Chapter 11. Determining which nodes a resource can run on

You can configure a location constraint to control which nodes a cluster resource can run on in a cluster. You can use a location constraint to make a resource prefer a specific node or to prevent it from running on certain nodes.

Before Pacemaker starts a resource anywhere, it first runs a one-time monitor operation (often referred to as a "probe") on every node, to learn whether the resource is already running. This process of resource discovery can result in errors on nodes that are unable to execute the monitor.

When configuring a location constraint on a node, you can use the resource-discovery option of the pcs constraint location command to indicate a preference for whether Pacemaker should perform resource discovery on this node for the specified resource. Limiting resource discovery to a subset of nodes the resource is physically capable of running on can significantly boost performance when a large set of nodes is present. When pacemaker_remote is in use to expand the node count into the hundreds of nodes range, this option should be considered.

The following command shows the format for specifying the resource-discovery option of the pcs constraint location command. In this command, a positive value for score corresponds to a basic location constraint that configures a resource to prefer a node, while a negative value for score corresponds to a basic location`constraint that configures a resource to avoid a node. As with basic location constraints, you can use regular expressions for resources with these constraints as well.

# pcs constraint location add id rsc node score [resource-discovery=option]

The following table summarizes the meanings of the basic parameters for configuring constraints for resource discovery.

When using location constraints, you can configure a general strategy for specifying which nodes a resource can run on.

Whether you should choose to configure your cluster as an opt-in or opt-out cluster depends on both your personal preference and the make-up of your cluster. If most of your resources can run on most of the nodes, then an opt-out arrangement is likely to result in a simpler configuration. On the other hand, if most resources can only run on a small subset of nodes an opt-in configuration might be simpler.

Configuring an "Opt-In" cluster

To create an opt-in cluster, set the symmetric-cluster cluster property to false to prevent resources from running anywhere by default.

# pcs property set symmetric-cluster=false

Enable nodes for individual resources. The following commands configure location constraints so that the resource Webserver prefers node example-1, the resource Database prefers node example-2, and both resources can fail over to node example-3 if their preferred node fails. When configuring location constraints for an opt-in cluster, setting a score of zero allows a resource to run on a node without indicating any preference to prefer or avoid the node.

# pcs constraint location Webserver prefers example-1=200
# pcs constraint location Webserver prefers example-3=0
# pcs constraint location Database prefers example-2=200
# pcs constraint location Database prefers example-3=0

Configuring an "Opt-Out" cluster

To create an opt-out cluster, set the symmetric-cluster cluster property to true to allow resources to run everywhere by default. This is the default configuration if symmetric-cluster is not set explicitly.

# pcs property set symmetric-cluster=true

The following commands will then yield a configuration that is equivalent to the example in "Configuring an "Opt-In" cluster". Both resources can fail over to node example-3 if their preferred node fails, since every node has an implicit score of 0.

# pcs constraint location Webserver prefers example-1=200
# pcs constraint location Webserver avoids example-2=INFINITY
# pcs constraint location Database avoids example-1=INFINITY
# pcs constraint location Database prefers example-2=200

Note that it is not necessary to specify a score of INFINITY in these commands, since that is the default value for the score.

Configure resource-stickiness to define a resource’s preference for remaining on its current node. Pacemaker compares this value against other scores, such as location constraints, to prevent unnecessary migrations. For setup details, see Configuring resource meta options.

With a resource-stickiness value of 0, a cluster may move resources as needed to balance resources across nodes. This may result in resources moving when unrelated resources start or stop. With a positive stickiness, resources have a preference to stay where they are, and move only if other circumstances outweigh the stickiness. This may result in newly-added nodes not getting any resources assigned to them without administrator intervention.

Newly-created clusters set the default value for resource-stickiness to 1. This small value can easily be overridden by other constraints that you create, but it is enough to prevent Pacemaker from needlessly moving healthy resources around the cluster. If you prefer cluster behavior that results from a resource-stickiness value of 0, you can change the resource-stickiness default value to 0 with the following command:

# pcs resource defaults update resource-stickiness=0

With a positive resource-stickiness value, no resources will move to a newly-added node. If resource balancing is desired at that point, you can temporarily set the resource-stickiness value to 0.

Note that if a location constraint score is higher than the resource-stickiness value, the cluster may still move a healthy resource to the node where the location constraint points.