9. Upgrading a Pacemaker Cluster

9.1. Pacemaker Versioning

Pacemaker has an overall release version, plus separate version numbers for certain internal components.

  • Pacemaker release version: This version consists of three numbers (x.y.z).

    The major version number (the x in x.y.z) increases when at least some rolling upgrades are not possible from the previous major version. For example, a rolling upgrade from 1.0.8 to 1.1.15 should always be supported, but a rolling upgrade from 1.0.8 to 2.0.0 may not be possible.

    The minor version (the y in x.y.z) increases when there are significant changes in cluster default behavior, tool behavior, and/or the API interface (for software that utilizes Pacemaker libraries). The main benefit is to alert you to pay closer attention to the release notes, to see if you might be affected.

    The release counter (the z in x.y.z) is increased with all public releases of Pacemaker, which typically include both bug fixes and new features.

  • CRM feature set: This version number applies to the communication between full cluster nodes, and is used to avoid problems in mixed-version clusters.

    The major version number increases when nodes with different versions would not work (rolling upgrades are not allowed). The minor version number increases when mixed-version clusters are allowed only during rolling upgrades. The minor-minor version number is ignored, but allows resource agents to detect cluster support for various features. [1]

    Pacemaker ensures that the longest-running node is the cluster’s DC. This ensures new features are not enabled until all nodes are upgraded to support them.

  • Pacemaker Remote protocol version: This version applies to communication between a Pacemaker Remote node and the cluster. It increases when an older cluster node would have problems hosting the connection to a newer Pacemaker Remote node. To avoid these problems, Pacemaker Remote nodes will accept connections only from cluster nodes with the same or newer Pacemaker Remote protocol version.

    Unlike with CRM feature set differences between full cluster nodes, mixed Pacemaker Remote protocol versions between Pacemaker Remote nodes and full cluster nodes are fine, as long as the Pacemaker Remote nodes have the older version. This can be useful, for example, to host a legacy application in an older operating system version used as a Pacemaker Remote node.

  • XML schema version: Pacemaker’s configuration syntax — what’s allowed in the Configuration Information Base (CIB) — has its own version. This allows the configuration syntax to evolve over time while still allowing clusters with older configurations to work without change.

9.2. Upgrading Cluster Software

There are three approaches to upgrading a cluster, each with advantages and disadvantages.

Upgrade Methods
Method Available between all versions Can be used with Pacemaker Remote nodes Service outage during upgrade Service recovery during upgrade Exercises failover logic Allows change of messaging layer [2]
Complete cluster shutdown yes yes always N/A no yes
Rolling (node by node) no yes always [3] yes yes no
Detach and reattach yes no only due to failure no no yes

9.2.1. Complete Cluster Shutdown

In this scenario, one shuts down all cluster nodes and resources, then upgrades all the nodes before restarting the cluster.

  1. On each node:
    1. Shutdown the cluster software (pacemaker and the messaging layer).
    2. Upgrade the Pacemaker software. This may also include upgrading the messaging layer and/or the underlying operating system.
    3. Check the configuration with the crm_verify tool.
  2. On each node:
    1. Start the cluster software.

Currently, only Corosync version 2 and greater is supported as the cluster layer, but if another stack is supported in the future, the stack does not need to be the same one before the upgrade.

One variation of this approach is to build a new cluster on new hosts. This allows the new version to be tested beforehand, and minimizes downtime by having the new nodes ready to be placed in production as soon as the old nodes are shut down.

9.2.2. Rolling (node by node)

In this scenario, each node is removed from the cluster, upgraded, and then brought back online, until all nodes are running the newest version.

Special considerations when planning a rolling upgrade:

  • If you plan to upgrade other cluster software – such as the messaging layer – at the same time, consult that software’s documentation for its compatibility with a rolling upgrade.
  • If the major version number is changing in the Pacemaker version you are upgrading to, a rolling upgrade may not be possible. Read the new version’s release notes (as well the information here) for what limitations may exist.
  • If the CRM feature set is changing in the Pacemaker version you are upgrading to, you should run a mixed-version cluster only during a small rolling upgrade window. If one of the older nodes drops out of the cluster for any reason, it will not be able to rejoin until it is upgraded.
  • If the Pacemaker Remote protocol version is changing, all cluster nodes should be upgraded before upgrading any Pacemaker Remote nodes.

See the ClusterLabs wiki’s release calendar to figure out whether the CRM feature set and/or Pacemaker Remote protocol version changed between the the Pacemaker release versions in your rolling upgrade.

To perform a rolling upgrade, on each node in turn:

  1. Put the node into standby mode, and wait for any active resources to be moved cleanly to another node. (This step is optional, but allows you to deal with any resource issues before the upgrade.)
  2. Shutdown the cluster software (pacemaker and the messaging layer) on the node.
  3. Upgrade the Pacemaker software. This may also include upgrading the messaging layer and/or the underlying operating system.
  4. If this is the first node to be upgraded, check the configuration with the crm_verify tool.
  5. Start the messaging layer. This must be the same messaging layer (currently only Corosync version 2 and greater is supported) that the rest of the cluster is using.

Note

Even if a rolling upgrade from the current version of the cluster to the newest version is not directly possible, it may be possible to perform a rolling upgrade in multiple steps, by upgrading to an intermediate version first.

Version Compatibility Table
Version being Installed Oldest Compatible Version
Pacemaker 2.y.z Pacemaker 1.1.11 [4]
Pacemaker 1.y.z Pacemaker 1.0.0
Pacemaker 0.7.z Pacemaker 0.6.z

9.2.3. Detach and Reattach

The reattach method is a variant of a complete cluster shutdown, where the resources are left active and get re-detected when the cluster is restarted.

This method may not be used if the cluster contains any Pacemaker Remote nodes.

  1. Tell the cluster to stop managing services. This is required to allow the services to remain active after the cluster shuts down.

    # crm_attribute --name maintenance-mode --update true
    
  2. On each node, shutdown the cluster software (pacemaker and the messaging layer), and upgrade the Pacemaker software. This may also include upgrading the messaging layer. While the underlying operating system may be upgraded at the same time, that will be more likely to cause outages in the detached services (certainly, if a reboot is required).

  3. Check the configuration with the crm_verify tool.

  4. On each node, start the cluster software. Currently, only Corosync version 2 and greater is supported as the cluster layer, but if another stack is supported in the future, the stack does not need to be the same one before the upgrade.

  5. Verify that the cluster re-detected all resources correctly.

  6. Allow the cluster to resume managing resources again:

    # crm_attribute --name maintenance-mode --delete
    

Note

While the goal of the detach-and-reattach method is to avoid disturbing running services, resources may still move after the upgrade if any resource’s location is governed by a rule based on transient node attributes. Transient node attributes are erased when the node leaves the cluster. A common example is using the ocf:pacemaker:ping resource to set a node attribute used to locate other resources.

9.3. Upgrading the Configuration

The CIB schema version can change from one Pacemaker version to another.

After cluster software is upgraded, the cluster will continue to use the older schema version that it was previously using. This can be useful, for example, when administrators have written tools that modify the configuration, and are based on the older syntax. [5]

However, when using an older syntax, new features may be unavailable, and there is a performance impact, since the cluster must do a non-persistent configuration upgrade before each transition. So while using the old syntax is possible, it is not advisable to continue using it indefinitely.

Even if you wish to continue using the old syntax, it is a good idea to follow the upgrade procedure outlined below, except for the last step, to ensure that the new software has no problems with your existing configuration (since it will perform much the same task internally).

If you are brave, it is sufficient simply to run cibadmin --upgrade.

A more cautious approach would proceed like this:

  1. Create a shadow copy of the configuration. The later commands will automatically operate on this copy, rather than the live configuration.

    # crm_shadow --create shadow
    
  1. Verify the configuration is valid with the new software (which may be stricter about syntax mistakes, or may have dropped support for deprecated features):

    # crm_verify --live-check
    
  2. Fix any errors or warnings.

  3. Perform the upgrade:

    # cibadmin --upgrade
    
  4. If this step fails, there are three main possibilities:

    1. The configuration was not valid to start with (did you do steps 2 and 3?).
    2. The transformation failed; report a bug.
    3. The transformation was successful but produced an invalid result.

    If the result of the transformation is invalid, you may see a number of errors from the validation library. If these are not helpful, visit the Validation FAQ wiki page and/or try the manual upgrade procedure described below.

  5. Check the changes:

    # crm_shadow --diff
    

    If at this point there is anything about the upgrade that you wish to fine-tune (for example, to change some of the automatic IDs), now is the time to do so:

    # crm_shadow --edit
    

    This will open the configuration in your favorite editor (whichever is specified by the standard $EDITOR environment variable).

  6. Preview how the cluster will react:

    # crm_simulate --live-check --save-dotfile shadow.dot -S
    # dot -Tsvg shadow.dot -o shadow.svg
    

    You can then view shadow.svg with any compatible image viewer or web browser. Verify that either no resource actions will occur or that you are happy with any that are scheduled. If the output contains actions you do not expect (possibly due to changes to the score calculations), you may need to make further manual changes. See Simulate Cluster Activity with crm_simulate for further details on how to interpret the output of crm_simulate and dot.

  7. Upload the changes:

    # crm_shadow --commit shadow --force
    

    In the unlikely event this step fails, please report a bug.

Note

It is also possible to perform the configuration upgrade steps manually:

  1. Locate the upgrade*.xsl conversion scripts provided with the source code. These will often be installed in a location such as /usr/share/pacemaker, or may be obtained from the source repository.

  2. Run the conversion scripts that apply to your older version, for example:

    # xsltproc /path/to/upgrade06.xsl config06.xml > config10.xml
    
  3. Locate the pacemaker.rng script (from the same location as the xsl files).

  4. Check the XML validity:

    # xmllint --relaxng /path/to/pacemaker.rng config10.xml
    

The advantage of this method is that it can be performed without the cluster running, and any validation errors are often more informative.

9.4. What Changed in 2.1

The Pacemaker 2.1 release is fully backward-compatible in both the CIB XML and the C API. Highlights:

  • Pacemaker now supports the OCF Resource Agent API version 1.1. Most notably, the Master and Slave role names have been renamed to Promoted and Unpromoted.
  • Pacemaker now supports colocations where the dependent resource does not affect the primary resource’s placement (via a new influence colocation constraint option and critical resource meta-attribute). This is intended for cases where a less-important resource must be colocated with an essential resource, but it is preferred to leave the less-important resource stopped if it fails, rather than move both resources.
  • If Pacemaker is built with libqb 2.0 or later, the detail log will use millisecond-resolution timestamps.
  • In addition to crm_mon and stonith_admin, the crmadmin, crm_resource, crm_simulate, and crm_verify commands now support the --output-as and --output-to options, including XML output (which scripts and higher-level tools are strongly recommended to use instead of trying to parse the text output, which may change from release to release).

For a detailed list of changes, see the release notes and the Pacemaker 2.1 Changes page on the ClusterLabs wiki.

9.5. What Changed in 2.0

The main goal of the 2.0 release was to remove support for deprecated syntax, along with some small changes in default configuration behavior and tool behavior. Highlights:

  • Only Corosync version 2 and greater is now supported as the underlying cluster layer. Support for Heartbeat and Corosync 1 (including CMAN) is removed.
  • The Pacemaker detail log file is now stored in /var/log/pacemaker/pacemaker.log by default.
  • The record-pending cluster property now defaults to true, which allows status tools such as crm_mon to show operations that are in progress.
  • Support for a number of deprecated build options, environment variables, and configuration settings has been removed.
  • The master tag has been deprecated in favor of using the clone tag with the new promotable meta-attribute set to true. “Master/slave” clone resources are now referred to as “promotable” clone resources.
  • The public API for Pacemaker libraries that software applications can use has changed significantly.

For a detailed list of changes, see the release notes and the Pacemaker 2.0 Changes page on the ClusterLabs wiki.

9.6. What Changed in 1.0

9.6.1. New

  • Failure timeouts.
  • New section for resource and operation defaults.
  • Tool for making offline configuration changes.
  • Rules, instance_attributes, meta_attributes and sets of operations can be defined once and referenced in multiple places.
  • The CIB now accepts XPath-based create/modify/delete operations. See cibadmin --help.
  • Multi-dimensional colocation and ordering constraints.
  • The ability to connect to the CIB from non-cluster machines.
  • Allow recurring actions to be triggered at known times.

9.6.2. Changed

  • Syntax

    • All resource and cluster options now use dashes (-) instead of underscores (_)
    • master_slave was renamed to master
    • The attributes container tag was removed
    • The operation field pre-req has been renamed requires
    • All operations must have an interval, start/stop must have it set to zero
  • The stonith-enabled option now defaults to true.

  • The cluster will refuse to start resources if stonith-enabled is true (or unset) and no STONITH resources have been defined

  • The attributes of colocation and ordering constraints were renamed for clarity.

  • resource-failure-stickiness has been replaced by migration-threshold.

  • The parameters for command-line tools have been made consistent

  • Switched to ‘RelaxNG’ schema validation and ‘libxml2’ parser

    • id fields are now XML IDs which have the following limitations:

      • id’s cannot contain colons (:)
      • id’s cannot begin with a number
      • id’s must be globally unique (not just unique for that tag)
    • Some fields (such as those in constraints that refer to resources) are IDREFs.

      This means that they must reference existing resources or objects in order for the configuration to be valid. Removing an object which is referenced elsewhere will therefore fail.

    • The CIB representation, from which a MD5 digest is calculated to verify CIBs on the nodes, has changed.

      This means that every CIB update will require a full refresh on any upgraded nodes until the cluster is fully upgraded to 1.0. This will result in significant performance degradation and it is therefore highly inadvisable to run a mixed 1.0/0.6 cluster for any longer than absolutely necessary.

  • Ping node information no longer needs to be added to ha.cf. Simply include the lists of hosts in your ping resource(s).

9.6.3. Removed

  • Syntax
    • It is no longer possible to set resource meta options as top-level attributes. Use meta-attributes instead.
    • Resource and operation defaults are no longer read from crm_config.

Footnotes

[1]Before CRM feature set 3.1.0 (Pacemaker 2.0.0), the minor-minor version number was treated the same as the minor version.
[2]Currently, Corosync version 2 and greater is the only supported cluster stack, but other stacks have been supported by past versions, and may be supported by future versions.
[3]Any active resources will be moved off the node being upgraded, so there will be at least a brief outage unless all resources can be migrated “live”.
[4]Rolling upgrades from Pacemaker 1.1.z to 2.y.z are possible only if the cluster uses corosync version 2 or greater as its messaging layer, and the Cluster Information Base (CIB) uses schema 1.0 or higher in its validate-with property.
[5]As of Pacemaker 2.0.0, only schema versions pacemaker-1.0 and higher are supported (excluding pacemaker-1.1, which was a special case).