6. Resource Operations

Operations are actions the cluster can perform on a resource by calling the resource agent. Resource agents must support certain common operations such as start, stop, and monitor, and may implement any others.

Operations may be explicitly configured for two purposes: to override defaults for options (such as timeout) that the cluster will use whenever it initiates the operation, and to run an operation on a recurring basis (for example, to monitor the resource for failure).

An OCF resource with a non-default start timeout

<primitive id="Public-IP" class="ocf" type="IPaddr" provider="heartbeat">
  <operations>
     <op id="Public-IP-start" name="start" timeout="60s"/>
  </operations>
  <instance_attributes id="params-public-ip">
     <nvpair id="public-ip-addr" name="ip" value="192.0.2.2"/>
  </instance_attributes>
</primitive>

Pacemaker identifies operations by a combination of name and interval, so this combination must be unique for each resource. That is, you should not configure two operations for the same resource with the same name and interval.

6.1. Operation Properties

The id, name, interval, and role operation properties may be specified only as XML attributes of the op element. Other operation properties may be specified in any of the following ways, from highest precedence to lowest:

• directly in the op element as an XML attribute

• in an nvpair element within a meta_attributes element within the op element

• in an nvpair element within a meta_attributes element within operation defaults

If not specified, the default from the table below is used.

Operation Properties

Name	Type	Default	Description
id	id		A unique identifier for the XML element (required)
name	text		An action name supported by the resource agent (required)
interval	duration	0	If this is a positive value, Pacemaker will schedule recurring instances of this operation at the given interval (which makes sense only with name set to monitor). If this is 0, Pacemaker will apply other properties configured for this operation to instances that are scheduled as needed during normal cluster operation. (required)
description	text		Arbitrary text for user’s use (ignored by Pacemaker)
role	enumeration		If this is set, the operation configuration applies only on nodes where the cluster expects the resource to be in the specified role. This makes sense only for recurring monitors. Allowed values: Started, Stopped, and in the case of promotable clone resources, Unpromoted and Promoted.
timeout	timeout	20s	If resource agent execution does not complete within this amount of time, the action will be considered failed. Note: timeouts for fencing agents are handled specially (see the Fencing chapter).
on-fail	enumeration	If name is stop: fence if stonith-enabled is true, otherwise block If name is demote: on-fail of the monitor action with role set to Promoted, if present, enabled, and configured to a value other than demote, or restart otherwise Otherwise: restart	How the cluster should respond to a failure of this action. Allowed values: ignore: Pretend the resource did not fail block: Do not perform any further operations on the resource stop: Stop the resource and leave it stopped demote: Demote the resource, without a full restart. This is valid only for promote actions, and for monitor actions with both a nonzero interval and role set to Promoted; for any other action, a configuration error will be logged, and the default behavior will be used. (since 2.0.5) restart: Stop the resource, and start it again if allowed (possibly on a different node) fence: Fence the node on which the resource failed standby: Put the node on which the resource failed in standby mode (forcing all resources away)
enabled	boolean	true	If false, ignore this operation definition. This does not suppress all actions of this type, but is typically used to pause a recurring monitor. This can complement the resource being unmanaged (is-managed set to false), which does not stop recurring operations. Maintenance mode, which does stop configured monitors, overrides this setting.
interval-origin	ISO 8601		If set for a recurring action, the action will be scheduled for this time plus a multiple of the action’s interval, rather than immediately after the resource gains the monitored role. For example, you might schedule an in-depth monitor to run once per day outside business hours, by setting this to the desired time (on any date) and setting interval to 24h. At most one of interval-origin and start-delay may be set.
start-delay	duration		If set, the cluster will wait this long before running the action (for the first time, if recurring). This is an advanced option that should generally be avoided. It can be useful for a recurring monitor if a resource agent incorrectly returns success from start before the service is actually ready, and the agent can’t be corrected, or for a start action if a service takes a very long time to start, and you don’t want to block the cluster from responding to other events during that time. If this delay is longer than 5 minutes, the cluster will pretend that the action succeeded when it is first scheduled for the purpose of other actions needed, then act on the result when it actually runs. At most one of interval-origin and start-delay may be set.
record-pending	boolean	true	Operation results are always recorded when the operation completes (successful or not). If this is true, operations will also be recorded when initiated, so that status output can indicate that the operation is in progress. (deprecated since 3.0.0)

Note

Only one action can be configured for any given combination of name and interval.

Note

When on-fail is set to demote, recovery from failure by a successful demote causes the cluster to recalculate whether and where a new instance should be promoted. The node with the failure is eligible, so if promotion scores have not changed, it will be promoted again.

There is no direct equivalent of migration-threshold for the promoted role, but the same effect can be achieved with a location constraint using a rule with a node attribute expression for the resource’s fail count.

For example, to immediately ban the promoted role from a node with any failed promote or promoted instance monitor:

<rsc_location id="loc1" rsc="my_primitive">
    <rule id="rule1" score="-INFINITY" role="Promoted" boolean-op="or">
      <expression id="expr1" attribute="fail-count-my_primitive#promote_0"
        operation="gte" value="1"/>
      <expression id="expr2" attribute="fail-count-my_primitive#monitor_10000"
        operation="gte" value="1"/>
    </rule>
</rsc_location>

This example assumes that there is a promotable clone of the my_primitive resource (note that the primitive name, not the clone name, is used in the rule), and that there is a recurring 10-second-interval monitor configured for the promoted role (fail count attributes specify the interval in milliseconds).

6.2. Monitoring Resources for Failure

When Pacemaker first starts a resource, it runs one-time monitor operations (referred to as probes) to ensure the resource is running where it’s supposed to be, and not running where it’s not supposed to be. (This behavior can be affected by the resource-discovery location constraint property.)

Other than those initial probes, Pacemaker will not (by default) check that the resource continues to stay healthy [1]. You must configure monitor operations explicitly to perform these checks.

An OCF resource with a recurring health check

<primitive id="Public-IP" class="ocf" type="IPaddr" provider="heartbeat">
  <operations>
     <op id="Public-IP-start" name="start" timeout="60s"/>
     <op id="Public-IP-monitor" name="monitor" interval="60s"/>
  </operations>
  <instance_attributes id="params-public-ip">
     <nvpair id="public-ip-addr" name="ip" value="192.0.2.2"/>
  </instance_attributes>
</primitive>

By default, a monitor operation will ensure that the resource is running where it is supposed to. The target-role property can be used for further checking.

For example, if a resource has one monitor operation with interval=10 role=Started and a second monitor operation with interval=11 role=Stopped, the cluster will run the first monitor on any nodes it thinks should be running the resource, and the second monitor on any nodes that it thinks should not be running the resource (for the truly paranoid, who want to know when an administrator manually starts a service by mistake).

Note

Currently, monitors with role=Stopped are not implemented for clone resources.

6.3. Custom Recurring Operations

Typically, only monitor operations should be configured as recurring. However, it is possible to implement a custom action name in an OCF agent and then configure that as a recurring operation.

This could be useful, for example, to run a report, rotate a log, or clean temporary files related to a particular service.

Failures of custom recurring operations will be ignored by the cluster and will not be reported in cluster status (since 3.0.0; previously, they would be treated like failed monitors). A fail count and last failure timestamp will be recorded as transient node attributes, and those node attributes will be erased by the crm_resource --cleanup command.

6.4. Setting Global Defaults for Operations

You can change the global default values for operation properties in a given cluster. These are defined in an op_defaults section of the CIB’s configuration section, and can be set with crm_attribute. For example,

# crm_attribute --type op_defaults --name timeout --update 20s

would default each operation’s timeout to 20 seconds. If an operation’s definition also includes a value for timeout, then that value would be used for that operation instead.

6.5. When Implicit Operations Take a Long Time

The cluster will always perform a number of implicit operations: start, stop and a non-recurring monitor operation used at startup to check whether the resource is already active. If one of these is taking too long, then you can create an entry for them and specify a longer timeout.

An OCF resource with custom timeouts for its implicit actions

<primitive id="Public-IP" class="ocf" type="IPaddr" provider="heartbeat">
  <operations>
     <op id="public-ip-startup" name="monitor" interval="0" timeout="90s"/>
     <op id="public-ip-start" name="start" interval="0" timeout="180s"/>
     <op id="public-ip-stop" name="stop" interval="0" timeout="15min"/>
  </operations>
  <instance_attributes id="params-public-ip">
     <nvpair id="public-ip-addr" name="ip" value="192.0.2.2"/>
  </instance_attributes>
</primitive>

6.6. Multiple Monitor Operations

Provided no two operations (for a single resource) have the same name and interval, you can have as many monitor operations as you like. In this way, you can do a superficial health check every minute and progressively more intense ones at higher intervals.

To tell the resource agent what kind of check to perform, you need to provide each monitor with a different value for a common parameter. The OCF standard creates a special parameter called OCF_CHECK_LEVEL for this purpose and dictates that it is “made available to the resource agent without the normal OCF_RESKEY prefix”.

Whatever name you choose, you can specify it by adding an instance_attributes block to the op tag. It is up to each resource agent to look for the parameter and decide how to use it.

An OCF resource with two recurring health checks, performing different levels of checks specified via OCF_CHECK_LEVEL.

<primitive id="Public-IP" class="ocf" type="IPaddr" provider="heartbeat">
   <operations>
      <op id="public-ip-health-60" name="monitor" interval="60">
         <instance_attributes id="params-public-ip-depth-60">
            <nvpair id="public-ip-depth-60" name="OCF_CHECK_LEVEL" value="10"/>
         </instance_attributes>
      </op>
      <op id="public-ip-health-300" name="monitor" interval="300">
         <instance_attributes id="params-public-ip-depth-300">
            <nvpair id="public-ip-depth-300" name="OCF_CHECK_LEVEL" value="20"/>
         </instance_attributes>
     </op>
   </operations>
   <instance_attributes id="params-public-ip">
       <nvpair id="public-ip-level" name="ip" value="192.0.2.2"/>
   </instance_attributes>
</primitive>

6.7. Disabling a Monitor Operation

The easiest way to stop a recurring monitor is to just delete it. However, there can be times when you only want to disable it temporarily. In such cases, simply add enabled=false to the operation’s definition.

Example of an OCF resource with a disabled health check

<primitive id="Public-IP" class="ocf" type="IPaddr" provider="heartbeat">
   <operations>
      <op id="public-ip-check" name="monitor" interval="60s" enabled="false"/>
   </operations>
   <instance_attributes id="params-public-ip">
      <nvpair id="public-ip-addr" name="ip" value="192.0.2.2"/>
   </instance_attributes>
</primitive>

This can be achieved from the command line by executing:

# cibadmin --modify --xml-text '<op id="public-ip-check" enabled="false"/>'

Once you’ve done whatever you needed to do, you can then re-enable it with

# cibadmin --modify --xml-text '<op id="public-ip-check" enabled="true"/>'

6.8. Handling Resource Failure

By default, Pacemaker will attempt to recover failed resources by restarting them. However, failure recovery is highly configurable.

6.8.1. Failure Counts

Pacemaker tracks resource failures for each combination of node, resource, and operation (start, stop, monitor, etc.).

You can query the fail count for a particular node, resource, and/or operation using the crm_failcount command. For example, to see how many times the 10-second monitor for myrsc has failed on node1, run:

# crm_failcount --query -r myrsc -N node1 -n monitor -I 10s

If you omit the node, crm_failcount will use the local node. If you omit the operation and interval, crm_failcount will display the sum of the fail counts for all operations on the resource.

You can use crm_resource --cleanup or crm_failcount --delete to clear fail counts. For example, to clear the above monitor failures, run:

# crm_resource --cleanup -r myrsc -N node1 -n monitor -I 10s

If you omit the resource, crm_resource --cleanup will clear failures for all resources. If you omit the node, it will clear failures on all nodes. If you omit the operation and interval, it will clear the failures for all operations on the resource.

Note

Even when cleaning up only a single operation, all failed operations will disappear from the status display. This allows us to trigger a re-check of the resource’s current status.

Higher-level tools may provide other commands for querying and clearing fail counts.

The crm_mon tool shows the current cluster status, including any failed operations. To see the current fail counts for any failed resources, call crm_mon with the --failcounts option. This shows the fail counts per resource (that is, the sum of any operation fail counts for the resource).

6.8.2. Failure Response

Normally, if a running resource fails, pacemaker will try to stop it and start it again. Pacemaker will choose the best location to start it each time, which may be the same node that it failed on.

However, if a resource fails repeatedly, it is possible that there is an underlying problem on that node, and you might desire trying a different node in such a case. Pacemaker allows you to set your preference via the migration-threshold resource meta-attribute. [2]

If you define migration-threshold to N for a resource, it will be banned from the original node after N failures there.

Note

The migration-threshold is per resource, even though fail counts are tracked per operation. The operation fail counts are added together to compare against the migration-threshold.

By default, fail counts remain until manually cleared by an administrator using crm_resource --cleanup or crm_failcount --delete (hopefully after first fixing the failure’s cause). It is possible to have fail counts expire automatically by setting the failure-timeout resource meta-attribute.

Important

A successful operation does not clear past failures. If a recurring monitor operation fails once, succeeds many times, then fails again days later, its fail count is 2. Fail counts are cleared only by manual intervention or failure timeout.

For example, setting migration-threshold to 2 and failure-timeout to 60s would cause the resource to move to a new node after 2 failures, and allow it to move back (depending on stickiness and constraint scores) after one minute.

Note

failure-timeout is measured since the most recent failure. That is, older failures do not individually time out and lower the fail count. Instead, all failures are timed out simultaneously (and the fail count is reset to 0) if there is no new failure for the timeout period.

There are two exceptions to the migration threshold: when a resource either fails to start or fails to stop.

If the cluster property start-failure-is-fatal is set to true (which is the default), start failures cause the fail count to be set to INFINITY and thus always cause the resource to move immediately.

Stop failures are slightly different and crucial. If a resource fails to stop and fencing is enabled, then the cluster will fence the node in order to be able to start the resource elsewhere. If fencing is disabled, then the cluster has no way to continue and will not try to start the resource elsewhere, but will try to stop it again after any failure timeout or clearing.

6.9. Reloading an Agent After a Definition Change

The cluster automatically detects changes to the configuration of active resources. The cluster’s normal response is to stop the service (using the old definition) and start it again (with the new definition). This works, but some resource agents are smarter and can be told to use a new set of options without restarting.

To take advantage of this capability, the resource agent must:

• Implement the reload-agent action. What it should do depends completely on your application!

  Note

  Resource agents may also implement a reload action to make the managed service reload its own native configuration. This is different from reload-agent, which makes effective changes in the resource’s Pacemaker configuration (specifically, the values of the agent’s reloadable parameters).

• Advertise the reload-agent operation in the actions section of its meta-data.

• Set the reloadable attribute to 1 in the parameters section of its meta-data for any parameters eligible to be reloaded after a change.

Once these requirements are satisfied, the cluster will automatically know to reload the resource (instead of restarting) when a reloadable parameter changes.

Note

Metadata will not be re-read unless the resource needs to be started. If you edit the agent of an already active resource to set a parameter reloadable, the resource may restart the first time the parameter value changes.

Note

If both a reloadable and non-reloadable parameter are changed simultaneously, the resource will be restarted.

6.10. Migrating Resources

Normally, when the cluster needs to move a resource, it fully restarts the resource (that is, it stops the resource on the current node and starts it on the new node).

However, some types of resources, such as many virtual machines, are able to move to another location without loss of state (often referred to as live migration or hot migration). In pacemaker, this is called live migration. Pacemaker can be configured to migrate a resource when moving it, rather than restarting it.

Not all resources are able to migrate; see the migration checklist below. Even those that can, won’t do so in all situations. Conceptually, there are two requirements from which the other prerequisites follow:

• The resource must be active and healthy at the old location; and

• everything required for the resource to run must be available on both the old and new locations.

The cluster is able to accommodate both push and pull migration models by requiring the resource agent to support two special actions: migrate_to (performed on the current location) and migrate_from (performed on the destination).

In push migration, the process on the current location transfers the resource to the new location where is it later activated. In this scenario, most of the work would be done in the migrate_to action and, if anything, the activation would occur during migrate_from.

Conversely for pull, the migrate_to action is practically empty and migrate_from does most of the work, extracting the relevant resource state from the old location and activating it.

There is no wrong or right way for a resource agent to implement migration, as long as it works.

Migration Checklist

• The resource may not be a clone.

• The resource agent standard must be OCF.

• The resource must not be in a failed or degraded state.

• The resource agent must support migrate_to and migrate_from actions, and advertise them in its meta-data.

• The resource must have the allow-migrate meta-attribute set to true (which is not the default).

If an otherwise migratable resource depends on another resource via an ordering constraint, there are special situations in which it will be restarted rather than migrated.

For example, if the resource depends on a clone, and at the time the resource needs to be moved, the clone has instances that are stopping and instances that are starting, then the resource will be restarted. The scheduler is not yet able to model this situation correctly and so takes the safer (if less optimal) path.

Also, if a migratable resource depends on a non-migratable resource, and both need to be moved, the migratable resource will be restarted.

Footnotes

[1]

Currently, anyway. Automatic monitoring operations may be added in a future version of Pacemaker.

[2]

The naming of this option was perhaps unfortunate as it is easily confused with live migration, the process of moving a resource from one node to another without stopping it. Xen virtual guests are the most common example of resources that can be migrated in this manner.