Class DelayedAllocationService
- All Implemented Interfaces:
Closeable
,AutoCloseable
,ClusterStateListener
,LifecycleComponent
,Releasable
DelayedAllocationService
listens to cluster state changes and checks
if there are unassigned shards with delayed allocation (unassigned shards that have
the delay marker). These are shards that have become unassigned due to a node leaving
and which were assigned the delay marker based on the index delay setting
UnassignedInfo.INDEX_DELAYED_NODE_LEFT_TIMEOUT_SETTING
(see AllocationService.disassociateDeadNodes(RoutingAllocation)
.
This class is responsible for choosing the next (closest) delay expiration of a
delayed shard to schedule a reroute to remove the delay marker.
The actual removal of the delay marker happens in
AllocationService.RerouteStrategy.removeDelayMarkers(RoutingAllocation)
, triggering yet
another cluster change event.-
Field Summary
Fields inherited from class org.elasticsearch.common.component.AbstractLifecycleComponent
lifecycle
-
Constructor Summary
ConstructorsConstructorDescriptionDelayedAllocationService
(ThreadPool threadPool, ClusterService clusterService, AllocationService allocationService) -
Method Summary
Modifier and TypeMethodDescriptionprotected void
void
Called when cluster state changes.protected long
override this to control time based decisions during delayed allocationprotected void
doClose()
Close this component.protected void
doStart()
Start this component.protected void
doStop()
Stop this component.Methods inherited from class org.elasticsearch.common.component.AbstractLifecycleComponent
addLifecycleListener, close, lifecycleState, start, stop
-
Constructor Details
-
DelayedAllocationService
@Inject public DelayedAllocationService(ThreadPool threadPool, ClusterService clusterService, AllocationService allocationService)
-
-
Method Details
-
doStart
protected void doStart()Description copied from class:AbstractLifecycleComponent
Start this component. Typically that means doing things like launching background processes and registering listeners on other components. Other components have been initialized by this point, but may not yet be started.If this method throws an exception then the startup process will fail, but this component will not be stopped before it is closed.
This method is called while synchronized on
AbstractLifecycleComponent.lifecycle
. It is only called once in the lifetime of a component, although it may not be called at all if the startup process encountered some kind of fatal error, such as the failure of some other component to initialize or start.- Specified by:
doStart
in classAbstractLifecycleComponent
-
doStop
protected void doStop()Description copied from class:AbstractLifecycleComponent
Stop this component. Typically that means doing the reverse of whateverAbstractLifecycleComponent.doStart()
does.This method is called while synchronized on
AbstractLifecycleComponent.lifecycle
. It is only called once in the lifetime of a component, after callingAbstractLifecycleComponent.doStart()
, although it will not be called at all if this component did not successfully start.- Specified by:
doStop
in classAbstractLifecycleComponent
-
doClose
protected void doClose()Description copied from class:AbstractLifecycleComponent
Close this component. Typically that means doing the reverse of whatever happened during initialization, such as releasing resources acquired there.This method is called while synchronized on
AbstractLifecycleComponent.lifecycle
. It is called once in the lifetime of a component. If the component was started then it will be stopped before it is closed, and once it is closed it will not be started or stopped.- Specified by:
doClose
in classAbstractLifecycleComponent
-
currentNanoTime
protected long currentNanoTime()override this to control time based decisions during delayed allocation -
clusterChanged
Description copied from interface:ClusterStateListener
Called when cluster state changes.Cluster states are applied one-by-one which means they can be a performance bottleneck. Implementations of this method should therefore be fast, so please consider forking work into the background rather than doing everything inline.
- Specified by:
clusterChanged
in interfaceClusterStateListener
-
assertClusterOrMasterStateThread
protected void assertClusterOrMasterStateThread()
-