Package org.apache.sling.discovery.oak

Class SlingIdCleanupTask

java.lang.Object
org.apache.sling.discovery.oak.SlingIdCleanupTask
All Implemented Interfaces:
Runnable, TopologyEventListener
public class SlingIdCleanupTask extends Object implements TopologyEventListener, Runnable
A background task that cleans up garbage slingIds after topology changes.

A slingId is considered garbage when:

  • it is not in the current topology
  • was not ever seen in previous topologies by the now leader instance
  • it is not in the current idmap (where clusterNodeIds are reused hence that list stays small and does not need cleanup)
  • its leaderElectionId was created more than 7 days ago (the leaderElectionId is created at activate time of the discovery.oak bundle - hence this more or less corresponds to the startup time of that instance)
The garbage accumulates at the following places, where it will thus be cleaned up from:
  • as child node under /var/discovery/oak/clusterInstances : this is the most performance critical garbage
  • as a property key in /var/discovery/oak/syncTokens
The task by default is executed:
  • only on the leader
  • 10min after a TOPOLOGY_INIT or TOPOLOGY_CHANGED event
  • with a maximum number of delete operations to avoid repository overload - that maximum is called batchSize and is 50 by default
  • in subsequent intervals of 10min after the initial run, if that had to stop at the batchSize of 50 deletions
All parameters mentioned above can be configured.

Additionally, the cleanup is skipped for 13 hours after a successful cleanup. This is to avoid unnecessary load on the repository. The number of 13 incorporates some heuristics such as : about 2 cleanup rounds per day maximum makes sense, if a leader is very long living, then the 1 additional hour makes it spread somewhat throughout the day. This is to further minimize any load side-effects.

  • Field Details

    • logger

      protected final org.slf4j.Logger logger

    • scheduler

      protected Scheduler scheduler

    • resourceResolverFactory

      protected ResourceResolverFactory resourceResolverFactory

  • Constructor Details

    • SlingIdCleanupTask

      public SlingIdCleanupTask()

  • Method Details

    • activate

      protected void activate(org.osgi.framework.BundleContext bc, SlingIdCleanupTask.Conf config)

    • modified

      protected void modified(org.osgi.framework.BundleContext bc, SlingIdCleanupTask.Conf config)

    • deactivate

      protected void deactivate()

    • handleTopologyEvent

      public void handleTopologyEvent(TopologyEvent event)
      Description copied from interface: TopologyEventListener
      Inform the service about an event in the topology - or in the discovery of the topology.

      Implementors of this method must ensure that this method returns quickly and that no locks are being acquired that could cause deadlocks (ie the framework might call this method holding locks).

      The TopologyEvent contains details about what changed. The supported event types are:

      • TOPOLOGY_INIT sent when the TopologyEventListener was first bound to the discovery service - represents the initial state of the topology at that time.
      • TOPOLOGY_CHANGING sent when the discovery service discovered a change in the topology and has started to settle the change. This event is sent before TOPOLOGY_CHANGED but is optional
      • TOPOLOGY_CHANGED sent when the discovery service discovered a change in the topology and has settled it.
      • PROPERTIES_CHANGED sent when the one or many properties have changed in an instance in the current topology
      A note on instance restarts: it is currently not a requirement on the discovery service to send a TopologyEvent should an instance restart occur rapidly (ie within the change detection timeout). A TopologyEvent is only sent if the number of instances or any property changes. Should there be a requirement to detect a restart in a guaranteed fashion, it is always possible to set a particular property (using the PropertyProvider) to the instance start time and have others detect a change in that property.
      Specified by:
      handleTopologyEvent in interface TopologyEventListener
      Parameters:
      event - The topology event

    • run

      public void run()
      Invoked via sling.commons.scheduler triggered from resetCleanupSchedule(). By default should get called at max every 5 minutes until cleanup is done or 10min after a topology change.
      Specified by:
      run in interface Runnable