. You may be able to avoid this problem by having a
copy of the software resident on all cluster nodes. Know whether your application
uses software that is licensed to a particular CPU ID.
Application dependencies:
Dependencies that in the past you had to worry about but now you may not have to:
i. One application must be up before another one
ii. Applications that must both run on the same node
These can now be handled by Run-Time Dependency options. An overview of these
is given later in this unit.
Interference
An application may execute properly on both the primary and standby nodes. However,
when HACMP is started, a conflict with the application or environment could arise that
prevents HACMP from functioning successfully. Two areas to lookout for are: Using
IPX/SPX Protocol and Manipulating Network Routes.
Robustness
Beyond basic stability, an application under HACMP should meet other robustness
characteristics, such as successful start after hardware failure and survival of real
memory loss. It should also be able to survive the loss of the kernel or processor state.
Implementation
There are a number of aspects of an application to consider as you plan for
implementing it under HACMP. Consider characteristics such as time to start, time to
restart after failure, and time to stop. Also consider:
Writing effective scripts.
Consider file storage locations.
Using inittab and cron Table: Inittab is processed before HACMP is started. Cron
table is local to a each node. Time/date should be synchronized.
We will look at writing scripts and data locations in the following visuals.
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Writing Start and Stop Scripts
Items to check:
Environment is what is expected
Multiple instances issue
Location of scripts
Handle Errors from previous termination
Correct coding
Using Assists
Notes:
Introduction:
Application start scripts should not assume the state of the environment; defensive
programming may correct any irregular conditions that occur. Remember that cluster
manager spawns these scripts off a separate job in the background and carries on
processing. The application start scripts must be able to handle an unknown previous
shutdown state.
Items to check
- Environment:
Verify the environment. Are the prerequisite conditions satisfied? These may include
access to a file system, adequate paging space, IP labels and free file system
space. The start script should exit and run a command to notify system
administrators if the requirements are not met.
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Using Assists
IBM provides a priced feature for HACMP that provides all the code and monitoring for
3 applications: WebSphere, DB2, and Oracle Real Application Server (RAC). In these
cases you would not have to write the scripts yourself.
There are also plug-in filesets that provide help for integrating print server, DHCP, and
DNS. These filesets are part of the base HACMP product.
Uempty
Where Should Data Go?
Private Storage
Operating system components
Shared Storage
Dynamic data
Web server content
Log Files
Files updated by application
It Depends
Configuration files
Application binaries
License files
Notes:
Introduction
Deciding where data should go should be thought out well. For some data, the answer
is clear. For other cases, it depends. Putting data on shared storage allows for only one
copy but may not be available when needed. Putting data on private storage is subject
to having different copies but upgrades can be done easier.
Private storage
Private storage must be used for the operating system components. It may also be
used for configuration files, license files and application binaries subject to the
trade-offs mentioned in the introduction above.
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Shared storage
Shared storage must be used for dynamic data, Web server content, data that is
updated by the application and application log files (be sure time is same on the nodes).
Again configuration files, application binaries, and license files could go here subject to
the trade-offs mentioned in the introduction above.
It depends
License files deserves a special mention. If using node locked, then you should use
private storage. In any case, you must learn the license requirements of the application
to make a proper determination.
Uempty
Shared Storage Questions
Some questions to ask your user or customer:
Notes:
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty 3) Fallback
If a node earlier in the list of nodes (that is, higher priority) for the resource
group is started after a fallover, then the Fallback policy determines if the
resource group should be stopped and started back up on the higher priority
node.
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Startup Policy
Online on Home Node Only
Notes:
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty simultaneously from multiple nodes. Oracle Real Application Server (RAC) is an
application that uses this type of startup policy.
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Fallover Policy
Notes:
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Fallback Policy
Never Fallback
Notes:
Uempty appropriate time. After starting the node, HACMP automatically starts the resource
group fallover at the specified time.
Run-time policies will be covered in more detail in the HACMP Administration II
Administration and Problem Determination course.
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Valid combinations
HACMP allows you to configure only valid combinations of startup, fallover, and fallback
behaviors for resource groups.
Uempty
Dependent Applications/Resource Groups
Node 1
Parent RG
Node 1 Node 2
Child/Parent Parent/Child
RG RG
Child RG
Notes:
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Checkpoint
1. True or False
Applications are defined to HACMP in a configuration file
that lists what binary to use.
2. What policies would be the best to use for a 2 node mutual
takeover cluster using IPAT to minimize both applications
running on the same node?
a. home, next, never
b. first, next, higher
c. distribution, next, never
d. all, error, never
e. home, next, higher
3. Which type of data should not be placed in private data
storage?
a. Log data
b. License file
c. Configuration files
d. Application binaries
4. Which policy is not a Run-time policy?
a. Settling
b. Delayed Fallback Timer
c. Dynamic Node Priority
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 4. Planning for Applications and Resource Groups 4-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Summary
To define an application to HACMP, you must:
Create an application server resource (start and stop scripts)
Create a resource group (node list, policies, resources)
Considerations for putting an application under HACMP control
Automation
Dependencies
Interference
Robustness
Implementation details
Monitoring
Shared storage requirements
Considerations for start and stop scripts:
Environment
Multiple instances
Script location
Error handling
Coding issues
Resource group policies control how HACMP manages the application
Startup policy (with optional Settling timer)
Fallover policy
Fallback policy (with optional Delayed fallback)
Notes:
References
SC23-4864-06 HACMP for AIX, Version 5.3: Concepts and Facilities
Guide
SC23-4861-06 HACMP for AIX, Version 5.3 Planning and Installation
Guide
SC23-4862-06 HACMP for AIX, Version 5.3 Administration Guide
SC23-5177-00 HACMP for AIX, Version 5.3 Troubleshooting Guide
SC23-4867-05 HACMP for AIX, Version 5.3 Master Glossary
www.ibm.com/servers/eserver/pseries/library/hacmp_docs.html#hacm
pv53HACMP Manuals
Unit Objectives
After completing this unit, you should be able to:
State where installation fits in the implementation process
Describe how to install HACMP 5.3
List the prerequisites for HACMP 5.3
List and explain the purpose of the major HACMP 5.3
components
Notes:
Uempty had a set of high availability daemons which provided most but not all of the
functionality required by HACMP, the SP variant of HACMP was developed to take
advantage of these high availability daemons within PSSP. The end result was that the
SP variant of HACMP did not include certain functionality, like the ability to monitor
networks and nodes using heartbeating, found in the non-SP variant because the SP
variant used PSSPs high availability daemons to provide this functionality.
The SP variant of HACMP came to be known as HACMP enhanced scalability or
HACMP/ES. The other HACMP became known as classic.
Over the years, the non-SP variant of HACMP and the SP variant of HACMP continued
to evolve as essentially separate products. By about the year 2000, the two products
provided roughly equivalent functionality although they were implemented in quite
different ways. The major functional difference between the two was that HACMP/ES
supported the SP switch and the non-SP variant did not. Another major difference
between the two was that the HACMP/ES variant required PSSP and the non-SP
variant, of course, did not.
In about 1998, a decision was made to reunite the two products. At about this time, the
non-SP variant came to be called HACMP classic to distinguish it from the HACMP/ES
variant. In order to make it possible to run HACMP/ES on non-PSSP capable servers,
the portions of PSSP used by HACMP were extracted into a new entity called Reliable
Scalable Cluster Technology (RSCT) and ported to conventional (that is, non-SP)
servers.
Topic 1 Objectives:
Installing the HACMP Software
After completing this Topic, you should be able to:
Where does Installation fit in the implementation process
Describe how to install HACMP 5.3
List the prerequisites for HACMP 5.3
Notes:
This topic covers the installation of the HACMP 5.3 filesets.
Uempty
Steps for Successful Implementation
HACMP should not be installed upon a system which is in
production
Notes:
Different opinions
Different people have different ideas about the exact order in which a cluster should be
configured. For example, some people prefer to leave the configuration of the shared
storage (step 5 above) until after theyve synchronized the clusters topology (step 7) as
this allows them to take advantage of HACMPs C-SPOC facility to configure the shared
storage.
One other area where different views are common is exactly when to install and
configure the application. If the application is installed, configured and tested
reasonably thoroughly prior to installing and configuring HACMP then most issues
which arise during later cluster testing are probably HACMP issues rather than
application issues. The other common perspective is that HACMP should be installed
and configured prior to installing and configuring the applications as this allows the
applications to be installed into the exact context that they will ultimately run in. There is
no correct answer to this issue. When to install and configure the applications is just
one more point that will have to be resolved during the cluster planning process.
Uempty
Where Are We in the Implementation?
9Plan for network, storage, and application resource groups
Eliminate single points of failure
9Define and configure the AIX environment
Storage (adapters, LVM volume group, filesystem)
Networks (IP interfaces, /etc/hosts, non-IP networks and
devices)
Application start and stop scripts
Install the HACMP filesets and reboot
Configure the HACMP environment
Topology
Cluster, node names, HACMP IP and non-IP networks
Resources:
Application Server
Service labels
Resource group:
Identify name, nodes, policies
Resources: Application Server, service label, VG, filesystem
Synchronize then start HACMP
Copyright IBM Corporation 2005
Notes:
Notes:
Uempty
What Is On the CD?
README_hacmp53
Directories
AIX5.2,5.3, pubs
Installp/ppc, usr/sys/inst.images
cluster.adt.es cluster.msg.<lang>.cspoc
cluster.doc.en_US.es cluster.msg.<lang>.es
cluster.es cluster.msg.<lang>.hativoli
cluster.es.cfs cluster.msg.<lang>.haview
cluster.es.clvm
cluster.es.cspoc
cluster.es.plugins rsct.core.lprm.2.4.2.1.bff
cluster.es.worksheets rsct.core.rmc.2.4.2.3.bff
cluster.hativoli rsct.core.sec.2.4.2.3.bff
cluster.haview rsct.core.sensorrm.2.4.2.1.bff
cluster.license rsct.core.utils.2.4.2.3.bff
cluster.man.en_US.es.data rsct.core.cimrm.2.4.2.1.bff
Notes:
Files on the CD
This visual shows the files that are on the CD. They will be expanded to show the table
of contents when using SMIT to do the install.
Notes:
Fileset considerations
Listed above are the filesets that you see when doing smit install_all in HACMP 5.3.
Using smit update_all will not show the msg filesets so you should use install_all and
select the filesets.
The same filesets should be installed on all nodes or Verify will give warnings every
time it executes.
You should install the documentation filesets on at least one non-cluster node (ensuring
that the HACMP PDF-based documentation is available even if none of the cluster
nodes will boot could prove REALLY handy someday).
Notice that some of the filesets require other products such as Tivoli or NetView. You
should not install these filesets unless you have these products. HAView is never
installed on the cluster node, it is installed on the NetView server.
Uempty The cluster.es.cfs fileset can only be used if GPFS is installed. You may not need the
plug-ins. The cluster.es.clvm fileset is required for Enhanced Concurrent Mode volume
group support and this fileset requires an RSCT fileset which you will see in lab.
Notes:
Uempty
Don't Forget the Prerequisites
AIX:
AIX 5L V5.2 ML4
AIX 5L V5.3 ML2
RSCT
rsct.compat.basic.hacmp 2.4.2.0 (AIX 5L V5.3) 2.3.6.0 (AIX 5L V5.2)
rsct.compat.clients.hacmp 2.4.2.1 (AIX 5L V5.3) 2.3.6.1 (AIX 5L V5.2)
rsct.core.sec 2.4.2.0 (AIX 5L V5.3) 2.3.6.0 (AIX 5L V5.2)
rsct.core.rmc 2.4.2.1 (AIX 5L V5.3) 2.3.6.1 (AIX 5L V5.2)
PSSP 3.5 on SP systems
Otherwise optional AIX filesets
(see student notes below):
Other prerequisites
Enhanced concurrent mode:
bos.rte.lvm 5.1.0.25 or higher
bos.clvm.enh 5.2.0.11
CSPOC with vpath
SDD 1.3.1.3 or later
Online Planning Worksheets
AIX 5L Java Runtime Environment
Copyright IBM Corporation 2005
Notes:
Installation suggestions
You might be able to get HACMP 5.3 installed without all of the above prerequisites, but
you are unlikely to enjoy the experience of working with HACMP 5.3 without the
appropriate prerequisites.
Since you are unlikely to want to upgrade a new cluster anytime soon, it is generally
wisest to start with the latest available AIX and HACMP patches (the URL for checking
on the latest patches is on the next foil).
Figure 5-10. Verify That You Have the Required APARs AU546.0
Notes:
Obtaining fixes
Use the above URL to check for and download the latest fixes.
Uempty
Some Final Things to Check
Code installation
Each node must be rebooted once HACMP has been installed
Correct filesets and prerequisites have been installed
Documentation is installed and accessible
Network setup
/etc/hosts file is configured on all nodes correctly
Name resolution works
IP and non-IP networks are configured
Subnets configured correctly
The subnet mask identical.
All interfaces on different subnets
Routing configured correctly
Test connectivity
Shared storage configured correctly
Notes:
Description of checklist
This is a checklist of items which you should verify before starting to configure an
HACMP cluster. It is not a complete list as each situation is different. It would probably
be wise to develop your own checklist during the cluster planning process and then
verify it just before embarking on the actual HACMP configuration of the cluster.
Code installation
Correct filesets includes making sure that the same HACMP filesets are installed on
each node. Documentation can be installed before installing HACMP. The
documentation is delivered as either html or pdf.
Network setup
The /etc/hosts file should have entries for all IP labels and all nodes. The file should be
the same on all nodes. Name resolution should be tested on all labels and nodes. To do
this you can use the host command. You should test address to name and name to
address and verify that they are the same on all nodes. You should ensure that a route
exists to all logical networks from all nodes. Finally, you should test connectivity by
pinging all nodes from all nodes on all interfaces.
Shared storage
Check to see that the disks are configured and recognized the same (if possible) and
can be accessed from all nodes that will share it.
Uempty
Install HACMP Client Machine
Setup Network
Configure network interface to reach cluster server node
Same subnet as service address
/etc/hosts file updated everywhere
Install prerequisites:
bos.adt.libm
bos.adt.syscalls
bos.data
Install HACMP client filesets:
cluster.adt.es
cluster.es
ES Client Libraries
ES Client Runtime
ES Client Utilities
cluster.msg.en_US.es
cluster.man.en_US.es
Configure /usr/es/sbin/cluster/clhosts
Can copy /usr/es/sbin/cluster/etc/clhosts.client
Test connectivity
Notes:
Lets Review
1. What is the first step in implementing a cluster?
a. Order the hardware
b. Plan the cluster
c. Install AIX and HACMP
d. Install the applications
e. Take a long nap
2. True or False?
HACMP 5.3 is compatible with any version of AIX 5L V5.x.
3. True or False?
Each cluster node must be rebooted after the HACMP software is installed.
4. True or False?
You should take careful notes while you install and configure HACMP so
that you know what to test when you are done.
Notes:
Notes:
Uempty
The Layered Look
Here are the layers of software on an HACMP 5.3 cluster node:
Application Layer
Contains the highly available applications
that use HACMP services
HACMP Layer
Provides highly available services to
applications
AIX Layer
Provides operating system services (SRC, snmpd)
Notes:
Uempty
HACMP Components and Features
The HACMP software has the following components:
Cluster Manager
Cluster Secure Communication Subsystem
IBM RS/6000 Reliable Scalable Cluster Technology Availability
Services (RSCT and RMC)
snmpd monitoring programs
Cluster Information Program
Highly Available NFS Server
Shared External Disk Access
Notes:
Uempty
Cluster Manager
A subsystem/daemon which runs on each cluster node
Primarily responsible for responding to unplanned events:
Recover from software and hardware failures
Respond to user-initiated events:
Request to online/offline a node
Request to move/online/offline a resource group
And so forth
A client to RSCT and RMC
Provides snmp retrievable status information
Implemented by the subsystem clstrmgrES
Started in /etc/inittab
Notes:
Notes:
Notes:
clcomd basics
The most obvious part of the cluster secure communication facility is the cluster
communication daemon (clcomd). This daemon replaces a number of ad hoc
communication mechanisms with a single facility thus funneling all cluster
communication through one point. This funneling, in turn, makes it feasible to then use
a VPN to actually send the traffic between nodes and to be sure that all the traffic is
going through the VPN.
Improves performance
The clcomds approach to supporting the verification and synchronization of cluster
configuration changes has an important additional benefit - by eliminating numerous rsh
calls across the cluster during the verification and synchronization operation and
replacing them with a purpose-built facility, the time that it takes to verify and
synchronize a cluster configuration change has improved noticeably.
Uempty Other aspects of clcomds implementation which further improve performance include:
- Caching coherent copies of each nodes ODMs which reduces the amount of
information which must be transmitted across the cluster during a verification
operation
- Maintaining long term socket connections between nodes avoids the necessity to
constantly create and destroy the short term sessions which are a natural result of
using rsh and other similar mechanisms
Notes:
RSCT
Included with AIX (originally part of PSSP for SP systems)
Provides:
Scalability to large clusters
Cluster Failure notification
Coordination of changes
Key components:
Topology Services
Heartbeat services
Group Services
Coordinates/monitors state changes of an application in the cluster
RMC: Resource Monitoring and Control
Provides resource threshold monitoring and Allows application to be
notified of resource
HACMP's Cluster Manager is an RSCT client/application
Notes:
Uempty
HACMP from an RSCT Perspective
AIX
Process
Monitor
HACMP
HA Recovery Driver
Database RSCT ~
Resource RMC Cluster Recovery
Monitor (ctrmc) Manager Programs
Switch
Resource
Monitor
Recovery
RSCT Commands
RSCT
Group ~
Topology
Services Services HACMP Event
Scripts
processor, LAN
heartbeats messages
membership
information
Notes:
Monitors
The monitors in the upper left of the diagram monitor various aspects of the local nodes
state including the status of certain processes (for example, the application if
application monitoring has been configured), database resources and the SP Switch (if
one is configured on the node). These monitors report state changes related to
monitored entities to the RSCT RMC Manager.
Topology Services
Also reporting state changes to the RSCT RMC Manager is the RSCT Topology
Services component which is responsible for failure detection/diagnosis of topology
components and the transmission of any RSCT-related messages between cluster
nodes.
Group Services
Associated with RSCT Topology Services is the RSCT Group Services daemon which
is responsible for coordinating and monitoring changes to the state of an application
running on multiple nodes. In the HACMP context, the application running on multiple
nodes is the HACMP cluster manager.
RMC Manager
The RSCT RMC Manager receives notification of events from the monitors and from
RSCT Topology Services. It analyzes these events and notifies RSCT clients of those
events which they have expressed an interest in.
The HACMP cluster manager, an RSCT client, registers itself with both the RSCT RMC
Manager and the RSCT Group Services components. The key interface is with the
RSCT RMC Manager which notifies the HACMP Cluster Manager of events that the
Cluster Manager has told the RMC Manager that it is interested in (for example, node
failures).
Cluster Manager
Once an event has been reported to the HACMP Cluster Manager, it responds to this
event the use of HACMPs recovery commands and event scripts to respond to the
event. The scripts are coordinated via the RSCT group services component.
Uempty
Heartbeat Rings
Heartbeat
25.8.60.6 25.8.60.5
25.8.60.2 25.8.60.4
25.8.60.3
Notes:
Example
For example, the IP addresses in the foil can be sorted as 25.8.60.6, 25.8.60.5,
25.8.60.4, 25.8.60.3 and 25.8.60.2. This ordering results in the following heartbeat path:
25.8.60.6 --> 25.8.60.5-->25.8.60.4-->25.8.60.3-->25.8.60.2-->25.8.60.6
Notes:
Uempty
Cluster Information Daemon (clinfo)
An SNMP-aware client to the cluster manager
Provides
A cluster information API to the HACMP SNMP manager
Focused on providing HACMP cluster information
Easier to work with than the SNMP APIs
Support for ARP cache issues
Is used by:
The clstat command
Customer written utility/monitoring tools
Implemented as the clinfoES subsystem
Notes:
Starting Clinfo
Starting Clinfo on an HACMP server node:
The clinfo daemon can be started in a number of ways (see the HACMP
Administration Guide) but probably the best way is to start it along with the rest of
the HACMP daemons by setting the Startup Cluster Information Daemon? field to
true when using the smit Start Cluster Services screen (which will be discussed in
the next unit).
Starting Clinfo on a Client:
Use the /usr/es/sbin/cluster/etc/rc.cluster script or the startsrc command to start
clinfo on a client.
You can also use the standard AIX 5L startsrc command startsrc -s clinfoES
Uempty
Highly Available NFS Server Support
Cluster administrator can:
Define NFS exports at directory level to all clients
Define NFS mounts and network to HACMP nodes
Specify export options for HACMP to set
HACMP preserves file locks and dupcache across fallovers
Limitations
Lock support is limited to two node clusters
Resource group is only active on one node at a time
Notes:
Limitations
- The locking function is available only for two-node clusters
- The resource group must behave as non-concurrent - active on one node at a time
Notes:
Uempty
Checkpoint
1. What component detects an adapter failure?
a. Cluster manager
b. RSCT
c. clcomd
d. clinfo
2. What component provides SNMP information?
a. Cluster manager
b. RSCT
c. clsmuxpd
d. clinfo
3. What component is required for clstat to work?
a. Cluster manager
b. RSCT
c. clcomd
d. clinfo
Notes:
Unit Summary
Notes:
References
SC23-4861-06 HACMP for AIX, Version 5.3 Planning and Installation
Guide
SC23-4862-06 HACMP for AIX, Version 5.3 Administration Guide
SC23-5177-00 HACMP for AIX, Version 5.3 Troubleshooting Guide
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Notes:
Objectives
This unit will show how to configure a two-node mutual takeover HACMP cluster with a
heartbeat over disk and rs232 non-IP network. It will then demonstrate how to startup
and shutdown HACMP, modify the configuration of the cluster by adding and deleting
nodes, adding and deleting resource groups, implementing hardware address takeover
and configuring target-mode SSA non-IP networks.
Uempty
What We Are Going to Achieve?
bondar hudson
X X
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
The Topology Configuration
Here's the key portion of the /etc/hosts file that we'll be using in this
unit:
192.168.15.29 bondar-if1 # bondar's first interface IP label
192.168.16.29 bondar-if2 # bondar's second interface IP label
192.168.5.29 bondar-per # persistent node IP label on bondar
192.168.15.31 hudson-if1 # hudson's first interface IP label
192.168.16.31 hudson-if2 # hudson's second interface IP label
192.168.5.31 hudson-per # persistent node IP label on hudson
192.168.5.92 cxweb # the IP label for the application
# normally resident on bondar
Hostnames: bondar, hudson
bondar's network configuration (defined via smit chinet) :
en0 - 192.168.15.29
en1 - 192.168.16.29
hudson's network configuration:
en0 - 192.168.15.31
en1 - 192.168.16.31
These network interfaces are all connected to the same
physical network
The subnet mask is 255.255.255.0 on all networks/NICs
An enhanced concurrent mode volume group "ecmvg" has
been created to support the xweb application and will be
used for a disk non-IP heartbeat network
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Configuration Methods
HACMP provides two menu paths with three methods to
configure topology and resources:
Initialization and Standard Configuration
Two-node cluster configuration assistant
Does everything (Topology, Resources, Resource Group)
Only supports a 2 node standby cluster
Standard configuration
Topology done in one step
You then must configure resource groups and synchronize
Extended Configuration
More flexible but with all the options
Notes:
Configuration methods
- Standard Configuration
With this method you must do the following:
i. Topology (simplified via Add Nodes to an HACMP Cluster)
ii. Configure Resources and Resource Groups
iii. Verify and Synchronize
- Two-Node Cluster Configuration Assistant
With this method, all the steps of Standard Configuration are done at once including
adding a non-IP disk heartbeat network if you created an enhanced concurrent
volume group.
- Extended Configuration
With this method you follow similar steps as the Standard Configuration but
Topology has more steps and there are many more options. Some options can only
be done using this method such as adding a non-IP network.
Uempty
Plan Two-Node Configuration Assistant
Plan configuration path to other node = hudson_if1
This node will be the home (primary) node
Plan Application Server name = xwebserver
Used to name the cluster and the resource group
Ensure Application Server start and stop scripts exist and
are put on Bondar (where 2 node assistant will be run from):
/mydir/xweb_start
/mydir/xweb_stop
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Almost There . . .
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
The Standard Configuration Menu
Initialization and Standard Configuration
Configuration Assistants
Add Nodes to an HACMP Cluster
Configure Resources to Make Highly Available
Configure HACMP Resource Groups
Verify and Synchronize HACMP Configuration
HACMP Cluster Test Tool
Display HACMP Configuration
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Caution
If changes are made on one node but not synchronized and then more changes are
made on a second node and then synchronized then the changes made on the first
node are lost.
If you want to avoid losing work, make sure that you dont flip back and forth between
nodes while doing configuration work (that is, work on only one node at least until
youve synchronized your changes).
Also be careful of having two persons work on two-nodes at the same time. The first
synchronization will wipe out the second persons changes.
Recommendation
Pick one of your cluster nodes to be the one node that you use to make changes.
Configuration Assistants
Besides the Two-Node Cluster Configuration Assistant, HACMP 5.3 provides, via an
additional feature, assistants for WebSphere, Oracle, and DB2.
Uempty
Two-Node Cluster Configuration Assistant
Two-Node Cluster Configuration Assistant
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Communication Path to Takeover Node [hudson_if1] +
* Application Server Name [xwebserver]
* Application Server Start Script [/mydir/xweb_start]
* Application Server Stop Script [/mydir/xweb_stop]
* Service IP Label [xweb] +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Configuration Assistants
Add Nodes to an HACMP Cluster
Configure Resources to Make Highly Available
Configure HACMP Resource Groups
Verify and Synchronize HACMP Configuration
HACMP Cluster Test Tool
Display HACMP Configuration
Notes:
Uempty
Add Nodes to an HACMP Cluster
Configure Nodes to an HACMP Cluster (standard)
[Entry Fields]
* Cluster Name [mycluster]
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
NODE hudson:
Network net_ether_01
hudson-if1 192.168.15.31
hudson-if2 192.168.16.31
Notes:
Uempty
What Is Left To Do?
Initialization and Standard Configuration
Configuration Assistants
Add Nodes to an HACMP Cluster
Configure Resources to Make Highly Available
Configure HACMP Resource Groups
Verify and Synchronize HACMP Configuration
HACMP Cluster Test Tool
Display HACMP Configuration
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Game plan
We are going to assume for the moment that these additional steps were done in order
to avoid duplication in the lecture material. We will show the menus for these steps as
part of defining the second resource group for the mutual takeover environment.
For the moment we will assume the Two-Node Cluster Configuration Assistant was
used to accomplish these additional steps.
So, now we have a cluster configured.
Uempty
Where Are We in the Implementation?
9Plan for network, storage, and application
Eliminate single points of failure
9Define and configure the AIX environment
Storage (adapters, LVM volume group, filesystem)
Networks (IP interfaces, /etc/hosts, non-IP)
Application start and stop scripts
9Install the HACMP filesets and reboot
9 Configure the HACMP environment
Topology
Cluster, node names, HACMP IP and non-IP networks
Resources, resource group, attributes:
Resources: Application Server, service label
Resource group: Identify name, nodes, policies
Attributes: Application Server, service label, VG, filesystem
Synchronize
Start HACMP
Test configuration
Save configuration
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Starting Cluster Services (2 of 4)
System Management (C-SPOC)
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Starting Cluster Services (4 of 4)
# smit clstart
Start Cluster Services
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Start now, on system restart or both now +
Start Cluster Services on these nodes [bondar,hudson] +
BROADCAST message at startup? true +
Startup Cluster Information Daemon? true +
Reacquire resources after forced down ? false +
Notes:
Startup choices
There are a few choices to make. For the moment we will just recommend the defaults
except selecting both nodes and turning on the Cluster Information Daemon. A more
detailed discussion of the options are discussed in the next unit.
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Configuring Mutual Takeover
Now let's add a second resource group called adventure,
which uses hudson as the primary node and bondar as the
backup.
X
bondar hudson
X X
A A
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Configure the Resources
smit hacmp -> Initialization and Standard Configuration
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Adding Adventure Service Label (1 of 3)
Add a Service IP Label/Address (standard)
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* IP Label/Address [] +
* Network Name [] +
+--------------------------------------------------------------------------+
IP Label/Address
Move cursor to desired item and press Enter.
(none) ((none))
bondar (192.168.5.29)
hudson (192.168.5.31)
yweb (192.168.5.70)
xweb (192.168.5.92)
F1=Help F2=Refresh F3=Cancel
F1 F8=Image F10=Exit Enter=Do
F5 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
+--------------------------------------------------------------------------+
Network Name
Move cursor to desired item and press Enter.
net_ether_01 (192.168.15.0/24 192.168.16.0/24)
F1=Help F2=Refresh F3=Cancel
F1 F8=Image F10=Exit Enter=Do
F5 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Uempty
Adding Adventure Service Label (3 of 3)
Add a Service IP Label/Address (standard)
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* IP Label/Address [yweb] +
* Network Name [net_ether_01] +
Notes:
Menu filled in
This screen shows the parameters for the Adventure resource groups service IP label.
Once were sure that this is what we intend to do, press Enter to define the service IP
label. The label is then available from a pick list when you add resources to a resource
group later.
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Add Adventure Application Server (2 of 2)
Add Application Server
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Server Name [ywebserver]
* Start Script [/usr/local/scripts/startyweb]
* Stop Script [/usr/local/scripts/stopyweb]
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The start script should then start the application. It should then probably wait until it is
sure that the application has completely started.
The stop scripts responsibility is to stop the application. It must not exit until the
application is totally stopped as HACMP will start to unmount filesystems and release
other resources as soon as the stop script terminates. The attempt to release these
resource might fail if there are remnants of the application still running.
The start and stop scripts must exist on all cluster nodes defined in the resource group
(that is, they must reside on a local non-shared filesystem) or you will not be able to
verify and synchronize the cluster.
HACMP 5.2 and later provides a file collection facility to help keep the start and stop
scripts in synch.
Uempty
Adding the Second Resource Group
smit hacmp -> Initialization and Standard Configuration
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
+--------------------------------------------------------------------------+
Select a Resource Group
Move cursor to desired item and press Enter.
xwebserver_group
| adventure
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Synchronize and Test the Changes
Initialization and Standard Configuration
Move cursor to desired item and press Enter.
Two-Node Cluster Configuration Assistant
Add Nodes to an HACMP Cluster
Configure Resources to Make Highly Available
Configure HACMP Resource Groups
Verify and Synchronize HACMP Configuration
HACMP Cluster Test Tool
Display HACMP Configuration
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
verification errors butt only if you are using Extended Configuration. Deciding to do so is
a decision which must be approached with the greatest of care as it is VERY unusual
for a verification error to occur which can be safely overridden.
Also, remember the earlier discussion about synchronization - any HACMP
configuration changes made on any other cluster node will be lost if you complete a
synchronization on this cluster node.
Uempty
Extended Configuration
Extended Configuration
Move cursor to desired item and press Enter.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Communication Interfaces and Devices
Configure HACMP Communication Interfaces/Devices
Move cursor to desired item and press Enter.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-45
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
+--------------------------------------------------------------------------+
Select a category
Move cursor to desired item and press Enter.
Add Discovered Communication Interface and Devices
Add Predefined Communication Interfaces and Devices
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Don't risk a potentially catastrophic partitioned cluster by using cheap rs232 cables!
Notes:
Uempty
Defining a Non-IP Network (2 of 3)
Configure HACMP Communication Interfaces/Devices
Move cursor to desired item and press Enter.
Add Communication Interfaces/Devices
Change/Show Communication Interfaces/Devices
Remove Communication Interfaces/Devices
Update HACMP Communication Interface with Operating System
Settings
+--------------------------------------------------------------------------+
Select a category
Move cursor to desired item and press Enter.
# Discovery last performed: (Feb 12 18:20)
Communication Interfaces
Communication Devices
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Is it an interface or a device?
Now we need to indicate whether we are adding a communication interface or a
communication device. Non-IP networks use communication devices as end-points
(dev/tty for example) so select Communication Devices to continue.
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-47
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Were now presented with a list of the discovered communication devices.
You can either choose to add an rs232 (using the /dev/tty entries) network or a diskhb
network (using the /dev/hdisk entries). We will cover SSA in the next unit.
rs232 networks
a. /dev/tty1 on bondar is connected to /dev/tty1 on hudson using a fully wired rs232
null-modem cable (dont risk a potentially catastrophic partitioned cluster by failing to
configure a non-IP network or by using cheap cables). Select these two devices and
press Enter to define the network.
b. Before you use this smit screen to define the non-IP network, make sure that you
verify that the link between the two-nodes is actually working.
c. For our example, the non-IP rs232 network connecting bondar to hudson can be
tested as follows:
i. Issue the command stty < /dev/tty1 on one node. The command should
hang.
Uempty ii. Issue the command stty < /dev/tty1 on the other node. The command
should immediately report the ttys status and the command that was hung on
the first node should also immediately report its ttys status.
iii. These commands should not be run while HACMP is using the tty.
iv. If you get the behavior described above (especially including the hang in the first
step that recovers in the second step) then the ports are probably connected
together properly (check the HACMP log files once the cluster is up to be sure). If
you get any other behavior then you probably are using the wrong cable or the
rs232 cable isnt connected the way that you think it is).
diskhb networks
a. Make sure you choose a pair of entries (such as /dev/hdisk5 shown in the figure),
one for each of two-nodes. Note that it is actually the pvids that must match.
b. You can test the connection using the command /usr/sbin/rsct/bin/dhb_read as
follows:
- On Node A, enter: dhb_read -p hdisk5 -r
- On Node B, enter: dhb_read -p hdisk5 -t
- You should then see on both nodes: Link operating normally
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-49
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Defining Persistent Node IP Labels (2 of 3)
Configure HACMP Persistent Node IP Label/Addresses
Move cursor to desired item and press Enter.
Add a Persistent Node IP Label/Address
Change / Show a Persistent Node IP Label/Address
Remove a Persistent Node IP Label/Address
+--------------------------------------------------------------------------+
Select a Node
Move cursor to desired item and press Enter.
bondar
hudson
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
/=Find n=Find Next
+--------------------------------------------------------------------------+
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-51
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Synchronize
smitty hacmp -> Extended Configuration
[Entry Fields]
* Verify, Synchronize or Both [Both] +
* Automatically correct errors found during [No] +
verification?
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-53
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Force synchronization if verification fails? - this is almost always a very bad idea.
Make sure that you really and truly MUST set this option to Yes before doing so.
Verify changes only? - setting this option to Yes will cause the verification to focus on
aspects of the configuration which changed since the last synchronization. As a result,
the verification will run slightly faster. This might be useful during the mid to early stages
of cluster configuration. It seems rather risky once the cluster is in production.
Logging - you can increase the amount of logging related to this verification and
synchronization by setting this option to Verbose. This can be quite useful if you are
having trouble figuring out what is going wrong with a failed verification.
Uempty
Save Configuration: snapshot
Snapshot Configuration
Move cursor to desired item and press Enter.
Add a Cluster Snapshot
Change/Show a Cluster Snapshot
Remove a Cluster Snapshot
Apply a Cluster Snapshot
Configure Custom Snapshot Method
Convert Existing Snapshot For Online Planning worksheets
Notes:
Creating a snapshot
smit hacmp --->Extended Configuration --->Snapshot Configuration
A snapshot captures the HACMP ODM files which allows you to recover the cluster
definitions. There is also an info file. The info file is discussed further in the AU57
course HACMP Administration II: Administration and Troubleshooting.
If necessary there is, from the Snapshot Configuration menu, another option to restore
(apply) a snapshot.
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-55
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
* File Name [/var/hacmp/log/cluster.haw] /
Cluster Notes []
Snapshot Configuration
Move cursor to desired item and press Enter.
Add a Cluster Snapshot
Change/Show a Cluster Snapshot
Remove a Cluster Snapshot
Apply a Cluster Snapshot
Configure Custom Snapshot Method
Convert Existing Snapshot For Online Planning Worksheets
Copyright IBM Corporation 2005
Notes:
Uempty
Removing a Cluster
Use Extended Topology Configuration
Notes:
Starting over
If you have to start over, you can:
- Stop cluster services on all nodes
- Use Extended Configuration, as shown above to remove the cluster (on all nodes)
- Remove the entries (but not the file) from /usr/es/sbin/cluster/etc/rhosts (on all
nodes)
If you really want to start over, then you can:
- installp -u cluster
- rm -r /usr/es/* (be very careful here)
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-57
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
We're There!
We've configured a two-node cluster with two resource
groups:
Each resource group has a different home (primary) node
Each resource group falls back to its home node on recovery
This is called a two-node mutual takeover cluster
bondar hudson
D D
A A
Notes:
Uempty
Checkpoint
1. True or False?
It is possible to configure a recommended simple two-node cluster
environment using just the standard configuration path.
2. In which of the top level HACMP menu choices is the menu for starting and
stopping cluster nodes?
a. Initialization and Standard Configuration
b. Extended Configuration
c. System Management (C-SPOC)
d. Problem Determination Tools
3. In which of the top level HACMP menu choices is the menu for defining a non-
IP heartbeat network?
a. Initialization and Standard Configuration
b. Extended Configuration
c. System Management (C-SPOC)
d. Problem Determination Tools
4. True or False?
It is possible to configure HACMP faster by having someone help you on
the other node.
5. True or False?
You must specify exactly which filesystems you want mounted when you
put resources into a resource group.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-59
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Break Time!
Notes:
Some notes from the developer :-)
This is a photograph of Lake Louise in the Canadian Rocky Mountains (located about a
90 minute drive west of Calgary). If you are ever there, make sure that you rent one of
the canoes in the photograph and go for a paddle out on the lake. Theres also a
number of quite spectacular and not particularly strenuous hikes that start from near the
point that this photograph was taken. The hike that goes up to the tea house is
definitely worth an afternoon (you can pay money to go up on horseback if you dont
feel like walking for free).
Also, can you read this? ;-)
Aoccdrnig to a rscheearchr at an Elingsh uinervtisy, it deosn't mttaer in waht oredr the
ltteers in a wrod are, the olny iprmoetnt tihng is taht the frist and lsat ltteer is at the rghit
pclae. The rset can be a toatl mses and you can sitll raed it wouthit porbelm. Tihs is
bcuseae we do not raed ervey lteter by it slef but the wrod as a wlohe.
Uempty
Unit Summary
Notes:
Copyright IBM Corp. 1998, 2005 Unit 6. Initial Cluster Configuration 6-61
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
References
SC23-4867-05 HACMP for AIX: HACMP Master Glossary
SC23-4864-06 HACMP for AIX: Concepts and Facilities Guide
SC23-4861-06 HACMP for AIX: Planning and Installation Guide
SC23-4862-06 HACMP for AIX: Administration Guide
SC23-5177-00 HACMP for AIX: Troubleshooting Guide
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Topic 1: Topology and resource group management
Use the SMIT Standard and Extended menus to make topology
and resource group changes
Topic 2: Cluster Single Point of Control
Describe the benefits and capabilities of C-SPOC
Perform routine administrative changes using C-SPOC
Start and stop cluster services
Perform resource group move operations
Topic 3: Dynamic Automatic Reconfiguration Event facility
Discuss the benefits and capabilities of DARE
Use the snapshot facility to return to a previous cluster
configuration or to rollback changes
Topic 4: Implementing Web SMIT
Configure and use Web SMIT
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 1 Objectives:
Topology and Resource Group Management
Figure 7-2. Topic 1 Objectives: Topology and Resource Group Management AU546.0
Notes:
Uempty
Yet Another Resource Group
The users have asked that a third application be added to the
cluster
The application uses very little CPU or memory and there's
money in the budget for more disk drives in the disk enclosure
Minimizing downtime is particularly important for this application
The resource group is called ballerina (nobody seems to know
why)
bondar hudson
D D
A A
B B
Notes:
Introduction
Were now going to embark on a series of hypothetical scenarios to illustrate a number
of routine cluster administration tasks. Some of these scenarios are more realistic than
others.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
* Resource Group Name [ballerina]
* Participating Node Names (Default Node Priority) [bondar hudson] +
Does the order in which the node names are specified matter?
Copyright IBM Corporation 2005
Notes:
Uempty
Adding a Third Service IP Label (1 of 2)
The extended configuration path screen for adding a service IP
label provides more options. We choose those which mimic the
standard path.
+--------------------------------------------------------------------------+
Select a Service IP Label/Address type
Move cursor to desired item and press Enter.
Configurable on Multiple Nodes
Bound to a Single Node
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Introduction
We need to define a service IP label for the ballerina resource group.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Network Name
The next step is to associate this Service Label with one of the HACMP networks. This
is not shown in the visual.
Uempty
Adding a Third Service IP Label (2 of 2)
The Alternate Hardware Address ... field is used for hardware address takeover
(which we'll configure later).
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Adding Resources to the Third RG (1 of 2)
The extended path's SMIT screen for updating the contents of a
resource group is MUCH more complicated!
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Synchronize Your Changes
The extended configuration path provides verification and
synchronization options.
HACMP Verification and Synchronization
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Verify, Synchronize or Both [Both] +
Force synchronization if verification fails? [No] +
* Verify changes only? [No] +
* Logging [Standard] +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
D D D
A A A
B B
Notes:
Uempty
Adding a New Cluster Node
1. Physically connect the new node
Connect to IP networks
Connect to the shared storage subsystem
Connect to non-IP networks to create a ring encompassing all nodes
2. Configure the shared volume groups on the new node
3. Add the new node's IP labels to /etc/hosts on one existing node
4. Copy /etc/hosts from this node to all other nodes
5. Install AIX, HACMP and application software on the new node:
Install patches required to bring the new node up to the same level as the
existing cluster nodes
Reboot the new node (always reboot after installing or patching HACMP)
6. Add the new node to the existing cluster (from one of the
existing nodes)
7. Add non-IP networks for the new node
8. Synchronize your changes
9. Start HACMP on the new node
10.Add the new node to the appropriate resource groups
11.Synchronize your changes again
12.Run through your (updated) test plan
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty separated.
This path will be taken to initiate communication with the node.
The command launched by this SMIT screen contacts the clcomd at each address and
asks them to come together in a new cluster. Obviously, HACMP must already be
installed on the new node(s).
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Add Node -- Extended Path
Add a Node to the HACMP Cluster
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Node Name [jones]
Communication Path to Node [jones_if1] +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Introduction
This visual, and the next one, show how to add two more non-IP networks to our cluster.
Make sure that the topology of the non-IP networks that you describe to HACMP
corresponds to the actual topology of the physical rs232 cables.
In the following notes, we discuss why we need to add two more non-IP links.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Synchronize Your Changes
HACMP Verification and Synchronization
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Verify, Synchronize or Both [Both] +
Force synchronization if verification fails? [No] +
* Verify changes only? [No] +
* Logging [Standard] +
Notes:
Synchronize
Once weve synchronized our changes, the jones node is an official member of the
cluster.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Add the Node to a Resource Group
Change/Show a Resource Group
[Entry Fields]
Resource Group Name adventure
New Resource Group Name []
Participating Node Names (Default Node Priority) [hudson bondar jones] +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Shrinking the Cluster
The Auditors aren't impressed with the latest investment
and force the removal of the jones node from the cluster so that it
can be transferred to a new project (some users suspect that
political considerations may have been involved)
D D
A A
B B
Notes:
Removing a node
In this scenario, we take a look at how to remove a node from an HACMP cluster.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Removing a node
While removing a node from a cluster is another fairly involved process, some of the
work has little if anything to do with HACMP itself.
Uempty
Removing an Application
The zwebserver application has been causing problems and a
decision has been made to move it out of the cluster
bondar hudson
D D
A A
B B
Notes:
Removing an application
In this scenarios, we get to remove a resource group
It looks like this imaginary organization could do with a bit more long range planning.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Introduction
The procedure for removing a resource group is actually fairly straightforward.
Cluster snapshot
HACMP supports something called a cluster snapshot. This would be an excellent time
to take a cluster snapshot, just in case we decide to go back to the old configuration.
We will discuss snapshots later in this unit.
Uempty the case of shared volume groups, tie up physical resources which could presumably
be better used elsewhere.
A cluster should not have any useless resources or components as anything which
simplifies the cluster tends to improve availability by reducing the likelihood of human
error.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
+--------------------------------------------------------------------------+
Select a Resource Group
Move cursor to desired item and press Enter.
adventure
ballerina
discovery
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Uempty
Removing a Resource Group (3 of 3)
+--------------------------------------------------------------------------+
ARE YOU SURE?
Continuing may delete information you may want
to keep. This is your last chance to stop
before continuing.
Press Enter to continue.
Press Cancel to return to the application.
F1=Help F2=Refresh F3=Cancel
F1 F8=Image F10=Exit Enter=Do
F9+--------------------------------------------------------------------------+
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Synchronize
As we will discuss later in this unit, its a good idea to synchronize after each change,
rather than making many changes and synchronizing once at the end.
Uempty
Implementing Target Mode SSA
The serial cable being used to implement the rs232 non-IP network
has been borrowed by someone and nobody noticed
A decision has been made to implement a target mode SSA (tmssa)
non-IP network as it won't fail unless complete access to the shared
SSA disks is lost by one of the nodes (and someone is likely to
notice that)
bondar hudson
D D
A A
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Use the smitty ssaa fastpath to get to AIX's SSA Adapters menu.
Copyright IBM Corporation 2005
Notes:
Required software
Target mode SSA support requires that the devices.ssa.tm.rte file set be installed on
all cluster nodes.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Introduction
Once each node has a unique SSA node number, the AIX configuration manager needs
to be used to define the tmssa devices. Each node must have tmssa devices which
refer to each of the other nodes that they can see via the SSA loops. When cfgmgr is
run on a node, it sets up the node to accept tmssa packets, and it then defines tmssa
devices referring to any other nodes which respond to tmssa packets. In order for this to
all work, the other nodes must all be set up to accept and respond to tmssa packets.
Procedure
The end result is that the following procedure gets all the required tmssa devices
defined:
Uempty 1. Run cfgmgr on each cluster node in turn. This sets up each node to handle tmssa
packets, and defines the tmssa devices on each node to refer to nodes which have
already been setup for tmssa.
2. Run cfgmgr on each node in turn again (depending upon exactly what order you do this
in, it is actually possible to skip running cfgmgr on one of the nodes, but it is probably
not worth the trouble of being sure that the last cfgmgr run wasnt required).
3. Verify the tmssar devices exist:
Run
# lsdev -C | grep tmssa
on each node. There should be a tmssar device (which is actually a target mode SSA
router acting as a pseudo device) configured on each node.
4. Verify the tmssa devices exist:
Run
# ls /dev/tmssa*
on each node. Note that each node has target mode SSA devices called
/dev/tmssa#.im and /dev/tmssa#.tm where # refers to the other nodes node number.
5. Test the target mode connection:
Enter the following command on the node with id 1 (make sure you specify the tm suffix
and not the im suffix):
# cat < /dev/tmssa2.tm
(This command should hang)
On the node with ID 2, enter the following command (make sure that you specify the im
suffix and not the tm suffix):
# cat /etc/hosts > /dev/tmssa1.im
(The /etc/hosts file should be displayed on the first node)
This validates that the target mode serial network in functional. Please note that any
text file may be substituted for /etc/hosts and you have to specify different tmssa
device names if you configured different SSA node numbers for each node. This is
simply an example.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Extended Configuration
Move cursor to desired item and press Enter.
Discover HACMP-related Information from Configured Nodes
Extended Topology Configuration
Extended Resource Configuration
Extended Event Configuration
Extended Performance Tuning Parameters Configuration
Security and Users Configuration
Snapshot Configuration
Extended Verification and Synchronization
Notes:
HACMP discover
By discovering the new devices, they will appear in SMIT pick lists when we configure
the tmssa non-IP network. Strictly speaking, it is not necessary to rerun the HACMP
discovery as it is possible to configure tmssa networks by entering in the tmssa device
names explicitly. As this is a rather error-prone process, it is probably best to use the
HACMP discovery mechanism to discover the devices for us.
Uempty
Defining a Non-IP tmssa Network (1 of 3)
This should look very familiar as it is the same procedure that was
used to define the non-IP rs232 network earlier.
Configure HACMP Communication Interfaces/Devices
Move cursor to desired item and press Enter.
Add Communication Interfaces/Devices
Change/Show Communication Interfaces/Devices
Remove Communication Interfaces/Devices
Update HACMP Communication Interface with Operating System Settings
+--------------------------------------------------------------------------+
Select a category
Move cursor to desired item and press Enter.
Add Discovered Communication Interface and Devices
Add Predefined Communication Interfaces and Devices
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
+--------------------------------------------------------------------------+
Select a category
Move cursor to desired item and press Enter.
# Discovery last performed: (Feb 12 18:20)
Communication Interfaces
Communication Devices
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Uempty
Defining a Non-IP tmssa Network (3 of 3)
Now, we need to define the tmssa network using a process
Configure HACMP Communication Interfaces/Devices
Move cursor to desired item and press Enter.
Add Communication Interfaces/Devices
+--------------------------------------------------------------------------+
Select Point-to-Point Pair of Discovered Communication Devices to Add
Move cursor to desired item and press F7. Use arrow keys to scroll.
ONE OR MORE items can be selected.
Press Enter AFTER making all selections.
# Node Device Device Path Pvid
> hudson tmssa1 /dev/tmssa1
> bondar tmssa2 /dev/tmssa2
bondar tty0 /dev/tty0
hudson tty0 /dev/tty0
bondar tty1 /dev/tty1
hudson tty1 /dev/tty1
F1=Help F2=Refresh F3=Cancel
F7=Select F8=Image F10=Exit
F1 Enter=Do /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Final step
Select the tmssa devices on each node and press Enter to define the network.
Refer to Chapter 13 of the HACMP v5.3 Planning and Installation Guide for information
on configuring all supported types of non-IP networks.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Removing a Cluster
Use Extended Topology Configuration
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-45
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-47
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 2 Objectives:
Cluster Single Point of Control
After completing this topic, you should be able to:
Discuss the need for change management when using
HACMP
Describe the benefits and capabilities of C-SPOC
Perform routine administrative changes using C-SPOC
Start and stop cluster services
Perform resource group move operations
Notes:
Uempty
Administering a High Availability Cluster
Administering a HA cluster is different from administering a
stand-alone server:
Changes made to one node need to be reflected on the
other node
Poorly considered changes can have far reaching
implications
Beware the law of unintended consequences
Aspects of the clusters configuration could be quite
subtle and yet critical
Scheduling downtime to install and test changes can be
challenging
Saying oops while sitting at a cluster console could get
you fired!
Notes:
Introduction
You must develop good change management procedures for managing an HACMP
cluster. As you will see, C-SPOC utilities can be used to help, but do not do the job by
themselves. Having well documented and tested procedures to follow, as well as
restricting who can make changes, (for example you should not have more than two or
three persons with root privileges) minimizes loss of availability when making changes.
The snapshot utility should be used before any change is made.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-49
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Recommendations
Implement and adhere to a change control/management
process
Wherever possible, use HACMP's C-SPOC facility to make
changes to the cluster (details to follow)
Document routine operational procedures in a step-by-step
list fashion (for example, shutdown, startup, increasing size
of a filesystem)
Restrict access to the root password to trained High
Availability cluster administrators
Always take a snapshot (explained later) of your existing
configuration before making a change
Notes:
Uempty
Change Control or Change Management
A real change control or management process requires a
serious commitment on the part of the entire organization:
Every change must be carefully considered
The onus should be on the requester of the change to
demonstrate that it is necessary
Not on the cluster administrators to demonstrate that it is unwise.
Management must support the process
Defend cluster administrators against unreasonable request or
pressure
Not allow politics to affect a change's priority or schedule
Every change, even the minor ones, must follow the
process
The cluster administrators must not sneak changes past the process
The notion that a change might be permitted without following the
process must be considered to be absurd
Notes:
Introduction
The visual describes some of the serious commitments that are required to successfully
implement a highly available cluster over time.
Many organizations have little or no experience with the sort of long-term commitment
and discipline which is required to successfully manage a highly available cluster over
the long term. The temptation to slip a little change through might seem to be worth the
risk (it almost certainly is not worth the risk) and political considerations can make it
very difficult to withstand pressure from certain interest groups for IMMEDIATE
CHANGES.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-51
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Choices
There is a choice to be made:
- There is a real long-term commitment by all parties to the change control process (or
the management backbone to stand up to those who lack the commitment).
This path has a strong likelihood of leading to a stable cluster with a satisfied user
community.
- Without a real long-term commitment, the change control process rapidly becomes
a farce.
This second path leads to disaster (the question is when, not if, disaster will strike).
Uempty
Change Considerations
Every change must be carefully considered:
Is the change necessary?
How urgent is the change?
How important is the change? (not the same thing as urgent)
What impact does the change have on other aspects of the
cluster?
What is the impact if the change is not allowed to occur?
Are all of the steps required to implement the change clearly
understood and documented?
How is the change to be tested?
What is the plan for backing out the change if necessary?
Is the appropriate expertise available should problems develop?
When is the change scheduled?
Have the users been notified?
Does the maintenance period include sufficient time for a full set
of backups prior to the change and sufficient time for a full
restore afterwards should the change fail testing?
Copyright IBM Corporation 2005
Notes:
Introduction
Develop a process which encompasses at least the above set of issues. This process
may and probably should include a form which requires appropriate approvals before
the change can go ahead.
Maintenance window
The point about requiring sufficient time during the scheduled outage window for a full
backup, the change, the testing and a full restore may seem a little extreme. Suffice it to
say that there are real customer organizations out there that consider the alternative to
be utter lunacy.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-53
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Continuous
Availability
Continuous High
Operations Availability
Notes:
Introduction
As you can see, HACMP addresses two aspects of availability: unplanned downtime
and planned downtime.
Planned downtime
Planned downtime is managed using one of following three methods:
- C-SPOC --
for AIX definition changes (such as users, disk, LVM) which have to be
made/modified on one node and then propagated to the other node
- C-SPOC: Resource Group and Application management --
for manually moving a resource group offline/online or to another node
Uempty - Synchronization/DARE --
for HACMP definition (configuration) changes where the HACMP ODM files must be
updated on all the nodes
In this topic, we address the first two methods. DARE is addressed in a later topic in this
unit.
The end result is overall continuous availability.
Without C-SPOC functionality, the system administrator must spend time executing
administrative tasks individually on each cluster node. For example, to add a user you
usually must perform this task on each node. With C-SPOC, a command executed on
one node is also executed on the other nodes. Thus, the use of C-SPOC minimizes
administrative overhead (repetition) and reduces the possibility of inconsistent node
states.
Unplanned downtime
Unplanned downtime is addressed via HACMP monitoring and resource group fallover
processing.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-55
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Initiating
node
Target Target
node node
Notes:
HACMP 4.X
In order to use rsh, you must create the ./rhosts file, adding the cluster IP Labels for all
cluster nodes. If kerberos has been implemented the ./rhosts file is not required.
HACMP 5.X
The clcomdES subsystem provides secure communications between nodes, /.rhosts or
Kerberos is no longer required. This daemon provides secure communication between
cluster nodes for all cluster utilities such as verification and synchronization and system
Uempty management (C-SPOC). The clcomd daemon is started automatically at boot time by
the init process.
More details
All the nodes in the resource group must be available or the C-SPOC command will not
be successful on any node (all or none).
As you saw in the LVM unit, LVM changes, if made through C-SPOC, are synchronized
automatically.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-57
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Adding a User to the Cluster (1 of 2)
Add a User to the Cluster
[Entry Fields]
Select nodes by Resource Group [] +
*** No selection means all nodes! ***
Notes:
User management
For users of an HACMP for AIX cluster, system administrators must create duplicate
accounts on each cluster node. The user account information stored in the /etc/passwd
file and in other files stored in the /etc/security directory should be consistent on all
cluster nodes. For example, if a cluster node fails, users should be able to log on to the
surviving nodes without experiencing problems caused by mismatches in the user or
group IDs.
System administrators typically keep user accounts synchronized across cluster nodes
by copying the key system account and security files to all cluster nodes whenever a
new account is created or an existing account is changed. For C-SPOC clusters, the
C-SPOC utility simplifies the cluster-wide synchronization of user accounts by
propagating the new account or changes to an existing account across all cluster nodes
automatically.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-59
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The SMIT screens for managing users are found under HACMP Security and Users
Management in the top-level C-SPOC menu.
Uempty
Adding a User to the Cluster (2 of 2)
Add a User to the Cluster
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-61
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
Select nodes by Resource Group
*** No selection means all nodes! ***
Notes:
AUTHENTICATION information
Remove AUTHENTICATION information indicates if the system should delete the user's
password and other user authentication information from the /etc/security/passwd file.
To change this value, use the Tab key to toggle the Yes/No values.
Uempty
Passwords in an HACMP Cluster
Passwords in an HACMP cluster
Notes:
Introduction
C-SPOC provides a utility to change user passwords across the cluster. As usual when
managing passwords, there are a number of considerations.
NIS or DCE
If you manage user accounts with a utility such as Network Information Service (NIS) or
Distributed Computing Environment (DCE) Manager, do not use HACMP user
management. Using HACMP user management in this environment might cause
serious system inconsistencies in the database.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-63
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Security
The security of the password propagated to other nodes is only as secure as the
network used to distribute the password. (In other words, the passwords are sent in
clear text.)
Uempty Both of these call the AIX 5L passwd command. The clpasswd command uses the
same arguments as the passwd command. For more information about the clpasswd
command, see its man page.
How it works
The table that follows is the expected behavior for non-root users either running the
clpasswd command, AIX passwd command or AIX passwd command when linked to
clpasswd.
/bin/passwd linked to
Authorized? /bin/passwd not linked to clpasswd
clpasswd
HACMP clpasswd is
run, or the Change
AIX passwd is run Current Users
Password SMIT
screen is used
The password is The password is The password is
Yes changed on all cluster changed only on the changed on all cluster
nodes. local node. nodes.
The password is The password is
Password not
No changed only on the changed only on the
changed
local node. local node.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-65
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
Node Name(s) to which disk is attached bondar,hudson +
Device type disk +
Disk Type hdisk
Disk interface ssar
Description SSA Logical Disk Drive
Parent ssar
* CONNECTION address [] +
Location Label []
ASSIGN physical volume identifier yes +
RESERVE disk on open yes +
Queue depth [] +
Maximum Coalesce [] +
Notes:
Uempty
Managing Shared LVM Components
Notes:
Introduction
This is the menu for using C-SPOC to perform LVM change management and
synchronization. As was mentioned in the LVM unit, you can make changes in AIX
directly and then synchronize OR. If you make the changes utilizing C-SPOC utilities,
the synchronization is automatic.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-67
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
How it works
Once you create a shared volume group, you must rerun the discovery mechanism
(refer to top-level menu in the enhanced configuration path) to get HACMP to know
about the volume group. You must then add the volume group to a resource group
before you can use C-SPOC to add shared logical volumes or filesystems.
Synchronization
Note that you only need to add the volume group to a resource group using SMIT from
one of the cluster nodes, and then you can start working with C-SPOC from the same
node. You do not need to synchronize the cluster between adding the volume group to a
resource group and working with it using C-SPOC unless you want to use C-SPOC
from some other node. Keep in mind that the volume group is not really be a part of the
resource group until you synchronize the addition of the volume group to the resource
group.
Uempty
Creating a Shared Volume Group
Create a Shared Volume Group
[Entry Fields]
Node Names bondar,hudson
PVID 00055207bbf6edab 0000>
VOLUME GROUP name [bernhardvg]
Physical partition SIZE in megabytes 64 +
Volume group MAJOR NUMBER [207] #
Enable Cross-Site LVM Mirroring Verification false +
HACMP 5.2--
Warning :
Changing the volume group major number may result
in the command being unable to execute
successfully on a node that does not have the
major number currently available. Please check
for a commonly available major number on all nodes
before changing this setting.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-69
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Creating a Shared File System (1 of 2)
First create mirrored logical volumes for the filesystem and jfslog.Do not forget to
logform the jfslog logical volume
The volume group must be online somewhere and listed in a resource group or it does not appear in
the pop-up list
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-71
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Creating a Shared File System (2 of 2)
Then create the filesystem in the now "previously defined logical volume"
[Entry Fields]
Node Names bondar,hudson
LOGICAL VOLUME name norbertfs
* MOUNT POINT [/norbert]
PERMISSIONS read/write +
Mount OPTIONS [] +
Start Disk Accounting? no +
Fragment Size (bytes) 4096 +
Number of bytes per inode 4096 +
Allocation Group Size (MBytes) 8 +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-73
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
VGDA = ODM
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-75
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
#
#importvg -V123 -L sharedvg hdisk3
#chvg -an sharedvg
#varyoffvg sharedvg
Notes:
After making a change to an LVM component such as increasing the size of a file system
as shown in the figure, you must propagate the change to the other nodes in the cluster
which are sharing the volume group using the steps above. Make sure that the auto
activate is turned off (chvg -an sharedvg) after the importvg command is executed
since the cluster manager will control the use of the varyonvg command on the node where
the volume group should be varied on.
Other than the sheer complexity of this procedure, the real problem with it is that it requires
that the resource group be down while the procedure is being carried out.
Fortunately, there are better ways...
Uempty
LVM Changes, Lazy Update
At fallover time, lazy update compares the time stamp value
in the VGDA with one stored in
/usr/sbin/cluster/etc/vg/<vgname>. If the time stamps are
the same, then the varyonvg proceeds.
If the timestamps do not agree, then HACMP does the
export/import cycle similar to a manual update.
NOTE: HACMP does change the VG auto vary on flag
AND it preserves permissions and ownership of the logical
volumes.
12 12
11 1 11 1
10 2 10 2
9 3 9 3
8 4 8 4
7 5 7 5
6 6
Notes:
HACMP has a facility called Lazy Update that it uses to ensure LVM changes are updated
during a fallover.
This works by HACMP keeping a copy of the timestamp from the volume groups VGDA.
AIX updates this timestamp whenever the LVM component is modified. When another
cluster node attempts to vary on the volume group, HACMP for AIX compares its copy of
the timestamp (kept in the /usr/sbin/cluster/etc/vg file) with the timestamp in the VGDA on
the disk. If the values are different, the HACMP for AIX software exports and re-imports the
volume group before activating it. If the timestamps are the same, HACMP for AIX
activates the volume group without exporting and re-importing. The time needed for
takeover expands by a few minutes if a Lazy Update occurs. A Lazy Update is always
performed the first time a takeover occurs in order to create the timestamp file on the
takeover node.
This method requires no downtime although, as indicated above, it does increase the
fallover time somewhat for the first fallover after the LVM change was made.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-77
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
The Best Method: C-SPOC LVM Changes
Notes:
You can use C-SPOC to both make the change and to distribute the change.
This approach has two major advantages: no downtime is required and you can be
confident that the nodes are in sync. It might take a little longer to run than the normal chfs
application, but it is well worth the wait.
Other C-SPOC screens exist for pretty much any operation that you are likely to want to do
with a shared volume group.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-79
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
LVM Changes, Select Your Filesystem
Journaled File Systems
+--------------------------------------------------------------------------+
File System Name
Move cursor to desired item and press Enter.
# Resource Group File System
adventure /norbert
discovery /ron
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-81
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
Resource Group Name discovery
File system name /ron
NEW mount point [/ron]
SIZE of file system [4000000]
Mount GROUP []
Mount AUTOMATICALLY at system restart? no +
PERMISSIONS read/write +
Mount OPTIONS [] +
Start Disk Accounting? no +
Fragment Size (bytes) 4096
Number of bytes per inode 4096
Compression algorithm no
Notes:
Uempty
Stopping Cluster Services
[Entry Fields]
* Stop now, on system restart or both now +
Stop Cluster Services on these nodes [hanode1] +
BROADCAST cluster shutdown? true +
* Shutdown mode graceful +
Shutdown mode
Move cursor to desired item and press Enter.
graceful
takeover
forced
F1=Help F2=Refresh F3=Cancel
F1 F8=Image F10=Exit Enter=Do
F5 /=Find n=Find Next
F9
Notes:
Shutdown mode
When stopping cluster services, you need to tell HACMP exactly what you want it to do:
- Graceful
Local machine shuts itself down gracefully. Remote machine(s) interpret this as a
graceful down and DO NOT takeover resources.
- Takeover
Local machine shuts itself down gracefully. Remote machine(s) interpret this as a
non-graceful down and TAKEOVER resources. This mode is useful for system
maintenance.
- Forced
Local machine shuts down cluster services without releasing any resources.
Remote machine(s) DO NOT takeover resources.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-83
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Priority Override Location (POL)
Assigned during a resource group move operation.
The destination node for a resource group online, offline or
move request becomes the resource group's POL
Remains in effect until:
A move to "Restore_Node_Priority_Order" is done
Cluster is restarted (unless option chosen to persist)
HACMP 4.x has the notion of a sticky location which is similar to
the notion of a persistent POL
Can be viewed with the command:
/usr/es/sbin/cluster/utilities/clRGinfo p
*This foil describes how priority override locations work for nonconcurrent
resource groups. Please refer to chapter 14 of the HACMP for AIX 5L
Administration Guide for information on how priority override locations work
for concurrent access resource groups.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-85
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Moving a Resource Group
[Entry Fields]
Resource Group to be Moved adventure
Destination Node hudson
Persist Across Cluster Reboot? false +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-87
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Bring a Resource Group Offline
-
Select an Online Node
Move cursor to desired item and press Enter.
t2toronto
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9-
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-89
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
Resource Group to Bring Offline adventure
Node On Which to Bring Resource Group Offline bondar
Persist Across Cluster Reboot? false +
Notes:
Uempty
Bring a Resource Group Back Online
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-91
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The discussion here refers to the behavior of non-concurrent access resource groups.
Please refer to chapter 14 of the HACMP for AIX Administration Guide for information
on how priority override locations work for concurrent access resource groups.
Uempty
Log Files Generated by HACMP
/usr/es/adm/cluster.log "High level view" of cluster activity.
Notes:
Log files
The visual summarizes the HACMP log files.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-93
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-95
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Figure 7-74. Topic 3 Objectives: Dynamic Automatic Reconfiguration Event Facility AU546.0
Notes:
Uempty
Dynamic Reconfiguration
HACMP provides a facility that allows changes to cluster
topology and resources to be made while the cluster is active. This
facility is known as DARE or to give it it's full name Dynamic
Automatic Reconfiguration Event. This requires three copies of the
HACMP ODM.
Notes:
How it works
Dynamic Reconfiguration is made possible by the fact that HACMP holds three copies
of the ODM, known as the Default, Staging and Active configuration directory. By
holding three copies of the ODM, HACMP can make changes on one node and
propagate them to other nodes in the cluster while an active configuration is currently
being used.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-97
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topology Changes
Adding or removing cluster nodes
Adding or removing networks
Adding or removing communication interfaces or
devices
Swapping a communication interface's IP address
Resource Changes
All resources can be changed
Copyright IBM Corporation 2005
Notes:
Uempty
What Limitations Does DARE Have?
DARE cannot change all cluster topology and resource group
components without the need to stop HACMP, take the application
offline or reboot a node
Here are some examples that require a stop and restart of HACMP
for the change to be made
Topology Changes
Change the name of the cluster
Change the cluster ID*
Change the name of a cluster node
Change a communication interface attribute
Changing whether or not a network uses IPAT via IP aliasing or via IP replacement
Change the name of a network module*
Add a network interface module*
Removing a network interface module*
Resource Changes
Change the name of a resource group
Change the name of an application server
Change the node relationship
DARE cannot run if two nodes are not at the same HACMP level
Copyright IBM Corporation 2005
Notes:
Limitations
Some changes require a HACMP restart.
Also, DARE requires that all nodes are at the same HACMP level.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-99
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
1 2 3 4 5
cluster manager reads SCD is deleted
change topology synchronize topology snapshot taken of
or resources in SMIT or resources in SMIT the current ACD ACD and refreshes
HACMP HACMP
Move cursor to desired item and press Enter. Move cursor to desired item and press Enter.
SCD
F1=Help F2=Refresh F3=Cancel Esc+8=Image F1=Help F2=Refresh F3=Cancel Esc+8=Image
Esc+9=Shell Esc+0=Exit Enter=Do Esc+9=Shell Esc+0=Exit Enter=Do
Type
text
Notes:
How it works
DARE uses three copies of the HACMP ODM to propagate live updates to the cluster
topology or resource configuration across the cluster. This is done in five steps detailed
above. Although it is possible to make a nearly arbitrarily large set of changes to the
configuration and then synchronize them all in one operation, it is usually better to make
a modest change, synchronize it, verify that it works, and then move on to more
changes.
Note that many changes are incompatible with the clusters current AIX configuration.
Such changes are, therefore, not possible to synchronize using DARE. Instead, the
cluster has to be taken down while the appropriate AIX configuration changes are
applied (it is sometimes possible to remove some resources from a resource group,
synchronize, change the AIX configuration of the resources, add them back into the
resource group and synchronize again although there is likely to be little point in running
the resource group without the resources).
Uempty HACMP 5.x synchronizes both topology changes and resource changes whenever it is
run. This is a change from previous releases of HACMP.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-101
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Verifying and Synchronizing (Extended)
HACMP Verification and Synchronization (Active Cluster on a Local Node)
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
(When NODE DOWN -- HACMP 5.2) [Entry Fields]
* Verify, Synchronize or Both [Both] +
* Automatically correct errors found during [No] +
verification?
* Force synchronization if verification fails? [No] +
* Verify changes only? [No] +
* Logging [Standard] +
(When NODE UP)
* Emulate or Actual [Actual] +
Force synchronization if verification fails? [No] +
* Verify changes only? [No] +
* Logging [Standard] +
Notes:
Emulate or Actual
The default of Actual causes the changes being verified and synchronized to take
effect (become the actual cluster configuration) if the verification succeeds. Setting this
field to Emulate causes HACMP to verify and then go through the motions of a
synchronize without actually causing the changes to take effect. This is useful to get a
sense of what side effects the synchronization is likely to result in. For example, if the
proposed change would trigger a fallover or a fallback (because node priorities have
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-103
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Logging
This field can be set to Standard to request the default level of logging or to Verbose to
request a more, ummmm, verbose level of logging! If you are having problems getting a
change to verify and do not understand why it will not verify then setting the logging
level to verbose may provide additional information which proves useful.
HACMP 4.x
HACMP 4.x provided an extra option in the Verify and Configure Resources SMIT
screen called Un/configure Cluster Resources. The default of yes resulted in a
verification and synchronization essentially as described above. A value of no caused
the verification and synchronization to proceed as usual except that the operation
stopped just before the step of loading the new ACD into the current cluster managers
and actually acting on the changes. This resulted in the change being staged. It would
then be caused to take effect by stopping HACMP on all cluster nodes and then
restarting HACMP on each cluster node (stopping and then restarting HACMP on one
cluster node while a staged change existed and while other cluster nodes remain
running would almost certainly result in an inconsistent cluster configuration across
cluster nodes which would, in turn, result in one or more nodes crashing).
This facility was used in a number of rather exotic cluster upgrade scenarios. It is no
longer available in either the standard or extended paths verification and
synchronization mechanisms. It is still available in the SMIT screen used to apply a
cluster snapshot (under Snapshot Configuration in the top level menu of the extended
configuration path).
Uempty Note that a configuration which has been staged as described above blocks all further
synchronizations until either the cluster is brought down (HACMP daemons stopped on
all cluster nodes) and then brought back up again. This is called a dynamic
reconfiguration lock. We see how to remove such a lock shortly.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-105
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Rolling back an unwanted change that has not yet been synchronized
If you have made changes which you have decided to not synchronize, they can be
discarded using the Restore HACMP Configuration Database from Active Configuration
menu entry shown above. It is located under the Problem Determination Tools menu
(accessible from the top-level HACMP SMIT menu).
Prior to rolling back the DCD on all nodes, the current contents of the DCD on the node
used to initiate the roll back is saved as a snapshot (in case they should prove useful in
the future). The snapshot will have a rather long name similar to:
Restored_From_ACD.Sep.18.19.33.58
This name can be interpreted to indicate that the snapshot was taken at 19:33:58 on
September 18th (the year is not preserved in the name).
Since the change being discarded is sometimes a change which has been emulated,
this operation is sometimes called rolling back an emulated change. This is a misnomer
Uempty as the operation rolls back ANY change which has not yet been verified and
synchronized by restoring all nodes DCDs to the contents of the currently active cluster
configuration.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-107
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-109
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
1 2 3 4 5
cluster manager reads SCD is deleted
change topology synchronize topology snapshot taken of
or resources in SMIT or resources in SMIT the current ACD ACD and refreshes
HACMP HACMP
Move cursor to desired item and press Enter. Move cursor to desired item and press Enter.
SCD
F1=Help F2=Refresh F3=Cancel Esc+8=Image F1=Help F2=Refresh F3=Cancel Esc+8=Image
Esc+9=Shell Esc+0=Exit Enter=Do Esc+9=Shell Esc+0=Exit Enter=Do
Type
text
Bang!
Notes:
Uempty different configurations, a situation which results in one or more cluster nodes
crashing).
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-111
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Lets Review: Topic 3
1. True or False?
DARE operations can be performed while the cluster is running.
2. Which operations can DARE not perform (select all that apply)?
a. Changing the name of the cluster.
b. Removing a node from the cluster.
c. Changing a resource in a resource group.
d. Change whether a network uses IPAT via IP aliasing or via IP
replacement.
3. True or False?
It is possible to roll back from a successful DARE operation using an
automatically generated snapshot.
4. True or False?
Running a DARE operation requires three separate copies of the
HACMP ODM.
5. True or False?
Cluster snapshots can be applied while the cluster is running.
6. What is the purpose of the dynamic reconfiguration lock?
a. To prevent unauthorized access to DARE functions.
b. To prevent further changes being made until a DARE operation
has completed.
c. To keep a copy of the previous configuration for easy rollback.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-113
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-115
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Topic 4 Objectives:
Implementing WebSMIT
Notes:
Uempty
Web-enabled SMIT (WebSMIT)
HACMP 5.2 and up includes a web-enabled user interface
that provides easy access to:
HACMP configuration and management functions
Interactive cluster status display and manipulation
HACMP online documentation
The WebSMIT interface is similar to the ASCII SMIT
interface. You do not need to learn a new user interface or
terminology and can easily switch between ASCII SMIT and
WebSMIT
To use the WebSMIT interface, you must configure and run
a Web server process on the cluster nodes to be
administered
Notes:
Introduction
WebSMIT combines the advantages of SMIT with the ease of access from any system
which runs a browser. In addition, it adds a few useful new features, not available in
text-based SMIT.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-117
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Introduction
To connect to WebSMIT, point your browser to the cluster node that you have
configured for WebSMIT.
WebSMIT uses port 42267 by default.
After authentication, this will be the first screen that you see. WebSMIT provides three
basic functions:
- Cluster Status
- Cluster Configuration and Management
- Online Documentation
Uempty
WebSMIT
Cluster Configuration and Management
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-119
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Tree View
If you click the Tree View button, a new window pops up. You can use the Tree View
window to quickly navigate between SMIT panels without having to step through all the
intervening menus. It may also be useful, if you have forgotten where to find a particular
panel.
Uempty
WebSMIT Online Documentation
Notes:
Online Documentation
This screen allows you to view the HACMP manuals in either HTML or PDF format. You
must install the HACMP documentation file sets.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-121
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
WebSMIT Fast Path
Notes:
WebSMIT FastPath
You can use the FastPath function to navigate to almost any SMIT panel. WebSMIT will
process just about any valid SMIT panel. There are exceptions due to the client-server
nature of an http interface, but in general it will process most any SMIT panel.
In the example, we have used the top_menu fast path to navigate to the top of the
SMIT menus.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-123
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
WebSMIT Configuration
/usr/es/sbin/cluster/wsm/README
Configure and run a Web server on cluster nodes
httpd.conf
Copy or link files
Security considerations
wsm_smit.conf
wsm_cmd_exec
Log files
wsm_smit.log
wsm_smit.script
Customizing WebSMIT status display
wsm_clstat.conf
Controlling which SMIT panels can be used
wsm_smit.allow
wsm_smit.deny
wsm_smit.redirect
Setting up WebSMIT online documentation
Install cluster.doc.en_US.es.html and cluster.doc.en_US.es.pdf
Create link
Notes:
Documentation
The primary source for information on configuring WebSMIT is the WebSMIT README
file as shown in the visual. The HACMP Planning and Installation Guide provides some
additional information on installation and the HACMP Administration Guide provides
information on using WebSMIT.
Web server
To use WebSMIT, you must configure one (or more) of your cluster nodes as a Web
server. You can use any Apache compliant server including the IBM HTTP Server
(IBMIHS). Refer to the specific documentation for the Web server you choose.
Once you have setup your server, you need to configure it to work with WebSMIT.
Nearly all modifications to Apache are made in the httpd.conf file. In addition, the Web
server software must be able to access the WebSMIT cgi-bin and htdoc files.
Uempty httpd.conf
If you are using IBMIHS, this file is found in the /usr/HTTPServer/conf directory.
HACMP provides a sample httpd.conf file:
/usr/es/sbin/cluster/wsm/httpd.conf.sample.
You can either copy this file to your Web servers directory, or you can copy relevant
portions to an existing httpd.conf file.
See the README file for details.
WebSMIT cgi-bin and htdocs directories
You can either copy these files to your Web servers directory, or link them. Creating
links is probably preferred since that allows any updates to the WebSMIT software to be
automatically applied. It also keeps the WebSMIT log files in their default location so
that they will be picked up by snap -e.
See the README file for details.
WebSMIT security
Since WebSMIT gives you root access to all the nodes in your cluster, you must
carefully consider the security implications.
WebSMIT uses a configuration file, wsm_smit.conf, that contains settings for
WebSMIT's security related features. This file is installed as
/usr/es/sbin/cluster/wsm/wsm_smit.conf, and it may not be moved to another
location. The default settings used provide the highest level of security in the default
AIX/Apache environment. However, you should carefully consider the security
characteristics of your system before putting WebSMIT to use. It may be possible to use
different combinations of security settings for AIX, Apache, and WebSMIT to improve
the security of the application in your environment.
WebSMIT uses the following configurable mechanisms to implement a secure
environment:
- Non-standard port
- Secure http (https)
- User authentication
- Session time-out
- wsm_cmd_exec setuid program
Use non-standard port
WebSMIT can be configured to allow access only over a specified port using the
wsm_smit.conf AUTHORIZED_PORT setting. If you do not specify an AUTHORIZED_PORT,
or specify a port of 0, then any connections via any port will be accepted. It is strongly
recommended that you explicitly specify the AUTHORIZED_PORT, and that you use a
non-standard port. The default setting for this configuration variable is 42267.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-125
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Log files
All operations of the WebSMIT interface are logged to the wsm_smit.log file and are
equivalent to the logging done with smitty -v. Script commands are also captured in
the wsm_smit.script log file.
WebSMIT log files are created by the CGI scripts using a relative path of <../logs>. If
you copy the CGI scripts to the default location for the IBM HTTP Server, the final path
to the logs is /usr/HTTPServer/logs.
The WebSMIT logs are not subject to manipulation by the HACMP Log Viewing and
Management SMIT panel. Also, just like smit.log and smit.script, the files grow
indefinitely.
The snap -e utility captures the WebSMIT log files if you leave them in the default
location (/usr/es/sbin/cluster/wsm/logs); but if you install WebSMIT somewhere else,
snap -e will not find them.
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-127
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Checkpoint
1. True or False?
A star configuration is a good choice for your non-IP networks.
2. True or False?
Using DARE, you can change from IPAT via aliasing to IPAT via
replacement without stopping the cluster.
3. The ___________ utility to allows users to change their passwords
across all nodes
4. True or False?
RSCT will automatically update /etc/filesystems when using enhanced
concurrent mode volume groups
5. True or False?
A resource groups priority override location can be cancelled by
selecting a destination node of Restore_Node_Priority_Order.
6. The basic steps to configure WebSMIT are:
a. Install a web server and edit the ____________ file to configure it to
server WebSMIT pages
b. Link the WebSMIT _________ and _________ directories to the
web servers directory
c. Edit the _________ file to configure WebSMIT security
d. Set 4511 permissions on the ______________ file.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 7. Basic HACMP Administration 7-129
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Summary
Implementing procedures for change management is a
critical part of administering a HACMP cluster
C-SPOC provides facilities for performing common cluster
wide administration tasks from any node within the cluster
Perform routine administrative changes
Start and stop cluster services
Perform resource group move operations
Start and stop cluster services
The SMIT Standard and Extended menus are used to make
topology and resource group changes
The Dynamic Automatic Reconfiguration Event facility
(DARE) provides the mechanism to make changes to
cluster topology and resources without stopping the cluster
The Cluster Snapshot facility allows the user to save and
restore a cluster configuration
WebSMIT provides access to HACMP SMIT menus from
any system with a web browser
Copyright IBM Corporation 2005
References
SC23-4861-06 HACMP for AIX, Version 5.3 Planning and Installation
Guide
SC23-4862-06 HACMP for AIX, Version 5.3 Administration Guide
Unit Objectives
After completing this unit, you should be able to:
Describe what an HACMP event is
Describe the sequence of events when:
The first node starts in a cluster
A new node joins an existing cluster
A node leaves a cluster voluntarily
Explain what happens when HACMP processes an event
Describe how to customize the event flow
State how to monitor other devices
Notes:
Notes:
Uempty
What is an HACMP Event?
An HACMP event is an incident of interest to HACMP:
A node joins the cluster
A node crashes
A NIC fails
A NIC recovers
Cluster administrator requests a resource group move
Cluster administrator requests a configuration change
(synchronization)
An HACMP event script is a script invoked by a recovery
program to perform the recovery function required.
node_up
node_down
fail_interface
join_interface
rg_move
reconfig_topology_start
Notes:
Recovery Command
Recovery Command
Recovery __
rules.hacmprd TE_JOIN_NODE
0
/usr/sbin/cluster/events/node_up.rp
2
file
0
# 6) Resource variable only used for event manager events
Group Services/ES
Topology Services/ES
Notes:
Uempty
Recovery Programs
reconfig_configuration.rp
cluster_notify.rp reconfig_configuration_dependency_acqui
external_resource_state_change. re.rp
rp reconfig_configuration_dependency_comp
external_resource_state_change lete.rp
_complete.rp reconfig_configuration_dependency_relea
fail_interface.rp se.rp
fail_standby.rp reconfig_resource.rp
join_interface.rp reconfig_topology.rp
join_standby.rp resource_state_change.rp
migrate.rp resource_state_change_complete.rp
network_down.rp rg_move.rp
network_up.rp rg_offline.rp
node_down.rp rg_online.rp
node_down_dependency.rp server_down.rp
node_down_dependency_comple server_restart.rp
te.rp site_down.rp
node_up.rp site_isolation.rp
node_up_dependency.rp site_merge.rp
node_up_dependency_complete. site_up.rp
rp swap_adapter.rp
Notes:
Recovery Programs
This visual lists the recovery programs that are called by the resource manager
component of the cluster manager services. These form the first step in processing an
event.
site_up.rp
# This file contains the HACMP/ES recovery program for site_up events
#
# format:
# relationship command to run expected status NULL
#
other "site_up" 0 NULL
#
barrier
#
event "site_up" 0 NULL
#
barrier
#
all "site_up_complete" 0 NULL
Notes:
Uempty
Event Scripts
(called by recovery programs) (called by other events)
Notes:
Event Scripts
This is the list of HACMP events which are managed by HACMP 5.3.
The events on the left are directly called by recovery programs in response to
unexpected happenings. The events on the right are invoked by primary or other
secondary events on an as-needed basis.
Each of these events can have an optional notify command, one or more pre-event
scripts, one or more post-event scripts and an optional recovery command associated
with it.
r
es
Start
Cluste
servic
1) node_up
lls node_up_local
ca acquire_service_addr
clstrmgrES
RC acquire_takeover_addr
Event
ca l get_disk_vg_fs
Manager ls
RC
2) node_up_complete
node_up_local_complete
start_server
run start script
Copyright IBM Corporation 2005
Notes:
Startup processing
Implicit in this example is the assumption that there is actually a resource group to start
on the node. If there are no resource groups to start on the node, then node_up_local
and node_up_local_complete do very little processing at all.
Uempty
Another Node Joins the Cluster
g
nnin
ru t
ar
clstrmgrES clstrmgrES St ster s
u e
Event Messages Event Cl rvic
e
Manager Manager ca s
ll
ca C
ll
R
R
C
call
1) node_up 2) node_up
ca
RC
node_up_remote node_up_local
ll
RC
stop_server acquire_service_address
run stop script acquire_takeover_address
release_takeover_address get_disk_vg_fs
release_vg_fs
4) node_up_complete
3) node_up_complete node_up_local_complete
node_up_remote_complete start_server
run start script
Copyright IBM Corporation 2005
Notes:
ning p
run Sto ter
s
clstrmgrES clstrmgrES
Clu vices
ca r
Event Event ll se
Manager Messages Manager
RC 1) node_down
ca C
node_down_local
ll
R
stop_server
3) node_down run stop script
ca
node_down_remote release_takeover_addr
ll
ca
acquire_service_addr RC release_vg_fs
ll
RC
Notes:
Node failure
The situation is only slightly different if the node on the right had failed suddenly. Since
it is not in a position to run any events, the events listed under the right hand node do
not get run and the events listed under the left hand node might need to be somewhat
more forceful (for example, the get_disk_vg_fs event may need to break some disk
reservations (see earlier shared storage unit for details).
Uempty
Lets Review
1. True or False?
HACMP 5.x supports a maximum of five pre and five post events per HACMP
event.
2. Which of the following are examples of primary HACMP events (select all that
apply)?
a. node_up
b. node_up_local
c. node_up_complete
d. start_server
e. Rg_up
3. When a node joins an existing cluster, what is the correct sequence for these
events?
a. node_up on new node, node_up on existing node, node_up_complete on new
node, node_up_complete on existing node
b. node_up on existing node, node_up on new node, node_up_complete on new
node, node_up_complete on existing node
c. node_up on new node, node_up on existing node, node_up_complete on existing
node, node_up_complete on new node
d. node_up on existing node, node_up on new node, node_up_complete on existing
node, node_up_complete on new node
Notes:
Notes:
In this topic, we examine how to customize events in HACMP.
Uempty
Event Processing Customization
Notify Command
Event Manager
Recovery
clcallev HACMP Event
HACMP Event Command
No Counter Yes
RC=0 >0
ODM Yes No
HACMP-
Classes
Boom!
Post-Event Script (1) Post-Event Script (n)
Notify Command
Notes:
Uempty
Adding/Changing Cluster Events (1 of 3)
Extended Event Configuration
Notes:
Notes:
[Entry Fields]
Event Name
node_down
Description Script run after the >
* Event Command [/usr/es/sbin/cluster/>
Notify Command []
Pre-event Command [] +
Post-event Command [stop_printq] +
Recovery Command []
* Recovery Counter [0] #
Notes:
Uempty
Recovery Commands
If an event script should fail to exit 0, Recovery commands can be
executed if an event script does not exit 0.
Recovery
HACMP Event
Command
No Counter Yes
RC=0
>0
Yes
Notes:
Notes:
Uempty
HACMP Events One More Time
(called by clstmgrES recovery programs) (called by other events)
Notes:
Points to Note
The execute bit must be set on all pre-, post-, notify and recovery
scripts.
Synchronization does not copy pre- and post-event script content
from one node to another.
You need to copy all your pre- and post-event scripts to all nodes.
Your pre- and post-event scripts must handle non-zero exit codes.
All scripts must declare the shell they will run in like:
#!/bin/ksh
Notes:
Uempty
Editing an HACMP Event Script (1 of 2)
It is not recommended that you modify an HACMP event
script.
If you do, please note the following:
Notes:
Notes:
Uempty
RG_Move Event and Selective Fallover
Selective Fallover allows fallover for 1 resource group
Cluster Manager uses rg_move event for selective fallover
CSPOC can also be used to use cause an rg_move event
Selective fallover can happen for the following failures:
NIC failures
Applications
Communication Links
Volume groups
Selective Fallover can be customized by resource group
Notes:
- In cases of volume group failures, the occurrence of the AIX error label
LVM_SA_QUORCLOSE indicates that a volume group went off-line on a node in the
cluster. This causes the selective fallover of the affected resource group.
Remember that in each case when HACMP uses Selective Fallover, an rg_move event
is launched as a response to a resource failure. You can recognize that HACMP uses
Selective Fallover when you identify that an rg_move event is run in the cluster.
Uempty
Customizing Event Flow for Other Devices
HACMP provides smit screens for managing the AIX error logging
facility's error notification mechanism.
CPU
Notes:
Notes:
Menu path
smitty hacmp -> Problem Determination Tools -> HACMP Error Notification
Uempty
Configuring Automatic Error Notification
Configure Automatic Error Notification
Notes:
Notes:
Notes:
Menu path
smitty hacmp -> Problem Determination Tools -> HACMP Error Notification ->
Add a Notify Method
Emulating Errors (1 of 2)
HACMP Error Notification
Mo+--------------------------------------------------------------------------+
Error Label to Emulate
Move cursor to desired item and press Enter.
[TOP]
SSA_DISK_ERR3 SSA_DISK_DET_ER
LVM_SA_QUORCLOSE bernhardvg
LVM_SA_QUORCLOSE xwebvg
LVM_SA_QUORCLOSE rootvg
SERVICE_EVENT diagela_SE
FCP_ARRAY_ERR6 fcparray_err
DISK_ARRAY_ERR2 ha_hdisk0_0
DISK_ARRAY_ERR3 ha_hdisk0_1
DISK_ARRAY_ERR5 ha_hdisk0_2
DISK_ERR2 ha_hdisk0_3
[MORE...39]
F1=Help F2=Refresh F3=Cancel
F8=Image F10=Exit Enter=Do
F1 /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
Menu path
smitty hacmp -> Problem Determination Tools -> HACMP Error Notification ->
Emulate Error Log Entry
Uempty it would be best to cause the actual hardware error that is of concern in order to verify
that the error notification method has been associated with the correct AIX error label.
Note that the emulated error does not have the same resource name as an actual
record but otherwise passes the same arguments to the method as the actual one.
Emulating Errors (2 of 2)
Emulate Error Log Entry
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
Error Label Name LVM_SA_QUORCLOSE
Notification Object Name xwebvg
Notify Method /usr/es/sbin/cluster/>
Notes:
Uempty
What Will This Cause?
# errpt -a
---------------------------------------------------------------------------
LABEL: LVM_SA_QUORCLOSE
IDENTIFIER: CAD234BE
Date/Time: Fri Sep 19 13:58:05 MDT
Sequence Number: 469
Machine Id: 000841564C00
Node Id: bondar
Class: H
Type: UNKN
Resource Name: LVDD
Resource Class: NONE
Resource Type: NONE
Location:
Description
QUORUM LOST, VOLUME GROUP CLOSING
Probable Causes
PHYSICAL VOLUME UNAVAILABLE
Detail Data
MAJOR/MINOR DEVICE NUMBER
00C9 0000
QUORUM COUNT
0
ACTIVE COUNT
0
SENSE DATA
0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000 0000
---------------------------------------------------------------------------
Notes:
Checkpoint
1. True or False?
HACMP event scripts are binary executables and cannot be
easily modified.
2. Which of the following runs if an HACMP event script fails? (select
all that apply)
a. Pre-event scripts.
b. Post-event scripts.
c. error notification methods.
d. recovery commands.
e. notify methods.
3. How does an event script get started?
a. Manually by an administrator
b. Called by cluster manager
c. Called by a recovery program
d. Called by the topology services daemon
4. True or False?
Pre event scripts are automatically synchronized.
5. True or False?
Writing error notification methods is a normal part of configuring a
cluster.
Notes:
Uempty
Unit Summary
Notes:
References
SC23-4867-05 HACMP for AIX: HACMP Master Glossary
SC23-4864-06 HACMP for AIX: Concepts and Facilities Guide
SC23-4861-06 HACMP for AIX: Planning and Installation Guide
SC23-4862-06 HACMP for AIX: Administration Guide
SC23-5177-00 HACMP for AIX: Troubleshooting Guide
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Explain the concepts of Network File System (NFS)
Configure HACMP to support NFS
Discuss why Volume Group major numbers must be unique
when using NFS with HACMP
Outline the NFS configuration parameters for HACMP
Notes:
Objectives
In this unit, we examine how NFS can be integrated in to HACMP in order to provide a
Highly Available Network File System.
Uempty
So, What Is NFS?
The Network File System (NFS) is a client/server
application that lets a computer user view and optionally
store and update files on a remote computer as though they
were on the user's own computer
NFS Client
NFS mount
NFS Server
read-write
NFS mount
read-only
JFS mount
read-only
NFS mount
NFS Client and Server
shared_vg
Copyright IBM Corporation 2005
Notes:
NFS
NFS is a suite of protocols which allow file sharing across an IP network. An NFS server
is a provider of file service (that is, a file, a directory or a file system). An NFS client is a
recipient of a remote file service. A system can be both an NFS client and server at the
same time.
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
NFS Server
n x nfsd and mountd
n x biod
Notes:
NFS processes
The NFS server uses a process called mountd to allow remote clients to mount a local
disk or CD resource across the network. One or more nfsd processes handle I/O on the
server side of the relationship.
The NFS client uses the mount command to establish a mount to a remote storage
resource which is offered for export by the NFS server. One or more block I/O
daemons, biod, run on the client to handle I/O on the client side.
The server maintains details of data resources offered to clients in the /etc/exports file.
Clients can automatically mount network file systems using the /etc/filesystems file.
Uempty
Combining NFS with HACMP
NFS exports can be made highly available by using the HACMP
resource group to specify NFS exports and mounts
client system
# mount aservice:/fsa /a
The A resource group specifies:
client system sees /fsa as /a
aservice as a service IP label resource
/fsa as a filesystem resource
/fsa as a NFS filesystem to export
A /fsa
# mount /fsa
Bondar Hudson
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
/fsa A
# mount /fsa
Bondar Hudson
Copyright IBM Corporation 2005
Notes:
Fallover
If the node offering the NFS export should fail, a standby node takes over the shared
disk resource, locally mounts the file system, and exports the file system or directory for
remote mount.
If the client was not accessing the disk resource during the period of the fallover, then it
is not aware of the change in which node is serving the NFS export.
Note that the aservice service IP label is in the resource group which is exporting /fsa.
The HACMP NFS server support requires that resource groups which export NFS
filesystems be configured to use IPAT since the client system is not capable of dealing
with two different IP addresses for its NFS server depending on which node the NFS
server service happens to be running on.
Uempty
Configuring NFS for High Availability
Change/Show All Resources and Attributes for a Resource Group
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[MORE...10] [Entry Fields]
Volume Groups [aaavg] +
Use forced varyon of volume groups, if necessary false +
Automatically Import Volume Groups false +
Filesystems (empty is ALL for VGs specified) [/fsa] +
Filesystems Consistency Check fsck +
Filesystems Recovery Method sequential +
Filesystems mounted before IP configured true +
Filesystems/Directories to Export [/fsa] +
Filesystems/Directories to NFS Mount [] +
Network For NFS Mount [] +
[MORE...10]
F1=Help F2=Refresh F3=Cancel F4=List
F5=Reset F6=Command F7=Edit F8=Image
F9=Shell F10=Exit Enter=Do
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
the filesystems in the aaavg volume group should be treated as resources within the
resource group.
Uempty
Cross-mounting NFS Filesystems (1 of 3)
A filesystem configured in a resource group can be made
available to all the nodes in the resource group:
One node has the resource group and acts as an NFS
server
Mounts the filesystem (/fsa)
Exports the filesystem (/fsa)
All nodes act as NFS clients
Mount the NFS filesystem (aservice:/fsa) onto a local mount point (/a)
aservice
/a /fsa /a
Notes:
Cross-mounting
We can use HACMP to mount an NFS exported filesystem locally on all the nodes
within the cluster. This allows two or more nodes to have access to the same disk
resource in parallel. An example of such a configuration might be a shared repository
for the product manuals (read only) or a shared /home filesystem (read-write). One
node mounts the filesystem locally, then exports the filesystem. All nodes within the
resource group then NFS mount the filesystem.
By having all nodes in the resource group act as an NFS client including the node which
holds the resource group, it is not necessary for the takeover node to unmount the
filesystem before becoming the NFS server.
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Cross-mounting NFS Filesystems (2 of 3)
When a fallover occurs, the role of NFS server moves with the
resource group
All (surviving) nodes continue to be NFS clients
aservice
/a /fsa /a
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
aservice
export /fsa
/fsa
A
Notes:
Cross-mounting details
The key change, compared to the configuration which did not use cross-mounting, is
that this configurations resource group lists /fsa as a NFS filesystem and specifies that
it is to be mounted on /a. This causes every node in the resource group to act as an
NFS client with aservice:/fsa mounted at /a. Only the node which actually has the
resource group is acting as an NFS server for the /fsa filesystem.
Uempty
Choosing the Network for Cross-mounts
In a cluster with multiple IP networks, it may be useful to specify
which network should be used by HACMP for cross-mounts
This is usually done as a performance enhancement
The A resource group specifies:
aservice as a service IP label resource
/fsa as a filesystem resource
/fsa as a NFS filesystem to export
/fsa as a NFS filesystem to mount on /a
net_ether_01 is the network for NFS mounts
net_ether_01
net_ether_02
aGservice aservice
export /fsa
A /fsa
# mount /fsa
# mount aservice:/fsa /a # mount aservice:/fsa /a
Bondar Hudson
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Cross-mount syntax
Note the rather strange /a;/fsa syntax for specifying the directory to be
cross-mounted. This rather unusual syntax is explained in the next foil.
Note that the resource group must include a service IP label which is on the
net_ether_01 network (aservice in the previous foil).
Uempty
Syntax for Specifying Cross-mounts
Where the filesystem should be mounted over
/a;/fsa
# mount aservice:/fsa /a
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The command lvlstmajor will list the available major numbers for each
node in the cluster
For example:
# lvlstmajor
43,45...99,101...
The VG major number may be set at the time of creating the VG using SMIT
mkvg or by using the -V flag on the importvg command, for example:
C-SPOC will "suggest" a VG major number which is unique across the nodes
when it is used to create a shared volume group
Notes:
VG major numbers
Volume group major numbers must be the same for any given volume group across all
nodes in the cluster. This is a requirement for any volume group that has filesystems
which are NFS exported to clients (either within or without the cluster).
Uempty
NFS with HACMP Considerations
Some points to note...
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Checkpoint
1. True or False?
HACMP supports all NFS export configuration options.
2. Which of the following is a special consideration when using
HACMP to NFS export filesystems? (select all that apply)
a. NFS exports must be read-write.
b. Secure RPC must be used at all times.
c. A cluster may not use NFS Cross-mounts if there are client
systems accessing the NFS exported filesystems.
d. A volume group which contains filesystems which are NFS
exported must have the same major device number on all cluster
nodes in the resource group.
3. What does [/abc;/xyz] mean when specifying a directory to cross-
mount?
a. /abc is the name of the filesystem which is exported and /xyz is
where it should be mounted at
b. /abc is where the filesystem should be mounted at and /xyz is the
name of the filesystem which is exported
4. True or False?
HACMP's NFS exporting feature only supports clusters of two
nodes.
5. True or False?
IPAT is required in resource groups which export NFS
filesystems.
Copyright IBM Corporation 2005
Notes:
Uempty
Unit Summary
HACMP provides a means to make Network File System
(NFS) highly available
Configure Filesystem/Directory to Export and Filesystems
mounted before IP started in resource group
VG major number must be the same on all nodes
Clients NFS mount using service address
In case of node failure, takeover node acquires the service address, acquires
the disk resource, mounts the file system and NFS exports the file system
Clients see NFS server not responding during the fallover
NFS file systems can be cross-mounted across all nodes
Faster takeover: takeover node does not have to unmount the file system
A preferred network can be selected
Really only for read only file systems: NFS cross-mounted file systems can be
mounted read-write, but concurrent write attempts will produce inconsistent
results
Use GPFS for true concurrent access
Non-default export options can be specified in
/usr/es/sbin/cluster/etc/exports
Notes:
Copyright IBM Corp. 1998, 2005 Unit 9. Integrating NFS into HACMP 9-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
References
SC23-4867-03 HACMP for AIX, Version 5.2: Master Glossary
SC23-4864-03 HACMP for AIX, Version 5.2: Concepts and Facilities
Guide
SC23-4861-03 HACMP for AIX, Version 5.2 Planning and Installation
Guide
SC23-4862-03 HACMP for AIX, Version 5.2: Administration and
Troubleshooting Guide
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
List reasons why HACMP can fail
Identify configuration and administration errors
List the problem determination tools available in smit
Understand why the Dead Man's Switch invokes
Know when the System Resource Controller kills a node
Isolate and recover from failed event scripts
Correctly escalate a problem to IBM support
Notes:
In this unit we examine some of the reasons why HACMP might fail, and how to perform
basic problem determination in order to recover from failure.
Uempty
Why Do Good Clusters Turn Bad?
Common reasons why HACMP fails:
A poor cluster design and lack of thorough planning
Basic TCP/IP and LVM configuration problems
HACMP cluster topology and resource configuration problems
Absence of change management discipline in a running cluster
Lack of training for staff administering the cluster
Performance/capacity problems
X
X
A
Halifax Vancouver
Notes:
Root causes
Often the root cause of problems with HACMP is the absence of design and planning at
the outset, or poor design and planning. As you will have now figured out, a couple of
hours spent in planning HACMP reaps rewards later on in terms of how easy it is to
configure, administer, and diagnose problems with the cluster.
HACMP verifies all topology and resource configuration parameters and most IP
configuration parameters before synchronization takes place. This means that provided
the cluster synchronizes and starts successfully, the cluster should remain stable.
The prime reason for cluster failure once the environment is in production is
administrative mistakes and an absence of change control.
Typically HACMP clusters are very stable. During the writing of this course, a customer
complained to IBM that his HACMP cluster had failed on him because a node had failed
and his workload did not get taken over by the standby node. Upon investigation it was
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
proven that in fact an earlier (undetected) failure had resulted in the standby node
taking over the workload and a subsequent component failure resulted in a second
point of failure. How many points of failure does HACMP handle?
Uempty
Test Your Cluster before Going Live!
Careful testing of your production cluster before going live reduces the
risk of problems later.
An example test plan might include:
Test Item How to test Checked
Node Fallover
Network Adapter Swap
IP Network Failure
Storage Adapter Failure
Disk Failure
Clstrmgr Killed
Serial Network Failure
SCSI Adapter for rootvg Failure
Application Failure
Node re-integration
Partitioned Cluster
Copyright IBM Corporation 2005
Notes:
Importance of testing
Every cluster should be thoroughly tested before going live. It is important that you
develop and document a cluster test plan for your environment. Start by taking your
cluster diagram and highlighting all the things that could go wrong, then write down
what you expect the cluster to do in response to that failure. Periodically test your
cluster to ensure that fallover works correctly and correct your test plan if your
assumptions about what will happen differ from that which HACMP actually performs
(for example shutdown -F does not cause fallover). HACMP 5.2 and later provide a
test tool which will be discussed later in this unit.
Although it is recommended that Graceful with Takeover testing of the cluster services
be performed, it is especially important to conduct this testing if HACMP is to be used to
reduce Planned Downtime (for upgrades/maintenance) as this will be the cluster
function that will be used. This Graceful with Takeover testing, however, should not
replace the testing of a node failure due to crash (for example, cat hello > /dev/kmem).
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
All efforts should be made to verify application functions (user level testing) as the
cluster function tests are being performed. Verifying that the cluster functions correctly
without verifying that the application functions correctly as part of the cluster function
test is not recommended. Getting the end-user commitment is sometimes the hardest
part of this process.
Use of emulation
You can emulate some common cluster status change events. Remember that
whenever you make a change to cluster configuration, test the change before putting
the cluster back into production if at all possible.
You should always emulate a DARE change before actually doing it. If a DARE change
does not succeed during emulation, then it will definitely not succeed when you actually
do it.
Uempty
Tools to Help You Diagnose a Problem
Most problems related to IP, LVM and cluster configuration errors
Tools:
Automatic Cluster Configuration Monitoring
Automatic Error Correction during verify
HACMP Cluster Test Tool
Emulation Tools
HACMP Troubleshooting manual
Log files: hacmp.out, cluster.log, clverify.log, clustrmgr.debug
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
HACMP Verification
View Current State
HACMP Log Viewing and Management
Recover From HACMP Script Failure
Restore HACMP Configuration Database from Active Configuration
Release Locks Set By Dynamic Reconfiguration
Clear SSA Disk Fence Registers
HACMP Cluster Test Tool
HACMP Trace Facility
HACMP Event Emulation
HACMP Error Notification
Notes:
Uempty
HACMP Verification
[Entry Fields]
* Automatic cluster configuration verification Enabled +
Node name Default +
* HOUR (00 - 23) [00] +#
Notes:
How it works
The clverify utility runs on one user-selectable HACMP cluster node once every 24
hours. By default, the first node in alphabetical order runs the verification at midnight.
When automatic cluster configuration, monitoring detects errors in cluster configuration,
clverify triggers a general_notification event. The output of this event is logged in
hacmp.out throughout the cluster on each node that is running cluster services.
clverify maintains the log file /var/hacmp/clverify/clverify.log.
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Automatic Correction
HACMP Verification and Synchronization
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Verify, Synchronize or Both [Both] +
* Automatically correct errors found during [No] +
verification?
* Force synchronization if verification fails? [No] +
* Verify changes only? [No] +
* Logging [Standard] +
Notes:
Uempty - Disks are available, but the volume group has not been imported to a node.
- Required HACMP snmpd entries are missing on a node.
Note that the autocorrection selection will not appear if cluster services are running.
Instead the top line of the menu will look like:
HACMP Verification and Synchronization (Active Cluster Nodes Exist)
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Currently, if a user attempts to start cluster services on a node on which the HACMP
topology has not yet been synchronized, the following error will be displayed
immediately upon entering SMIT:
smitty clstart
Local node not properly configured for HACMP for AIX.
This message appears when the HACMP cluster ODM stanza does not contain a node
name in the nodename field. Rather than continue to display this message, the existing
clstart SMIT interface will be updated to include a message that indicates verification /
synchronization will occur prior to starting cluster services. This message will appear in
the SMIT interface in place of the existing message.
The assumption is there is a valid cluster configuration on the local node the user is
attempting to start, but the user has not synchronized which leaves the HACMP cluster
node handle field blank and results in the error message, above. If cluster services are
not running on any node in the cluster (known to the local node) then the local cluster
configuration will be synchronized to all nodes attempting to start cluster services after
successfully verifying the local DCD configuration. If, however cluster services are
running on a node in the cluster then the local DCD will be compared against an ACD of
a running cluster node where the local node participates in the ACD's configuration. If
the DCD and ACD match, then verification is run. If the DCD and ACD do not match,
then a snapshot is made of the DCD, and the active node's ACD will be copied to the
DCD on the local node and verification will be run prior to starting cluster services.
This feature can be disabled such that verification and synchronization does not occur
during cluster startup. The smit path to disable is:
smitty hacmp -> Extended Configuration -> Extended cluster services
Uempty
HACMP Cluster Test Tool
HACMP Cluster Test Tool
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Event Emulation Tools
HACMP provides tools to emulate common cluster events.
Only certain events are emulated
Multiple events cannot be emulated
Each event runs in isolation; results do not impact upon the next
emulated event
The results are logged in /tmp/emuhacmp.out
If an event fails when emulated, it's not going to work when it
happens for real
Failed/Joined Network
Swap Adapter Failed/Joined Standby
Failed/Joined Node
A
Halifax Vancouver
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Menu path
smitty hacmp -> Problem Determination Tools -> HACMP Event Emulation
Uempty
Emulating a Node Down Event
Emulate Node Down Event
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Node Name [hudson] +
* Node Down Mode graceful +
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
[Entry Fields]
* Network Name [net_ether_01] +
Node Name [] +
Notes:
Uempty
Checking Cluster Subsystems (1 of 2)
# /usr/es/sbin/cluster/utilities/clcheck_server \
clstrmgrES ; echo $?
Mandatory clstrmgrES
Cluster
Components clinfoES
Optional
Notes:
clstart subsystems
Listed here are the processes that are listed in the startup smit menu for HACMP. Its
interesting to note that these cluster processes are not displayed by the command
when they are inactive. This was a display option (or probably better a non-display
option) that HACMP chose to use when the subsystems were defined during the install
process. This option can be changed (one subsystem at a time) using the chssys -s
subystem_name -a -D command.
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
An alternative command which works in HACMP 5.3 but is not guaranteed for the future
is easier. It is lssrc -ls clstrmgrES | grep state. Another command that will give you
state information in HACMP 5.3 is the command /usr/es/sbin/cluster/utilities/cldump.
Finally, you can use the smit path: Problem Determination Tools -> View Current
State
Uempty
Checking Cluster Subsystems (2 of 2)
Check rsct, clcomd, ctrmc subsystems
#
# lssrc a | grep svc
topsvcs topsvcs 258248 active
grpsvcs grpsvcs 434360 active
emsvcs emsvcs 335994 active
emaixos emsvcs 307322 active
# lssrc -s clcomdES
Subsystem Group PID Status
clcomdES clcomdES 13420 active
# lssrc -s ctrmc
Subsystem Group PID Status
ctrmc rsct 2954 active
#
Notes:
Supporting subsystems
Listed here are the additional processes we would expect to find running on an HACMP
5.2 and later cluster node.
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Why?
Clstrmgr starved of CPU
Excessive I/O traffic
Excessive TCP/IP traffic over an interface
Was it DMS?
Copy the system dump to a file
kdb on the dump file
stat subcommand
look for 'HACMP dms timeout halting...'
Notes:
Uempty
Avoiding Deadman Switch Timeouts
Steps to avoid DMS timeout problems
1. Isolate the cause of excessive I/O or TCP/IP traffic and fix it, and if that
does not work...
2. Turn on I/O pacing, and if that does not work...
3. Increase the frequency of the syncd, and if that does not work...
4. Reduce the failure detection rate for the slowest network, and if that does
not work...
5. Buy a bigger machine
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
The command /usr/sbin/rsct/bin/hatsdmsinfo can be used to see how often the DMS
timer is being reset
Although we don't recommend changing the DMS time-out value, we are sometimes
asked about how to increase the time-out period on the deadman switch in order to
make it less likely that the DMS will pop and crash the node. There is no strict time-out
setting, it is monitored by RSCT and is calculated as twice the value of the longest
failure detection rate of all configured HA network in the cluster. If, for example, you
have 2 networks, an Ethernet, and a disk heartbeat network, the Ethernet has the
longer failure detection rate, 10 seconds versus 8 for the diskhb network, so the DMS
time-out is set to 2*10, or 20 seconds. If the failure detection rate is being modified to
extend the DMS time-out, it is best to ensure that all networks have the same failure
detection period. To set the DMS time-out value to 30 seconds, while making the failure
detection the same for both networks, the custom NIM settings would be: Ethernet:
Failure Cycle 16
Interval between Heartbeats (seconds) 1
Uempty
Setting Performance Tuning Parameters
Extended Performance Tuning Parameters Configuration
Move cursor to desired item and press Enter.
Change/Show I/O pacing
Change/Show syncd frequency
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Changing the Frequency of syncd
The HACMP documentation recommends a value of 10.
Change/Show syncd frequency
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
syncd frequency (in seconds) [10] #
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Partitioned Clusters and Node Isolation
When
Heartbeats are received from a node that was marked as
failed
HACMP ODM configuration is not the same on a joining
node as nodes already active in the cluster
Two clusters with the same ID appear in the same logical
network
The rogue recovering or joining node is halted
What happens
Group Services and clstrmgr exit on some node(s)
Notes:
Node isolation
When you have a partitioned cluster, the node(s) on each side of the partition detect this
and run a node_down for the node(s) on the opposite side of the partition. If, while
running this or after communication is restored, the two sides of the partition do not
agree on which nodes are still members of the cluster, a decision is made as to which
partition should remain up, and the other partition is shutdown by a Group Services
(GS) merge from nodes in the other partition or by a node sending a GS merge to itself.
In clusters consisting of more than two nodes, the decision is based on which partition
has the most nodes left in it, and that partition stays up. With an equal number of nodes
in each partition (as is always the case in a two-node cluster) the node(s) that remain(s)
up is determined by the node number (lowest node number in cluster remains) which is
also generally the first in alphabetical order.
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Avoiding Partitioned Clusters
Have a non IP (serial) network
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
It means that an event script has failed, hung or is taking too long.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
smit menu
smit hacmp -> Extended Configuration -> Extended Event Configuration ->
Change/Show Time Until Warning
Uempty
Recovering From an Event Script Failure
Notes:
The procedure
The procedure is outlined in the visual above. Using the /usr/es/adm/cluster.log file
with the command grep EVENT /usr/es/adm/cluster.log | more makes it easier to
find when the config too long event 1st occurred. Be sure to find the earliest AIX error
message -- not just the first AIX error message. You must manually complete what the
event would have done before doing recover from script failure which is described on
the next visual. You can also use the cluster.log in combination with hacmp.out.
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
A Troubleshooting Methodology
Save the log files from all available nodes as soon as possible
Attempt to duplicate the problem
Approach the problem methodically
Distinguish between what you know and what you assume
Keep an open mind
Isolate the problem
Go from the simple to the complex
Make one change at a time
Stick to a few simple troubleshooting tools
Do not neglect the obvious
Watch for what the cluster is not doing
Keep a record of the tests you have completed
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Item Checked
EXACT error messages that appear in HACMP logs such as
hacmp.out or on the console
Your cluster diagram / Planning Worksheets (updated)
A snapshot of your current cluster configuration (not a photo)
Details of any customization performed to HACMP events
Details of current AIX, HACMP and application software levels
Details of any PTFs applied to HACMP or AIX the cluster
The adapter microcode levels (especially for SSA adapters)
Cluster planning worksheets, with all components clearly labeled
A network topology diagram for the network as far as the users
Copies of all HACMP log files (snap e command)
Notes:
Uempty
Checkpoint
1. What is the most common cause of cluster failure?
a. Bugs in AIX or HACMP
b. Cluster administrator error
c. Marauding space aliens from another galaxy
d. Cosmic rays
e. Poor/inadequate cluster design
2. True or False?
Event emulation can emulate all cluster events.
3. If the cluster manager process should die, what will happen to the
cluster node?
a. It continues running but without HACMP to monitor and protect it.
b. It continues running AIX but any resource groups will fallover.
c. Nobody knows because this has never happened before.
d. The System Resource Controller sends an e-mail to root and issue a "halt -q".
e. The System Resource Controller sends an e-mail to root and issue a
"shutdown -F".
4. True or False?
A non-IP network is strongly recommended. Failure to include a non-IP
network can cause the cluster to fail or malfunction in rather ugly ways.
5. (bonus question) My favorite graphic in the lower right hand corner
of a foil was: ____________________________________
Notes:
Copyright IBM Corp. 1998, 2005 Unit 10. Problem Determination and Recovery 10-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Summary
Notes:
References
SC23-4861-06 HACMP for AIX, Version 5.3 Planning and Installation
Guide
/usr/es/sbin/cluster/release_notes
www.ibm.com/servers/eserver/pseries/haHACMP 5.3
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
List the methods to update a cluster to HACMP 5.3
Describe the migration steps for each migration method
List the new features of HACMP 5.3
Notes:
Uempty
Okay, So You Want to Upgrade
Various versions of HACMP work with different release levels of AIX.
As of September 2005 only HACMP 5.x releases are supported.
HACMP 5.3 is not supported on AIX 5L V5.1.
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Cold (Snapshot)
HACMP 4.5 HACMP 5.3
Notes:
Terminology
HACMP defines 4 migration methods:
i. Rolling: A type of upgrade from one HACMP version to another during which
cluster services are not stopped on all nodes in the cluster. Rolling migration lets
you update a cluster that has an ES (Enhanced Scalability) option to a cluster
with a higher version of the ES option. HACMP is not uninstalled and data is
saved.
ii. Snapshot: A type of upgrade from one HACMP version to another during which
you take a snapshot of the current cluster configuration, stop cluster services on
all nodes, install the next version of HACMP and then convert the snapshot by
running the clconvert_snapshot utility. HACMP is uninstalled and no data is
saved.
Uempty iii. Node-by-Node: A type of a rolling migration that enables you to upgrade the
cluster from one version to another while also moving the cluster from HACMP
(HAS) to HACMP/ES (HACMP 5.x). HACMP is not uninstalled and data is saved.
iv. Offline: Here the cluster services are stopped. HACMP is not uninstalled and
data is saved.
Note that migration of a node while the applications are running is not an acceptable
method for upgrading HACMP.
Recommended reading
Chapter 10 HACMP Planning and Installation Guide
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_b
res_grp_a
TMSSA
RS232
Vancouver
Halifax
Notes:
We will be upgrading a running cluster with an application. The goal is to keep the
application running while we upgrade HACMP/ES 4.5 on AIX 5L V5.1 to HACMP 5.3 on
AIX 5L V5.3.
Uempty
Upgrade Steps
The steps are generic
Step Description Resources Resources available
available on Halifax on Vancouver
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Risks?
To assume the world is perfect and software always functions as
designed would put most of us out of a job
If changing the level of AIX, will the applications still function?
Is the combination of AIX and HACMP supported?
Upgraded systems may not use "new features" until the rest of the cluster is
aware of the "features".
Nodes may run at different versions of HACMP, but ONLY during migration
and only ES to ES or classic to classic.
Keep the time that mismatched versions are running to a minimum.
Adapter and Storage microcode may require upgrading; don't forget to check.
Make a plan. Make sure you can go back.
Old code at latest level
Notes:
Uempty
Step 1/Roll: Backup and Snapshots
Make an extra backup and mksysb before starting
Test that backups are useable
Back up the shared data separate from the non-shared storage
Verify the cluster and take a snapshot; move the files to a safe place
Ethernet
res_grp_b
res_grp_a
TMSSA
RS232
Vancouver
Halifax
Notes:
Get ready
The cluster should be working! Save the configuration and make a mksysb. The
HACMP software on the cluster nodes should be at the current level. If patches are
applied, make sure you reboot and repeat saving the configuration and mksysb. The
saved snapshot should be placed in a directory not used by HACMP.
The backup steps can't be stressed enough. This is the best insurance policy should
things go drastically wrong during the migration/upgrade process.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
res_grp_b
res_grp_a
TMSSA
RS232
Vancouver
Halifax
Notes:
During the installation, HACMP creates the hacmp group on all nodes. By
default, the hacmp group has permission to read the HACMP ODMs, but does
not have any other special authority. For security reasons, do not to expand the
authority of the hacmp group.
If you use programs that access the HACMP ODMs directly, you may need to
rewrite them if they are intended to be run by non-root users:
All access to the ODM data by non-root users should be handled via the
provided HACMP utilities.
In addition, if you are using the PSSP File Collections facility to maintain the
consistency of /etc/group, the new hacmp group that is created at installation
time on the individual cluster nodes may be lost when the next file
synchronization occurs.
To prevent overwriting your hacmp group, before installing HACMP 5.3,
either:
- Turn off the PSSP File Collections synchronization of the /etc/group file
OR
- Include the hacmp group in the master /etc/group file and propagate this
change to all cluster nodes.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Step 3/Roll: Migrate Resources
The resource groups are to be moved to the other nodes while the
originating node is worked on.
This will cause a short outage to move the resource group.
Shutdown Cluster Services with Takeover.
Verify the resource group is taken over successfully.
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Vancouver
Halifax
AIX 5.1
AIX 5.1 HACMP/ES 4.5
HACMP/ES 4.5
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Vancouver
Halifax
AIX 5.1
AIX 5.1 HACMP/ES 4.5
HACMP/ES 4.5
Notes:
Stopping HACMP
If an rg_move command (C-SPOC) was used to move the resource groups (as
opposed to a shutdown with fallover), it is now time to stop HACMP on the node to be
upgraded.
Uempty
Step 5/Roll: Upgrade AIX
If AIX is to be upgraded, do it now.
To keep HACMP ODM entries in tact:
Use "Migration Install" if installing a new version/release of AIX.
Use "update all" if installing a new modification level of AIX.
Apply the necessary AIX ptf's.
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Halifax Vancouver
AIX 5.1
AIX 5.3 ML 2 HACMP/ES 4.5
HACMP/ES 4.5
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Halifax Vancouver
AIX 5.1
AIX 5.3 ML 2 HACMP/ES 4.5
HACMP 5.3
Notes:
Uempty
Step 7/Roll: Convert ODM Classes
If a ODM class upgrade failure occurred in the previous step, it
may be necessary to manually update the HACMP ODM classes.
HACMP 5.3 will automatically attempt to run clconvert. Check the output
after the install for positive or negative messages.
The clconvert can be found in /usr/es/sbin/cluster/conversion if required to
run manually.
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Halifax Vancouver
AIX 5.1
AIX 5.3 ML2 HACMP/ES 4.5
HACMP 5.3
Notes:
This step is only necessary if the conversion routine did not complete successfully.
The /usr/es/sbin/cluster/conversion/clconvert command is documented in the HACMP
Administration Guide Appendix C.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Vancouver
Halifax
AIX 5.1
AIX 5.3 ML2 HACMP/ES 4.5
HACMP 5.3
Notes:
Uempty
Step 9/Roll: Restart HACMP
Restart HACMP with smitty clstart.
Verify clstrmgr starts
Depending on the resource groups configuration and options, a takeover
may occur. Be prepared for an outage on the application
Use clstat or cldisp to verify the cluster becomes stable
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Vancouver
Halifax
AIX 5L V5.1
AIX 5L V5.3 ML2 HACMP/ES 4.5
HACMP 5.3
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Vancouver
Halifax
AIX 5L V5.1
AIX 5L V5.3 ML2 HACMP/ES 4.5
HACMP 5.3
Notes:
Uempty
What About Failures?
In this type of mixed environment, HACMP will function but:
Applications may not appreciate differences in HACMP or AIX levels
Care must be taken not to deploy options or enhancements that the older
version of HACMP or AIX cannot deal with
There may be compatibility issues in the event of a failover so don't rest too
long
HACMP will not be able to synchronize
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Halifax Vancouver
AIX 5.1
AIX 5.3 ML2 HACMP/ES 4.5
HACMP 5.3
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Steps 11 to 19/Roll
Repeat the steps 3 through 10 only on the other node.
Move the resources to the Halifax node
Stop HACMP
Upgrade AIX and HACMP; convert ODM if necessary
Reboot, verify and restart HACMP
The process is exactly the same, it is just the other node.
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Halifax Vancouver
Notes:
Uempty
Step 20/Roll: Backup and Testing
Okay, all the upgrades are done. The resource groups are back
where they belong.
Verify the cluster will synchronize both topology and resources
Verify the cluster is stable
Create a new mksysb for both nodes
Create new cluster snapshot
Fully test the cluster
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Halifax Vancouver
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
cold convert
HACMP 5.X, HACMP 5.3
HACMP/ES 4.5
Notes:
Uempty
Snapshot/Cold Convert
The cold convert method has its advantages.
Conversion can be accomplished from HACMP version 4.1 upwards
The entire cluster is converted at once
A snapshot may be used or you can just start all over
The configuration may be reentered in a different configuration
The disadvantages:
The entire cluster must be down and all the nodes will require a reboot
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Uempty
Step 1/Snap: Backup and Snapshots
The backups and snapshots are very important. All of the previous
reasons apply and a few extra considerations this time.
Copy the snapshot out of the default directory
/usr/es/sbin/cluster/snapshots/<name>.odm and .info to another directory
Move all customized events and configuration information out of
/usr/es/sbin/cluster and /usr/lpp/cluster
Ethernet
res_grp_b
res_grp_a
TMSSA
RS232
Vancouver
Halifax
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Halifax Vancouver
Notes:
Uempty
Step 3/Snap: Stop HACMP
Stop HACMP
No point in doing a failover, both cluster nodes have to be down
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Vancouver
Halifax
AIX 5L V5.1
AIX 5L V5.1 HACMP/ES 4.5
HACMP/ES 4.5
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
TMSSA
RS232
Vancouver
Halifax
AIX 5L V5.1
AIX 5L V5.1
Notes:
Uempty
Step 5/Snap: Install HACMP 5.3
Migrate/update AIX 5L V5.1 to AIX 5L V 5.3
Install the appropriate AIX prerequisite file sets
Install the HACMP 5.3 filesets
Use smitty install_all
Only install the required file sets
Reboot the node. It is a requirement of any HACMP installation
Ethernet
TMSSA
RS232
Vancouver
Halifax
Notes:
Installation notes
- As in any installation process performed using SMIT, the preview option can be
selected. If any required filesets are missing, this will list them. HACMP 5.3 has a
couple of additional prerequisites that HACMP 4.5 (classic) did not have.
- Any prerequisites should have been discovered in the planning stages.
- All of the nodes in the cluster can be installed at the same time.
- Reboot each node in the cluster with a shutdown Fr command. When you reboot
cluster nodes, the first node up runs a network_down event for each of its non-IP
networks, even if the network is healthy:
Error: EVENT START: network_down -1 SERIAL
This event appears in the /tmp/hacmp.out file whether or not the network is
functional. Ignore the message and let cluster services continue to function. You
should see this error message corrected in a healthy cluster as functional network
communication is eventually established between other nodes in the cluster.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
TMSSA
RS232
Halifax Vancouver
Notes:
Snapshot considerations
Copy the saved snapshot into the /usr/es/sbin/cluster/snapshots directory.
Applying the snapshot on one system will cause synchronization of all the nodes in the
cluster.
For additional info on the snapshot utility consult the HACMP Administration guide.
You may need to force apply the snapshot. The following messages can be ignored:
WARNING: The NFS mount/Filesystem specified for resource group rg1 is using
incorrect syntax for specifying an NFS cross mount: /mnt/fs1.
ERROR: Disk Heartbeat Networks have been defined, but no Disk Heartbeat devices.
You must configure one device for each node in order for a Disk Heartbeat network to
function.
Uempty
Step 7/Snap: Reinstall Event Customization
If there has been primary or secondary event customization, it has
been removed and may require reinstallation
This assumes the event customization is documented and stored in a location
that is still accessible
Manual intervention will be required to check the compatibility of the previous
event customization and the new event scripts
Ethernet
TMSSA
RS232
Halifax Vancouver
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Resources
Ethernet
res_grp_b
res_grp_a TMSSA
RS232
Halifax Vancouver
Notes:
Uempty
Step 9/Snap: Verify Cluster Operation
Use the original planning information to verify the cluster operation
Perform fall over test per the test plan
Verify customization is working as planned
Check for errors in the various logs
Verify the cluster becomes stable
Take a new snapshot and mksysb
Ethernet
res_grp_b
res_grp_a
TMSSA
RS232
Vancouver
Halifax
Notes:
Test
A complete test of cluster functionality should be performed. Don't assume that it
worked fine before the upgrade, so it will work fine now. Build enough time into the plan
to fully test the cluster.
Perform the steps in the test plan. You may want to also use the Test facility of HACMP.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-35
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Node-by-node migration
Now that we've seen the snapshot method we'll take a look at the steps required to
perform a node-by-node conversion.
Node by node migration first appeared in HACMP 4.3.1. Since it is the goal to finish
on HACMP/ES 4.5 then it is suggested to upgrade HACMP to 4.5 first, then perform
the node by node migration.
Node-by-Node migration is supported in HACMP 5.3 only from HACMP 4.5
(classic).
Uempty
Node by Node Migration Steps
The steps are for a node by node migration from HACMP 4.5 to
HACMP 5.3
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-37
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
- The HACMP 5.3 recovery driver sends a message to the HACMP clstrmgr telling it
to run the waiting and waiting_complete events
- HACMP 5.3 uses the RSCT Group Services to verify cluster stability and
membership
- The firstboot file on each cluster node is moved to an active directory (/etc)
- The migration flag (.mig file) created during installation is transferred from the
HACMP 5.3 directory to the HACMP 4.5 directory on all nodes
When the firstboot file is moved to the active directory and the .mig file transfer is
complete on all nodes, transfer of control to HACMP 5.3 continues with the HACMP 5.3
migrate event.
- HACMP 5.3 recovery driver issues the migrate event
- HACMP 5.3 stops the HACMP daemons using the forced option
- HACMP 5.3 clinfoES daemons are all activated, reusing the ports previously used
by the HACMP versions of those daemons
- HACMP 5.3 recovery driver runs the migrate_complete event
- HACMP is deinstalled; configuration files common to both products are left
untouched
- Base directories are relinked
- /etc/firstboot files are removed
- The migration flag (.mig file) in the HACMP /usr/sbin/cluster directory is removed
- Migration is now complete
You should verify and test the clusters proper fallover and recovery functionality.
Uempty
Step 1/Node. Backup and Snapshots
Node by Node migration has a "step of no return", backups are
critical in this environment
Copy the snapshot out of the default directory
/usr/sbin/cluster/snapshots/<name>.odm and .info to another directory
Ethernet
res_grp_b
res_grp_a
TMSSA
RS232
Vancouver
Halifax
Notes:
As we saw during the cold upgrade, it is important to make a copy of the snapshot files in a
location which won't be effected by the upgrade. If this isn't done and the snapshot is
converted to HACMP/ES format, it can't be unconverted.
The deinstallation of HACMP and implementation of HACMP/ES will happen under the
control of the migrate event script. It is critical that a good mksysb is created before the
process is started.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-39
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Halifax Vancouver
Notes:
Points to consider
Having two copies of the mksysb image is probably not a bad idea before you start this
process.
Verifying prerequisites can be accomplished using the preview option from the SMIT
install software menu.
The cluster will require a complete test for functionality once the migrations have been
completed. Testing is a must; don't assume everything will work fine.
The clsmuxpd is not used in HACMP 5.3. During the node-by-node migration, it will not
be installed on the node. Before upgrading, make sure your applications previously
monitored by clsmuxpd now use application monitoring.
The cl_registerwithclsmuxpd() API routine is removed in HACMP 5.3; application
monitoring effectively supersedes this function. Ensure your applications previously
monitored by clsmuxpd now use application monitoring.
Uempty
Step 3/Node. Stop HACMP
Stop HACMP 4.5 on the node to be upgraded
Stop cluster services with takeover to have the resource groups migrate to
the next node in the resource group
Check the resources have successfully transferred to another node
Verify there are no resource groups on the node to be upgraded
Ethernet
res_grp_a
TMSSA res_grp_b
RS232
Vancouver
Halifax
Notes:
Use clstat on the surviving node to verify the cluster becomes stable after the node is
stopped. If the cluster does not become stable, stop the process here and troubleshoot the
problem.
Verify all the cluster resources are active on the second node.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-41
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Ethernet
res_grp_a
TMSSA res_grp_b
RS232
Vancouver
Halifax
Notes:
Review the smit.out log file after completing the install process. Make sure each fileset
installed completed successfully.
The reboot is not optional.
Note: The clsmuxpd is not used in HACMP 5.3. During the node-by-node migration, it
will not be installed on the node.
Note: The cl_registerwithclsmuxpd() API routine is removed; application monitoring
effectively supersedes this function. Ensure your applications previously monitored by
clsmuxpd now use application monitoring.
Uempty
Step 5/Node. Restart HACMP
Restart HACMP services on the hybrid node
Both HACMP 4.5 and HACMP 5.3 clstrmgr will start
HACMP 4.5 will be in control as the node rejoins the cluster
The node will reacquire resources according to their configuration parameters
Ensure the cluster becomes stable
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Vancouver
Halifax
Notes:
Take a look at all the processes that have started as the hybrid node joins the cluster.
Commands like lssrc -g cluster will show the node is in hybrid mode.
Use the clstat utility to verify cluster stability.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-43
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
res_grp_b
TMSSA
res_grp_a
RS232
Vancouver
Halifax
Notes:
Now repeat the steps previously performed to the upgraded node.
Uempty
Step 7/Node. Install HACMP 5.3
Install the appropriate HACMP 5.3 file sets
Only install the required file sets
Verify the filesets installed correctly
Reboot the node
Ethernet
res_grp_a
TMSSA
res_grp_b
RS232
Vancouver
Halifax
Notes:
Perform the same step as on the other node.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-45
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
res_grp_a res_grp_b
TMSSA
RS232
Vancouver
Halifax
Notes:
Review the log files carefully.
The migration will start automatically when the last node joins the cluster in the hybrid
mode.
This is interesting to observe by tail -f /tmp/hacmp.out. The TCP/IP processes may
appear to hang while the migration is in progress.
Make sure the conversion process completed successfully. Also verify that the HACMP
software was uninstalled correctly.
Uempty
Step 9/Node. Backup and Snapshots
The cluster will be running HACMP 5.3 only
The migration is complete
It is a good time to make a new snapshot and mksysb
Verify correct operation of the cluster
Check for errors during conversion in /tmp/hacmp.out and /tmp/cm.log
Ethernet
res_grp_a res_grp_b
TMSSA
RS232
Vancouver
Halifax
Notes:
Make new backups. This is important, the complete configuration had been changed from
HACMP to HACMP/ES.
Fully test the cluster functionality. Make sure everything stills performs as expected.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-47
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Packaging
HACMP comes as a base CD with additional fee based feature CDs -- one for Smart
Assists (DB2, WebSphere, and Oracle) and the other for HACMP/XD. The clsmuxpd
daemon has been removed. The function is now in the clstrmgrES subsystem. The
HACMP for AIX Administration and Problem Determination Manual has been broken up
and problem determination is now in a separate manual. Dont forget the
RELEASE_NOTES file (see copy in Appendix to this course).
Ease of use
XML files can be created to keep configuration data. HACMP can be configured from
them via the online planning work sheets or the utilities command cl_opsconfig.
Snapshots can also be converted to the xml formatted file. HACMP will allow
administrators to use Veritas Volume groups in an HACMP cluster and opens the
interface to other OEM vendors. Paging notifications to cell phones is now supported.
Uempty Finally during synchronization HACMP will propagate a clhosts file for HACMP client
nodes. The file created is /usr/es/sbin/cluster/etc/clhosts.client
Configuration changes
You can now have choices about how to distribute alias addresses among the
interfaces as well as how to distribute dependent resource groups such as the ability to
ensure that the parent and child resource groups do not run on the same node. If you
configured a resource group with a distribution policy of network, it is no longer
supported in HACMP 5.3. HACMP 5.3 now supports the use of virtual I/O for network
adapters (must use netmon file) as well as DLPAR for power 5.
Verification changes
HACMP 5.3 will ensure all nodes are configured/synchronized the same way as the first
node up (see additional rules in the Planning and Installation Guide). It is no longer
necessary to manually invoke the Automated Error Notification. The cldiag command is
now deprecated and moved to the samples directory. Additional warnings, errors and
auto corrections have been added:
- Report Error if HACMP network type does not match CuAt interface type
- Auto correct ERROR: Network option: tcp_pmtu_discover has different settings
between nodes: nodeA and nodeB. Please make sure that the command no o
tcp_pmtu_discover provides the same output on all nodes (also for
udp_pmtu_discover and ipignoreredirects)
- Auto correct WARNING: Network option: routerevalidate is set to 0 on node
nodeA. Please be aware that this setting will be changed to 1 during HACMP
startup.
- Auto correct WARNING: Network option: nonlcsrcroute is set to 0 on node
nodeA. Please be aware that this setting will be changed to 1 during HACMP
startup (also for ipsrcroutesend, ipsrcrouterecv, and Ipsrcrouteforward)
- Report ERROR: The MTU sizes do not match for communication interface:
ip_label_mtu1500 and ip_label_mtu2000. The NIC en1 on node nodeA has an MTU
size of 1500, and the NIC on node nodeB has an MTU size of 2000. To correct this
error, make sure that the MTU size is consistent across all NICs on the same
HACMP network.
- Report WARNING: The RSCT level is different on nodes: nodeA and nodeB. Both
nodes have AIX level 5.3.0.1 installed, and RSCT software is at 2.3.0.0 on node
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-49
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Uempty
Checkpoint
1. True or False?
AIX 5.1 is compatible with HACMP 5.3.
2. To be able to do a node-by-node migration from HACMP 4.4.1
to HACMP 5.3 when AIX is at level 4.3.3 you should:
a. You can do node-by-node migration to HACMP 5.3
b. First upgrade to HACMP 4.5 and then upgrade to HACMP 5.3.
c. Nobody knows because this has never happened before.
d. None of the above
3. True or False?
When upgrading levels of HACMP, it is recommended that you upgrade one
node and then run the cluster for a week or two to see if any problems occur
before upgrading the other nodes.
4. List 3 features of HACMP 5.3
5. True or False
Always make a new mksysb of the system and create a
snapshot before upgrading to a new level of HACMP
6. If the nodes of an HACMP 4.5 cluster running AIX 5.1 need to
be upgraded to HACMP 5.3 then which should be done first?
a. HACMP 4.5 to HACMP 5.1
b. AIX 5.1 to AIX 5.2
c. HACMP 4.5 to HACMP 5.3
Notes:
Write down your answers here:
1.
2.
3.
4.
5.
6.
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-51
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Exercise
Exercise:
Migration
Notes:
Uempty
Unit Summary
Notes:
Copyright IBM Corp. 1998, 2005 Unit 11. HACMP Migration 11-53
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Checkpoint Solutions
1. True or False?
Resource Groups may be moved from node to node
2. True or False?
HACMP XD is a complete solution for building
geographically distributed clusters.
3. Which of the following capabilities does HACMP not
provide (select all that apply)?:
a. Time synchronization.
b. Automatic recovery from node and network adapter failure.
c. System Administration tasks unique to each node. Backup and
restoration.
d. Fallover of just a single resource group.
4. True or False?
Resource Groups may be moved from node to node.
5. True or False?
All nodes in a resource group must have equivalent
performance characteristics.
Checkpoint Solutions
1. True or False?
Lazy update keeps VGDA constructs in sync between cluster nodes.
(reserve/release-based shared storage protection)
2. Which of the following commands will bring a volume group
online?
a. getvtg <vgname>
b. mountvg <vgname>
c. attachvg <vgname>
d. varyonvg <vgname>
3. True or False?
Quorum should always be disabled on shared volume groups.
4. True or False?
filesystem and logical volume attributes cannot be changed while the
cluster is operational.
5. True or False?
An enhanced concurrent volume group is required for the heartbeat over
disk feature.
Checkpoint Solutions
1. True or False?
Clients are required to exit and restart their application
after a fallover.
2. True or False?
All client systems are potentially directly affected by the
ARP cache issue.
3. True or False?
clinfo must not be run both on the cluster nodes and
on the client systems.
4. If clinfo is run by cluster nodes to address ARP cache
issues, you must add the list of clients to ping to either the
/etc/cluster/ping_client_list or the
/usr/es/sbin/cluster/etc/clinfo.rc file
Checkpoint Solutions
1. True or False
Applications are defined to HACMP in a configuration file
that lists what binary to use.
2. What policies would be the best to use for a 2 node mutual
takeover cluster using IPAT to minimize both applications
running on the same node?
a. home, next, never
b. first, next, higher
c. distribution, next, never
d. all, error, never
e. home, next, higher
3. Which type of data should not be placed in private data
storage?
a. Log data
b. License file
c. Configuration files
d. Application binaries
4. Which policy is not a Run-time policy?
a. Settling
b. Delayed Fallback Timer
c. Dynamic Node Priority
Copyright IBM Corporation 2005
Checkpoint Solutions
1. What component detects an adapter failure?
a. Cluster manager
b. RSCT
c. clcomd
d. clinfo
2. What component provides SNMP information?
a. Cluster manager
b. RSCT
c. clsmuxpd
d. clinfo
3. What component is required for clstat to work?
a. Cluster manager
b. RSCT
c. clcomd
d. clinfo
Checkpoint Solutions
1. True or False?
It is possible to configure a recommended simple two-node cluster
environment using just the standard configuration path.
2. In which of the top level HACMP menu choices is the menu for starting and
stopping cluster nodes?
a. Initialization and Standard Configuration
b. Extended Configuration
c. System Management (C-SPOC)
d. Problem Determination Tools
3. In which of the top level HACMP menu choices is the menu for defining a non-
IP heartbeat network?
a. Initialization and Standard Configuration
b. Extended Configuration
c. System Management (C-SPOC)
d. Problem Determination Tools
4. True or False?
It is possible to configure HACMP faster by having someone help you on
the other node.
5. True or False?
You must specify exactly which filesystems you want mounted when you
put resources into a resource group.
Checkpoint Solutions
1. True or False?
A star configuration is a good choice for your non-IP networks.
2. True or False?
Using DARE, you can change from IPAT via aliasing to IPAT via
replacement without stopping the cluster.
3. The clpasswd utility allows users to change their passwords across
all nodes
4. True or False?
RSCT will automatically update /etc/filesystems when using enhanced
concurrent mode volume groups
5. True or False?
A resource groups priority override location can be cancelled by
selecting a destination node of Restore_Node_Priority_Order.
6. The basic steps to configure WebSMIT are:
a. Install a web server and edit the httpd.conf file to configure it to
server WebSMIT pages
b. Link the WebSMIT cgi-bin and htdocs directories to the web
servers directory
c. Edit the wsm_smit.conf file to configure WebSMIT security
d. Set 4511 permissions on the wsm_cmd_exec file.
AP Unit 8 - Events
Unit 8 - Events
Checkpoint Solutions
1. True or False?
HACMP event scripts are binary executables and cannot be
easily modified.
2. Which of the following runs if an HACMP event script fails? (select
all that apply)
a. Pre-event scripts.
b. Post-event scripts.
c. error notification methods.
d. recovery commands.
e. notify methods.
3. How does an event script get started?
a. Manually by an administrator
b. Called by cluster manager
c. Called by a recovery program
d. Called by the topology services daemon
4. True or False?
Pre event scripts are automatically synchronized.
5. True or False?
Writing error notification methods is a normal part of configuring a
cluster.
Checkpoint Solutions
1. True or False?*
HACMP supports all NFS export configuration options.
2. Which of the following is a special consideration when using HACMP to NFS
export filesystems? (select all that apply)
a. NFS exports must be read-write.
b. Secure RPC must be used at all times.
c. A cluster may not use NFS Cross-mounts if there are client systems
accessing the NFS exported filesystems.
d. A volume group which contains filesystems which are NFS exported must
have the same major device number on all cluster nodes in the resource
group.
3. What does [/abc;/xyz] mean when specifying a directory to cross-mount?
a. /abc is the name of the filesystem which is exported and /xyz is where it
should be mounted at
b. /abc is where the filesystem should be mounted at and /xyz is the name of
the filesystem which is exported
4. True or False?**
HACMP's NFS exporting feature only supports clusters of two nodes.
5. True or False?
IPAT is required in resource groups which export NFS filesystems.
*/usr/es/sbin/cluster/exports must be used to specify NFS export options if
the default of "read-write to the world" is not acceptable.
**Resource groups larger than two nodes which export NFS filesystems do not
provide full NFS functionality (for example, NFS file locks are not preserved
across a fallover).
Copyright IBM Corporation 2005
Checkpoint Solutions
1. What is the most common cause of cluster failure?
a. Bugs in AIX or HACMP
b. Cluster administrator error
c. Marauding space aliens from another galaxy
d. Cosmic rays
e. Poor/inadequate cluster design
2. True or False?
Event emulation can emulate all cluster events.
3. If the cluster manager process should die, what will happen to the
cluster node?
a. It continues running but without HACMP to monitor and protect it.
b. It continues running AIX but any resource groups will fallover.
c. Nobody knows because this has never happened before.
d. The System Resource Controller sends an e-mail to root and issue a "halt -q".
e. The System Resource Controller sends an e-mail to root and issue a
"shutdown -F".
4. True or False?
A non-IP network is strongly recommended. Failure to include a non-IP
network can cause the cluster to fail or malfunction in rather ugly ways.
*The correct answer is almost certainly "cluster administrator error" although
"poor/inadequate cluster design" would be a very close second.
Checkpoint Solutions
1. True or False?
AIX 5.1 is compatible with HACMP 5.3.
2. To be able to do a node-by-node migration from HACMP 4.4.1
to HACMP 5.3 when AIX is at level 4.3.3 you should:
a. You can do node-by-node migration to HACMP 5.3
b. First upgrade to HACMP 4.5 and then upgrade to HACMP 5.3.
c. Nobody knows because this has never happened before.
d. None of the above
3. True or False?
When upgrading levels of HACMP, it is recommended that you upgrade one
node and then run the cluster for a week or two to see if any problems occur
before upgrading the other nodes.
4. List 3 features of HACMP 5.3
5. True or False
Always make a new mksysb of the system and create a
snapshot before upgrading to a new level of HACMP
6. If the nodes of an HACMP 4.5 cluster running AIX 5.1 need to
be upgraded to HACMP 5.3 then which should be done first?
a. HACMP 4.5 to HACMP 5.1
b. AIX 5.1 to AIX 5.2
c. HACMP 4.5 to HACMP 5.3
Checkpoint Solutions
1. For IPAT via replacement (select all that apply)
a. Each service IP address must be in the same subnet as one
of the non-service addresses
b. Each service IP address must be in the same subnet
c. Each service IP address cannot be in any non-service address
subnet
2. True or False?
If the takeover node is not the home node for the resource group
and the resource group does not have a Startup policy of Online
Using Distribution Policy, the service IP address replaces the IP
address of a NIC with an IP address in the same subnet as the
subnet of the service IP address
3. True or False?
In order to use HWAT, you must enable and complete the
ALTERNATE ETHERNET address field in the SMIT devices
menu
4. True or False?
You must stop the cluster in order to change from IPAT via
aliasing to IPAT via replacement
These release notes contain the latest information about version 5.3 of HACMP for AIX 5L.
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
PRODUCT CONSTRAINTS
- Required BOS level
- APARs Required
=======================================
NEW/ENHANCED FUNCTIONALITY IN HACMP 5.3
=======================================
This section lists only the titles of the features added in this release.
For brief descriptions of each feature, see Chapter 8 HACMP 5.3: Summary of Changes
in the Concepts and Facilities Guide.
You can access the publications directly from the /pubs directory of the HACMP Installation
CD without having to first install the filesets.
The documentation is available in HTML and PDF formats.
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
AP - If you are upgrading from a pre 5.2 release, the cllockd or cllockdES (the Cluster
Lock Manager) is no longer supported as of HACMP 5.2.
During node-by-node migration, it is uninstalled. Installing HACMP 5.2 or 5.3
removes the Lock Manager binaries and definitions.
- The network-based distribution policy for resource groups is removed. See also the
section: Network-based Distribution Policy Replaced by Node-Based Distribution.
=============================
OTHER CHANGES OR ENHANCEMENTS
=============================
NOTE: Prior to HACMP version 5.1, HACMP for AIX included four features: HAS and CRM
(with core filesets named cluster.base*); and ES and ESCRM (with core filesets named
cluster.es*). Starting with HACMP 5.1, the HAS, CRM and ES features are no longer
shipped, and the ESCRM feature is now called HACMP. That is, HACMP 5.3 refers to the
ESCRM feature of HACMP.
Since the Cluster Manager daemon is now a long running process, you cannot use lssrc -s
clstrmgrES to determine the state of the cluster.
Instead, we recommend several alternative methods:
1. Use /usr/es/sbin/cluster/utilities/clcheck_server grpsvcs instead.
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
If the grpsvcs daemon is running, then the Cluster Manager has begun to process
events.
2. Use the SMIT path Problem Determination Tools > View Current State to view the state
of the cluster.
Note that the Cluster Manager is never permanently stopped. Once it is stopped, for
example, even during a forced down event, the Cluster Manager process will automatically
be restarted. It can be in one of several states, as displayed by executing the command:
lssrc -ls clstrmgrES
ST_INIT (HACMP is configured but no events have run)
ST_NOTCONFIGURED (HACMP has yet to be configured)
For a complete table of how pre-5.2 resource group characteristics are mapped to the
combinations of startup, fallover and fallback policies available in HACMP 5.3, see chapter
10: Upgrading an HACMP Cluster in the Planning and Installation Guide.
NOTE: If you are upgrading from HACMP 5.2 to HACMP 5.3, then there is no conversion of
resource groups (they already use startup, fallover and fallback policies).
AP This is now enforced by SMIT. It is suggested that the names of resources relate to the
application they serve, as well as any corresponding device, such as:
DB2_Filesystem1, or Websphere_service_address.
NOTE: The SMIT screen for specifying the distribution policy is removed, since the only
distribution policy that HACMP 5.3 utilizes is node-based distribution. To achieve the
behavior similar to network-based distribution, use the Configure Service IP
Labels/Addresses Distribution Preferences panel in SMIT.
For those resource groups that are configured to use nodes at more than one site in their
nodelist, the clRGinfo utility now displays information about the secondary (backup)
instance of the resource group, in addition to information about primary instances.
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
You configure this setting when you use the SMIT path HACMP Resource Group and
Application Management > Move a Resource Group to Another Node/Site > Move
Resource Groups to Another Site, and select another site.
The site-specific Priority Override Location (POL) specifies the site where you want to
move the resource group.
This site POL setting overrides the settings for the nodelist for the resource group, that is,
HACMP attempts to move the resource group to the available nodes at another site, if
these nodes are included in the nodelist for this resource group. Site-specific POL
overrides node POL, if node POL is specified for the node at the local site.
The default setting is Fallover, that is, for a specified resource group, HACMP will
automatically use Selective Fallover to move the resource group to another node, at either
site. You can change this option and then HACMP will not automatically use selective
fallover for a particular resource group in case of failure. It notifies you when the resource
group goes into the error state.
Note that even if the Cluster Manager is not enabled to initiate a selective fallover across
sites, it will still move the resource group within a site if a node_down or node_up event
occurs. You can also manually move a resource group between sites.
Also, see the section Inter-Site Selective Fallover After an Upgrade to HACMP 5.3"
Online Planning Worksheets Application Can be Run from the HACMP CD-ROM
***********************************************************************
You can run the Online Planning Worksheets Application directly from the HACMP
CD-ROM.
1. Make sure that the path to the Java 2 Runtime Environment is in your PATH
environment variable.
On AIX 5.1 by default this is /usr/java130/bin.
On AIX 5.2 by default this is /usr/java131/bin.
On AIX 5.3 by default this is /usr/java140/bin.
Where cd_location is the location of the CD, and mount_directory is the name of the
directory to be mounted.
For example:
mount -vcdrfs -p -r /dev/cd0 /mnt
1. Make sure that the Java2 Runtime Environment version 1.3.0 or greater is installed on
your system.
or
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
On systems that have an association configured for .jar files, you can double-click the
worksheets.jar file.
NOTE: Do not close the command window used to launch the application.
Closing this window closes the application.
If you use the /usr/es/sbin/cluster/etc/rc.cluster script to start the Clinfo utility on client
nodes, HACMP issues an error (/usr/es/sbin/cluster/etc/ha_odm_constants: not found) and
Clinfo is not started. To enable Clinfo to start, comment out the line
. /usr/es/sbin/cluster/etc/ha_odm_constants in the rc.cluster script, or start the Clinfo utility
manually.
However, XD_data does not support any type of IPAT (IP Address Takeover), including
IPAT via IP replacement. In other words, it does not take over the IP address of ANOTHER
node. It does support local IP swapping.
Also, the XD_data network requires that you configure a node-bound service IP label for it.
===================
PRODUCT CONSTRAINTS
===================
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
******************
For detailed information on software requirements, see the migration and upgrade chapters
in the Planning and Installation Guide.
The specific HACMP 5.3 requirements for AIX 5.2 and AIX 5.3 are:
AIX 5L V5.2 with the 5200-04 Recommended Maintenance package (or later
maintenance levels) with RSCT version 2.3.6 or higher.
AIX 5L V5.3 with the 5300-02 Recommended Maintenance package (or later
maintenance levels) with RSCT version 2.4.2 or higher.
The RSCT filesets delivered with AIX 5L must be installed.
APARs Required
**************
The following APARs are required for proper operation of HACMP version 5.3.
This list represents the most current information available at this writing.
Refer to the Announcement materials for a list of the latest known fixes.
APAR IY71500
*************
This is an APAR for AIX 5L 5.3. The fix enables conversion of a volume group that contains
no logical volumes from non-concurrent capable to concurrent capable using the chvg -C
command.
APAR IY72928
************
AP Apply this APAR to enable the DARE (Dynamic Reconfiguration) functions for resource
groups with dependencies, or resource groups that have replicated resources and are
configured in clusters with sites.
================================
INSTALLATION AND MIGRATION NOTES
================================
Installation and Migration Information Included in the Planning and Installation Guide
***************************************************************
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
When you do not respond to the IA prompt (for instance, when the installation or reboot is
performed remotely), no other processes are started from /etc/inittab, and you may notice
that certain HACMP daemons have not been started. HACMP cannot control this situation.
To ensure that these and other daemons are started from /etc/inittab on nodes that you
rebooted after installing AIX 5L, respond to the AIX 5L Installation Assistant prompt and
manually enter the information at the system console. Alternatively, you can remove the IA
entry from /etc/inittab. This will enable the process of reading the /etc/inittab to proceed and
will start the clcomdES and the clstrmgrES daemons.
This format was introduced in HACMP 4.3.1 and was used along with the old format (where
the NFS mount point was not specified).
Starting with HACMP 5.2, the following changes apply:
If you configure new filesystems or change existing resource groups, only the new
format of cross-mounting is allowed.
If you upgrade from previous releases to HACMP 5.3, the old format for configuring
cross-mounted filesystems is still allowed. However, HACMP issues an error and
requires changing the format if you attempt to change the resource group, or create a
new filesystem after an upgrade.
A warning message notifies you about this change during the installation process, after
HACMP runs the cluster verification process, and during the NFS mount process.
Network types not supported in HACMP 5.3 flagged during migration from
***********************************************************************
HACMP 4.5
*********
Some network types are supported in HACMP (HAS) 4.5 but not in HACMP/ES 4.5 or
HACMP 5.1 and higher. If any of the following network types are detected during a
migration from HACMP (HAS) 4.5 to HACMP 5.3, the migration will fail during the
pre-install checks of cluster.es.server.rte:
* SOCC
* SLIP
* 802_ether (Ethernet Protocol 802.3)
* IP (Generic IP)
* FCS (Fiber Channel Switch)
An error message will alert you that the unsupported network types are detected in the
configuration and must be removed or redefined.
The only supported network types are Ether, Token, FDDI, HPS, ATM, RS232, TMSCSI,
TMSSA, DISKHB.
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
To further improve security, the HACMP Configuration Database (ODM) has the following
enhancements:
Ownership. All HACMP ODM files are owned by user root and group hacmp.
In addition, all HACMP binaries that are intended for use by non-root users are also
owned by user root, group hacmp.
Permissions. All HACMP ODM files, except for the hacmpdisksubsystem file with 600
permissions, are set with 640 permissions (that is, readable by user root and group
hacmp, writable by user root).
All HACMP binaries that are intended for use by non-root users are installed with 2555
permissions (that is, readable and executable by all users, with the setgid bit turned on
so that the program runs as group hacmp).
During the installation, HACMP creates the group hacmp on all nodes if it does not
already exist. By default, group hacmp has permission to read the HACMP ODMs, but
does not have any other special authority. For security reasons, it is recommended not to
expand the authority of group hacmp.
If you use programs that access the HACMP ODMs directly, you may need to rewrite them
if they are intended to be run by non-root users: All access to the ODM data by non-root
users should be handled via the provided HACMP utilities.
In addition, if you are using the PSSP File Collections facility to maintain the consistency of
/etc/group, the new group hacmp that is created at installation time on the individual
cluster nodes may be lost when the next file synchronization occurs.
There are two possible solutions to this problem. Before HACMP 5.3 is installed:
1) Turn off PSSP File Collections synchronization of /etc/group, or
2) Ensure that group hacmp is included in the master /etc/group file and ensure that the
change is propagated to all cluster nodes.
AP *******************************************************************
If you upgrade to HACMP 5.3 using the previously created snapshot, the migration to the
new version may get stalled if the cluster configuration in the snapshot is not 100%
accurate according to the verification check. (Also, dynamic cluster reconfiguration does
not work if verification finds any errors).
This is due to the fact that the cluster verification utility in HACMP 5.3 checks for a wider
variety of issues than in previous releases. The issues that HACMP now identifies as
incorrect could have been misconfigured in the cluster configuration in the previous
releases without necessarily breaking the cluster.
If you apply a snapshot and see an error, review the log files to check if it can be
automatically corrected in HACMP 5.3. A list of errors for which HACMPs verification utility
takes corrective actions is included in Chapter 6 of the Administration Guide.
If the error is included in the list, to continue an upgrade process, Force Apply the snapshot
and run the cluster synchronization and verification process, with the option Automatically
Correct Errors during the Cluster Verification set to Interactively.
NOTE: Be careful when Force Applying the snapshot: only use this option if you know that
the error you encountered can be automatically corrected.
Also, you may see some warnings and errors that will cause the upgrade process via a
snapshot to fail (there is no automatic corrective action for them, and they do not break the
cluster, so Force Applying the snapshot is safe). In these cases, Force Apply the snapshot
to continue an upgrade process to HACMP 5.3.
NOTE: While you are upgrading to HACMP 5.3, selective fallover of resource groups
between sites is disabled. This is the pre-5.3 release behavior for resource groups with a
non-IGNORE site management policy.
A particular instance of a resource group can fall over within one site, but cannot move
between sites.
During migration, if no nodes are available on the site where the affected instance resides,
that instance goes into ERROR or ERROR_SECONDARY state. It does not stay on the
node where it failed.
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
This behavior applies to both primary and secondary instances in releases prior to HACMP
5.3.
For a new install of HACMP 5.3, inter-site resource group recovery is instead enabled by
default. HACMP will automatically attempt to move the resource group to another node
(first at the local site and then at the remote site), in case of a resource failure.
However, if you migrated from HACMP 5.2, the inter-site resource group recovery is
disabled. After the migration to HACMP 5.3 is complete, you can change the default
behavior and enable HACMP to move a resource group between sites.
Use the HACMP SMIT path Extended Configuration > Extended Resource Configuration >
Customize Resource Group and Resource Recovery > Customize Inter-Site Resource
Group Recovery.
We recommended that when you upgrade to HACMP 5.3, you do so from/to the same
version of SNMP, otherwise your SNMP-based applications may not function correctly.
Once the upgrade has completed, you can switch to a different version of SNMP, if needed.
For example, if you are migrating from an environment using SNMPv1, and you are
upgrading to AIX 5L 5.3, then before upgrading to HACMP 5.3, run the following command:
stopsrc -s snmpd
See also the note SNMP version 3 Agents are Used with HACMP 5.3" later in this
document.
====================
POTENTIAL AIX ISSUES
====================
AP HACMP offers a suite of SMIT commands to manage the operation of HACMP cluster
environments. As part of this function, the necessary logic is included to control and
synchronize the execution of the commands on all HACMP cluster nodes.
=======================================
NOTES ON OTHER UTILITIES AND FACILITIES
=======================================
The following sections describe notes for product utilities:
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Note: Importing the HACMP definition is supported only from the cluster nodes (and not, for
instance, from a Windows 2000 machine).
In addition, you can also create a cluster definition file from an active HACMP cluster.
To create a cluster definition file from an active HACMP cluster, export the definition file
from SMIT (Extended Configuration > Export Definition File for Online Planning
Worksheets), and open the file from the Online Planning Worksheets application.
Also, if you choose to create a cluster definition as XML, ensure that constructions of the
type <tag/> are written as follows: <tag></tag>. This affects such entities as
<VALIDATED/>.
On networks configured to use IP Aliasing for IP Address Takeover (aka, IPAT via Aliasing),
HACMP will briefly define the Service IP address on both the old (failed) network interface
and the new (takeover) network interface in order to properly preserve network routes. The
Service IP address is then removed from the old network interface. During the brief time
AP that the service IP address exists on both network interfaces, AIX 5L may detect this
situation and add an entry in the error log similar to:
This condition is only temporary (during the IPAT operation) and the error log entry can
safely be ignored.
NOTE: The filesets cluster.rpv and cluster.rpv.msg.en_US are renamed. For details, see
the Installation and Migration Notes section earlier in this file.
We recommended that when you upgrade to HACMP 5.3, you do so from/to the same
version of SNMP.
For details, see the note The Same Version of SNMP Should be Used During Upgrades
earlier in this document.
======================================
USER DOCUMENTATION LOCATION AND TITLES
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
======================================
Insert the HACMP installation CD and find the documents under the top-level /pubs
directory. See step 7 below for an explanation of the files contained in the HTML directory.
1. At the command line, enter: smit install_selectable_all SMIT asks for the input
device/directory for software.
2. Select the CD ROM drive from the picklist and press Enter.
3. On the next SMIT screen with the cursor on Software to Install, press the F4 key.
NOTE: Installing all of the documentation requires about 27 MB of space in the /usr
filesystem. (PDF files = 15 MB, HTML files = 12 MB.)
/usr/share/man/info/en_US/cluster/HAES
6. For the PDF files, each book directory contains the books PDF file.
7. For the HTML files, each book directory contains a number of files: separate .html files
for each chapter, .gifs, and a .css file for each book. The MAIN FILE (the one you
should click on) for each book is an .htm file with a filename that begins with ha and is
related to the name of the book (for example, ha_concepts.htm).
When you click on this main file, the book opens in the browser window with a Contents
frame that allows you to navigate easily to any chapter.
See the Document Titles and Filenames list below for exact filenames for the book
titles (both .htm and .pdf).
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
==========================
PRODUCT DIRECTORIES LOADED
==========================
Use the following command to determine the exact files loaded into product directories
when installing the HACMP for AIX 5L, version 5.3:
lslpp -f cluster*
/etc/inetd.conf
/etc/inittab
/etc/group
/etc/objrepos/SRCnotify
/etc/objrepos/SRCsubsvr
/etc/objrepos/SRCsubsys
/etc/rc.net
/etc/services
/etc/snmpd.conf
/etc/snmpd.peers
/etc/syslog.conf
/var/spool/cron/crontabs/root
=================
PRODUCT MAN PAGES
=================
Man pages for HACMP commands and utilities are installed in the following directory:
/usr/share/man/cat1
/usr/share/man/cat1/clmixver.1
/usr/share/man/cat1/clwahs_import.1
/usr/share/man/cat1/cl_opsconfig.1
Copyright IBM Corp. 1998, 2005 Appendix B. Release Notes for HACMP 5.3 B-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
References
SC23-4867-05 HACMP for AIX: HACMP Master Glossary
SC23-4864-06 HACMP for AIX: Concepts and Facilities Guide
SC23-4861-06 HACMP for AIX: Planning and Installation Guide
SC23-4862-06 HACMP for AIX: Administration Guide
SC23-5177-00 HACMP for AIX: Troubleshooting Guide
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-1
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Objectives
After completing this unit, you should be able to:
Explain and set up IP Address Takeover (IPAT) via IP
replacement
Notes:
AP
IPAT via IP Replacement Configuration
Define each networks boot IP addresses in the AIX ODM.
Each interface IP address on a given node must be in a different logical IP subnet*
and there must be a common subnet among the nodes
Define these address in the /etc/hosts file and configure them in HACMP topology
Define service IP addresses in /etc/hosts and HACMP resources
The address must be in the SAME subnet as a common interface subnet
HACMP configures them to AIX as required
* See earlier discussion of heartbeating and failure diagnosis for explanation of why
Notes:
Requirements
Keep the following items in mind when you configure a network for IPAT via IP
replacement:
- There must be at least one logical IP subnet which has a communication interface
(NIC) on each node. (In HACMP 4.5 terminology, these were called boot adapters.)
- Each service IP address must be in the same logical IP subnet as one of the
non-service addresses. Contrast with IPAT via IP aliasing, where service addresses
are required to NOT be in a boot subnet.
- If you have more than one service IP address, they must all be in the same subnet.
The reason for this will become clear when we discuss what happens during a
takeover, see IPAT via IP Replacement after a Node Fails on page 8.
- None of the other non-service addresses may be in the same subnet as the service
IP address (this is true regardless of whether IPAT via IP replacement is being used
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-3
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
subnet IP labels
n1-if1, n2-if1, appA-svc,
192.168.10/24
appB-svc
192.168.11/24 n1-if2, n2-if2
AP
IPAT via IP Replacement in Operation
When the resource group comes up on a node, HACMP replaces
an boot (ODM) IP label with the service IP label
It replaces the boot IP label on the same subnet if the resource
group is on its startup node or if the distribution startup
policy is used.
It replaces a boot IP label on a different subnet otherwise
Notes:
Operation
When the resource group comes up on its home node, the resource groups service IP
address replaces the interface IP address of the NIC (AIX ODM) which is in the same
subnet as the service IP label (that is, the boot adapter in HACMP 4.x terminology).
Note that this approach implies that there cannot be two resource groups in the cluster
which both use IPAT via IP replacement and use the same node as their home node
unless their respective service IP addresses are in different subnets (in other words,
associated with different physical networks).
Also, since the service IP address replaces the existing IP address on the NIC, it is not
possible to have two or more service IP addresses in the same resource group which
are in the same IP subnet (as there will not be an adapter to assign the second service
IP address to).
When the resource group comes up on any node other than its home node, the
resource groups service IP address replaces the interface IP address of one of the
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-5
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
NICs which is not in the same subnet as the service IP address (this is primarily to allow
some other resource group to use the node as its home node).
AP
IPAT via IP Replacement After an I/F Fails
If the communication interface being used for the service IP label
fails, HACMP swaps the service IP label with a boot (ODM) IP label
on one of the node's remaining available (that is, currently
functional) communication interfaces
The IP labels remain swapped when the failed interface recovers
NIC A NIC B
9.47.11.1 (ODM) 9.47.10.22 (service) 9.47.10.2 (ODM) 9.47.11.2 (ODM)
Notes:
Interface failure
If a communications interface (NIC A) which is currently assigned an IPAT via IP
replacement service IP address fails, then HACMP moves the service IP address to
one of the other communication interfaces (NIC B) on the same node (to one of the
standby adapters using HACMP 4.x terminology).
If there are no available (that is, functional) NICs left the relevant network then HACMP
initiates a fallover.
Interface swap
The failed communications interface (NIC A) is then reconfigured with the address of
the communication interface (NIC B) as this allows the heartbeat mechanism to watch
for when the failed communication interface (NIC A) recovers.
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-7
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Node failure
If the node currently responsible for an IPAT via IP replacement using resource group
fails then HACMP initiates a fallover. When the resource group comes up on the
takeover node, the service IP addresses are assigned to NICs on the fallover node:
- Home node or Startup policy of Online Using Distribution Policy (rotate in
HACMP 4.x terminology)
If the takeover node is the home node for the resource group or the resource group
has a Startup policy of Online Using Distribution Policy (rotate in HACMP 4.x
terminology), the Service IP addresses replace the IP addresses of a
communications interface (NIC) with an IP address in the same subnet as the
service IP address.
- Not the home node and not Online Using Distribution Policy
If the takeover node is not the home node for the resource group and the resource
group does not have a Startup policy of Online Using Distribution Policy, the
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-9
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Advantages
Probably the most significant advantage of IPAT via IP replacement is that it supports
hardware address takeover (HWAT), which will be discussed in a few pages.
Another advantage is that it requires fewer subnets. If you are limited in the number of
subnets available for your cluster, this may be important.
Note: Another alternative, if you are limited on the number of subnets you have
available, is to use heartbeating via IP aliases. See Heartbeating Over IP Aliases in the
HACMP for AIX 5L Planning and Installation Guide.
Disadvantages
Probably the most significant disadvantages are that IPAT via IP replacement limits the
number of service IP labels per subnet per resource group on one communications
interface to one and makes it rather expensive (and complex) to support lots of
AP resource groups in a small cluster. In other words, you need more network adapters to
support more applications.
Also, IPAT via replacement usually takes more time than IPAT via aliasing.
Note that HACMP tries to keep the service IP Labels available by swapping IP
addresses with other communication interfaces (standby adapters in HACMP 4.x
terminology) even if there are no resource groups currently on the node which uses
IPAT via IP replacement.
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-11
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Review
When using IPAT via aliasing, you can use AIXs gratuitous ARP features to update
client and router ARP caches after a takeover. However, there may be issues.
AP to be causing the cluster and the cluster administrator far more serious problems than
the ARP cache issue involves).
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-13
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Suggestion:
Do not get involved with using either clinfo or HWAT to deal with
ARP cache issues until you've verified that there actually are ARP
issues which need to be dealt with.
Notes:
AP
Option 3: Hardware Address Takeover
HACMP can be configured to swap a service IP label's hardware
address between network adapters.
HWAT is incompatible with IPAT via IP aliasing because each
service IP address must have its own hardware address and a NIC
can support only one hardware address at any given time.
Cluster implementer designates a Locally Administered Address
(LAA) which HACMP assigns to the NIC which has the service IP
label
Notes:
HWAT considerations
There are a few points which must be kept in mind when contemplating HWAT:
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-15
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
- The hardware address which is associated with the service IP address must be
unique within the physical network that the service IP address is configured for.
- HWAT is not supported by IPAT via IP aliasing because each NIC can have more
than one IP address but each NIC can only have one hardware address.
- HWAT is only supported for Ethernet, token ring and FDDI networks (MCA FDDI
network cards do not support HWAT). ATM networks do not support HWAT.
- HWAT increases the takeover time (usually by just a few seconds).
- HWAT is an optional capability which must be configured into the HACMP cluster
(we will see how to do that in a few minutes).
- Cluster nodes using HWAT on token ring networks must be configured to reboot
after a system crash as the token ring card will continue to intercept packets for its
hardware address until the node starts to reboot.
AP
Hardware Address Takeover (1 of 3)
bondar-if1 hudson-if1
9.47.9.1 9.47.9.2
tr1 255.255.255.0 tr1
255.255.255.0
tr1
00:04:ac:48:22:f4 00:04:ac:62:72:61
Before hudson-if2
bondar-if2
tr0 9.47.5.3
255.255.255.0
resource group 9.47.5.2
255.255.255.0 tr0
00:04:ac:48:22:f6
00:04:ac:62:72:49
is started
Bondar Hudson
hudson-if1
bondar-if1 9.47.9.2
9.47.9.1 255.255.255.0
tr1 255.255.255.0 00:04:ac:62:72:61
00:04:ac:48:22:f4
After hudson-if2
xweb
tr0 9.47.5.1
255.255.255.0
resource group 9.47.5.2
255.255.255.0 tr0
00:04:ac:48:22:f6
40:04:ac:62:72:49
is started
Bondar Hudson
Notes:
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-17
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Bondar Hudson
xweb
bondar-if1 9.47.5.1
9.47.9.1 255.255.255.0
255.255.255.0 40:04:ac:62:72:49
00:04:ac:48:22:f4
Node failure hudson-if2
xweb 9.47.5.2
9.47.5.1 255.255.255.0
255.255.255.0 00:04:ac:48:22:f6
40:04:ac:62:72:49
Bondar Hudson
Notes:
AP
Hardware Address Takeover (3 of 3)
xweb
bondar-if1 9.47.5.1
9.47.9.1 255.255.255.0
tr1 255.255.255.0 40:04:ac:62:72:49 tr1
00:04:ac:48:22:f4
When a failed node
hudson-if2
bondar-if2 comes back to life, 9.47.5.2
tr0 9.47.5.3
tr0
255.255.255.0 the burned in ROM 255.255.255.0
00:04:ac:48:22:f6
00:04:ac:62:72:49 Address is used on
the service network
adapter.
Bondar Hudson
hudson-if1
bondar-if1 9.47.9.2
9.47.9.1 255.255.255.0
tr1 255.255.255.0 00:04:ac:62:72:61 tr1
00:04:ac:48:22:f4
After HACMP is
xweb
started the node hudson-if2
9.47.5.2
tr0 9.47.5.1 reintegrates 255.255.255.0 tr0
255.255.255.0
40:04:ac:62:72:49 according to its 00:04:ac:48:22:f6
resource group
parameters
Bondar Hudson
Notes:
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-19
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
bondar hudson
D D
A A
Notes:
Reality check
A side note is probably in order: although most TCP/IP-capable systems respect
gratuitous ARP, there are strange devices out there that do not. This scenario is phoney
but it presents a real if rather unlikely problem. For example, the ATM network does not
support gratuitous ARP and so could be a candidate for the use of HWAT.
AP
Our Plan for Implementing HWAT
1. Stop cluster services on both cluster nodes
Use the graceful shutdown option to bring down the resource groups and
their applications
2. Remove the alias service labels from the Resources
They are in the wrong subnet for replacement
They are automatically removed from the RG
3. Convert the net_ether_01 Ethernet network to use IPAT via IP
replacement:
Disable IPAT via IP aliasing on the Ethernet network.
Update /etc/hosts on both cluster nodes to describe service IP labels and
addresses on the 192.168.15.0 subnet
Use the procedure described in the networking to select the (Locally
Administered Addresses (LAA) addresses
Configure new service IP labels with these LAA addresses in the HACMP
SMIT screens
4. Define resource groups to use the new service IP labels.
5. Synchronize the changes
6. Restart HACMP on the two nodes.
Notes:
Implementing HWAT
In order to use HWAT, we must use IPAT via replacement.
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-21
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
AP
Stopping HACMP
# smit clstop
Stop Cluster Services
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* Stop now, on system restart or both now +
Stop Cluster Services on these nodes [bondar,hudson] +
BROADCAST cluster shutdown? true +
* Shutdown mode graceful +
Notes:
Stop HACMP
Make sure that HACMP is shut down gracefully, as we cant have the application
running while we are changing service IP addresses.
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-23
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
+--------------------------------------------------------------------------+
Select Service IP Label(s)/Address(es) to Remove
Move cursor to desired item and press F7.
ONE OR MORE items can be selected.
Press Enter AFTER making all selections.
xweb
yweb
zweb
F1=Help F2=Refresh F3=Cancel
F7=Select F8=Image F10=Exit
F1 Enter=Do /=Find n=Find Next
F9+--------------------------------------------------------------------------+
Notes:
AP
Disable IPAT via Aliases
Set the "Enable IP Address Takeover via IP Aliases" setting to "No"
and press Enter.
Notes:
Introduction
Here we change the net_ether_01 network to disable IPAT via aliasing.
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-25
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
AP
Creating a
Locally Administered Address (LAA)
Each service IP label using HWAT will need an LAA
The LAA must be unique on the cluster's physical network
The MAC address based technologies (Ethernet, Token ring and
FDDI) use six byte hardware addresses of the form:
xx.xx.xx.xx.xx.xx
The factory-set MAC address of the NIC will start with 0, 1, 2 or 3
A MAC address that starts with 0, 1, 2 or 3 is called a Globally
Administered Address (GAA) because it is assigned to the
NIC's vendor by a central authority
Incrementing this first digit by 4 transforms the GAA into a Locally
Administered Address (LAA) which will be unique worldwide
(unless someone has already used the same GAA to create an
LAA which isn't likely since GAAs are unique worldwide)
Notes:
Hardware addresses
Hardware addresses must be unique, at a minimum, on the local network to which they
are connected. The factory set hardware address for each network interface card (NIC)
is administered by a central authority and should be unique in the world. These
addresses are called Globally Administered Addresses (GAAs).
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-27
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
address is a GAA. Setting this bit to one indicates that this address is locally
administered.
AP
Creating Two LAAs for Our Cluster
Here are two Globally Administered Addresses (GAAs)
taken from Ethernet adapters in the cluster:
0.4.ac.17.19.64
0.6.29.ac.46.8
First we make sure that each number is two digits long by
adding leading zeros as necessary:
00.04.ac.17.19.64
00.06.29.ac.46.08
Verify that the first digit is 0, 1, 2 or 3:
Yep!
Add 4 to the first digit of each GAA:
40.04.ac.17.19.64
40.06.29.ac.46.08
Done! These two addresses are now LAAs
Notes:
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-29
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Issues
The main thing to remember is that you do NOT configure the ALTERNATE hardware
address field in the SMIT devices panel.
You MUST leave that blank and configure this using the SMIT HACMP menus.
AP
Redefining the Service IP Labels for HWAT
Redefine the two service IP labels. Note that the periods are stripped
out before the LAA is entered into the HW Address field.
Add a Service IP Label/Address configurable on Multiple Nodes (extended)
Type or select values in entry fields.
Press Enter AFTER making all desired changes.
[Entry Fields]
* IP Label/Address [xweb] +
* Network Name net_ether_01
Alternate HW Address to accompany IP Label/Address [4004ac171964]
Don't forget to specify the second LAA for the second service IP label.
Copyright IBM Corporation 2005
Notes:
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-31
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Notes:
Synchronize
Dont forget to synchronize.
AP
Checkpoint
1. For IPAT via replacement (select all that apply)
a. Each service IP address must be in the same subnet as one of
the non-service addresses
b. Each service IP address must be in the same subnet
c. Each service IP address cannot be in any non-service address
subnet
2. True or False?
If the takeover node is not the home node for the resource group
and the resource group does not have a Startup policy of Online
Using Distribution Policy, the service IP address replaces the IP
address of a NIC with an IP address in the same subnet as the
subnet of the service IP address
3. True or False?
In order to use HWAT, you must enable and complete the
ALTERNATE ETHERNET address field in the SMIT devices
menu
4. True or False?
You must stop the cluster in order to change from IPAT via
aliasing to IPAT via replacement
Notes:
Copyright IBM Corp. 1998, 2005 Appendix C. IPAT via IP Replacement C-33
Course materials may not be reproduced in whole or in part
without the prior written permission of IBM.
Student Notebook
Unit Summary
IPAT via IP replacement:
May require fewer subnets than IPAT via aliasing
May require more NICs than IPAT via aliasing
Supports hardware address takeover
HACMP replaces non-service IP labels with service IP labels on the
same subnet as the service IP label when the resource group is
started on its home node or if the Startup Policy is distributed
HACMP replaces non-service IP labels with service IP labels on a
different subnet from the service IP label when the resource group is
moved to any other node
IPAT via IP replacement configuration issues
Service IP address must be the same subnet as one of the non-service subnets
All service IP addresses must be in the same subnet
You must have at least as many NICs on each node as service IP addresses
Hardware Address Takeover (HWAT) issues
Alternate hardware address (Locally Administered Address or LAA) must be
configured in HACMP. Do NOT use standard SMIT field.
Alternate hardware address must be unique.
Notes:
backpg
Back page