High availability (with corosync/pacemaker)

Configuration

The inventory may define these hosts:

  • observers: Set of hosts to observe the cluster (only the first is considering)
  • Hypervisors: Set of machines to hosts VMs.

Remember that the cluster must contain an odd number of machines. For example, three hypervisors or one observer and two hypervisors.

Node redundancy

All nodes in the cluster have access to a shared storage via Ceph (see Shared storage section). With it, the cluster is in N to N redundancy mode.

  • Corosync will provides messaging and membership services.
  • Pacemaker will manage the cluster (synchronize resources between each node).

More details on Pacemaker here and Corosync here.

pacemaker-remote

pacemaker-remote is a component which can be installed inside a VM to allow Pacemaker to manage and monitor resources inside this VM.

For instance, with pacemaker-remote pacemaker can monitor services and containers directly inside a VM.

For more information about pacemaker-remote refer to https://clusterlabs.org/pacemaker/doc/2.1/Pacemaker_Remote/singlehtml/.

Management tool

The vm_manager project is an high-level interface of pacemaker and Ceph to manage the VM like a resource. He is installed during the installation step and provides the vm-mgr command.

Sub-command

All sub-commands has -n, --name required option to specify which resource should be used.

  • add_colocation: Adds a colocation constraint between resources

  • clone: Creates a copy of the VM

  • create: Generates a new resource from a VM

  • create_snapshot: Creates a snapshot of a resource

  • disable: Stops and removes the resource on the cluster

  • enable: Adds and starts the resource on the cluster

  • get_metadata: Gets a metadata of a resource

  • list: Lists all resources

  • list_metadata: Lists all keys of a resource

  • list_snapshots: Lists all created snapshots

  • purge: Deletes all snapshots of a resource

  • remove: Removes the resource

  • remove_snapshot: Removes a snapshot of a resource

  • rollback: Rollbacks to a snapshot for a resource

  • set_metadata: Sets a metadata of a resource

  • start: Start a resource

  • status: Gets the status of a resource

  • stop: Stops a resource

Resources status

  • Undefined:
  • Disabled:
  • Failed:
  • Started:
  • Starting:
  • Stopping:
  • Stopped: