Chapter 13. Determining the order in which cluster resources are run

download PDF

To determine the order in which the resources run, you configure an ordering constraint.

The following shows the format for the command to configure an ordering constraint.

pcs constraint order [action] resource_id then [action] resource_id [options]

The following table summarizes the properties and options for configuring ordering constraints.

Table 13.1. Properties of an Order Constraint


The name of a resource on which an action is performed.


The action to be ordered on the resource. Possible values of the action property are as follows:

* start - Order start actions of the resource.

* stop - Order stop actions of the resource.

* promote - Promote the resource from an unpromoted resource to a promoted resource.

* demote - Demote the resource from a promoted resource to an unpromoted resource.

If no action is specified, the default action is start.

kind option

How to enforce the constraint. The possible values of the kind option are as follows:

* Optional - Only applies if both resources are executing the specified action. For information about optional ordering, see Configuring advisory ordering.

* Mandatory - Always enforce the constraint (default value). If the first resource you specified is stopping or cannot be started, the second resource you specified must be stopped. For information about mandatory ordering, see Configuring mandatory ordering.

* Serialize - Ensure that no two stop/start actions occur concurrently for the resources you specify. The first and second resource you specify can start in either order, but one must complete starting before the other can be started. A typical use case is when resource startup puts a high load on the host.

symmetrical option

If true, the reverse of the constraint applies for the opposite action (for example, if B starts after A starts, then B stops before A stops). Ordering constraints for which kind is Serialize cannot be symmetrical. The default value is true for Mandatory and Optional kinds, false for Serialize.

Use the following command to remove resources from any ordering constraint.

pcs constraint order remove resource1 [resourceN]...

13.1. Configuring mandatory ordering

A mandatory ordering constraint indicates that the second action should not be initiated for the second resource unless and until the first action successfully completes for the first resource. Actions that may be ordered are stop, start, and additionally for promotable clones, demote and promote. For example, "A then B" (which is equivalent to "start A then start B") means that B will not be started unless and until A successfully starts. An ordering constraint is mandatory if the kind option for the constraint is set to Mandatory or left as default.

If the symmetrical option is set to true or left to default, the opposite actions will be ordered in reverse. The start and stop actions are opposites, and demote and promote are opposites. For example, a symmetrical "promote A then start B" ordering implies "stop B then demote A", which means that A cannot be demoted until and unless B successfully stops. A symmetrical ordering means that changes in A’s state can cause actions to be scheduled for B. For example, given "A then B", if A restarts due to failure, B will be stopped first, then A will be stopped, then A will be started, then B will be started.

Note that the cluster reacts to each state change. If the first resource is restarted and is in a started state again before the second resource initiated a stop operation, the second resource will not need to be restarted.

13.2. Configuring advisory ordering

When the kind=Optional option is specified for an ordering constraint, the constraint is considered optional and only applies if both resources are executing the specified actions. Any change in state by the first resource you specify will have no effect on the second resource you specify.

The following command configures an advisory ordering constraint for the resources named VirtualIP and dummy_resource.

# pcs constraint order VirtualIP then dummy_resource kind=Optional

13.3. Configuring ordered resource sets

A common situation is for an administrator to create a chain of ordered resources, where, for example, resource A starts before resource B which starts before resource C. If your configuration requires that you create a set of resources that is colocated and started in order, you can configure a resource group that contains those resources.

There are some situations, however, where configuring the resources that need to start in a specified order as a resource group is not appropriate:

  • You may need to configure resources to start in order and the resources are not necessarily colocated.
  • You may have a resource C that must start after either resource A or B has started but there is no relationship between A and B.
  • You may have resources C and D that must start after both resources A and B have started, but there is no relationship between A and B or between C and D.

In these situations, you can create an ordering constraint on a set or sets of resources with the pcs constraint order set command.

You can set the following options for a set of resources with the pcs constraint order set command.

  • sequential, which can be set to true or false to indicate whether the set of resources must be ordered relative to each other. The default value is true.

    Setting sequential to false allows a set to be ordered relative to other sets in the ordering constraint, without its members being ordered relative to each other. Therefore, this option makes sense only if multiple sets are listed in the constraint; otherwise, the constraint has no effect.

  • require-all, which can be set to true or false to indicate whether all of the resources in the set must be active before continuing. Setting require-all to false means that only one resource in the set needs to be started before continuing on to the next set. Setting require-all to false has no effect unless used in conjunction with unordered sets, which are sets for which sequential is set to false. The default value is true.
  • action, which can be set to start, promote, demote or stop, as described in the "Properties of an Order Constraint" table in Determining the order in which cluster resources are run.
  • role, which can be set to Stopped, Started, Promoted, or Unpromoted.

You can set the following constraint options for a set of resources following the setoptions parameter of the pcs constraint order set command.

pcs constraint order set resource1 resource2 [resourceN]... [options] [set resourceX resourceY ... [options]] [setoptions [constraint_options]]

If you have three resources named D1, D2, and D3, the following command configures them as an ordered resource set.

# pcs constraint order set D1 D2 D3

If you have six resources named A, B, C, D, E, and F, this example configures an ordering constraint for the set of resources that will start as follows:

  • A and B start independently of each other
  • C starts once either A or B has started
  • D starts once C has started
  • E and F start independently of each other once D has started

Stopping the resources is not influenced by this constraint since symmetrical=false is set.

# pcs constraint order set A B sequential=false require-all=false set C D set E F sequential=false setoptions symmetrical=false

13.4. Configuring startup order for resource dependencies not managed by Pacemaker

It is possible for a cluster to include resources with dependencies that are not themselves managed by the cluster. In this case, you must ensure that those dependencies are started before Pacemaker is started and stopped after Pacemaker is stopped.

You can configure your startup order to account for this situation by means of the systemd resource-agents-deps target. You can create a systemd drop-in unit for this target and Pacemaker will order itself appropriately relative to this target.

For example, if a cluster includes a resource that depends on the external service foo that is not managed by the cluster, perform the following procedure.

  1. Create the drop-in unit /etc/systemd/system/ that contains the following:

  2. Run the systemctl daemon-reload command.

A cluster dependency specified in this way can be something other than a service. For example, you may have a dependency on mounting a file system at /srv, in which case you would perform the following procedure:

  1. Ensure that /srv is listed in the /etc/fstab file. This will be converted automatically to the systemd file srv.mount at boot when the configuration of the system manager is reloaded. For more information, see the systemd.mount(5) and the systemd-fstab-generator(8) man pages.
  2. To make sure that Pacemaker starts after the disk is mounted, create the drop-in unit /etc/systemd/system/ that contains the following.

  3. Run the systemctl daemon-reload command.

If an LVM volume group used by a Pacemaker cluster contains one or more physical volumes that reside on remote block storage, such as an iSCSI target, you can configure a systemd resource-agents-deps target and a systemd drop-in unit for the target to ensure that the service starts before Pacemaker starts.

The following procedure configures blk-availability.service as a dependency. The blk-availability.service service is a wrapper that includes iscsi.service, among other services. If your deployment requires it, you could configure iscsi.service (for iSCSI only) or as the dependency instead of blk-availability.

  1. Create the drop-in unit /etc/systemd/system/ that contains the following:

  2. Run the systemctl daemon-reload command.
Red Hat logoGithubRedditYoutubeTwitter


Try, buy, & sell


About Red Hat Documentation

We help Red Hat users innovate and achieve their goals with our products and services with content they can trust.

Making open source more inclusive

Red Hat is committed to replacing problematic language in our code, documentation, and web properties. For more details, see the Red Hat Blog.

About Red Hat

We deliver hardened solutions that make it easier for enterprises to work across platforms and environments, from the core datacenter to the network edge.

© 2024 Red Hat, Inc.