ai.grakn.engine.backgroundtasks.distributed

Class SchedulerElector

java.lang.Object

org.apache.curator.framework.recipes.leader.LeaderSelectorListenerAdapter

ai.grakn.engine.backgroundtasks.distributed.SchedulerElector

All Implemented Interfaces:

org.apache.curator.framework.recipes.leader.LeaderSelectorListener, org.apache.curator.framework.state.ConnectionStateListener

public class SchedulerElector
extends org.apache.curator.framework.recipes.leader.LeaderSelectorListenerAdapter

There is one "Scheduler" that will be constantly running on the "Leader" machine. This class registers this instance of Engine with Zookeeper & the Leader election process. If this machine is required to take over the Scheduler, the "takeLeadership" function will be called. If this machine controls the Scheduler and leadership is relinquished/lost, the "stateChanged" function is called and the Scheduler is stopped before a error is thrown. The Scheduler thread is a daemon thread

Author:

Denis Lobanov, alexandraorth

Constructor Summary

Constructors

Constructor and Description
SchedulerElector(TaskStateStorage storage, ZookeeperConnection zookeeperConnection)

Method Summary

Modifier and Type	Method and Description
Scheduler	getScheduler()
void	stateChanged(org.apache.curator.framework.CuratorFramework client, org.apache.curator.framework.state.ConnectionState newState)
void	stop() When stopping the ClusterManager we must: 1.
void	takeLeadership(org.apache.curator.framework.CuratorFramework client) On leadership takeover, start a new Scheduler instance and wait for it to complete.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

SchedulerElector

public SchedulerElector(TaskStateStorage storage,
                        ZookeeperConnection zookeeperConnection)

Method Detail

stop

public void stop()

When stopping the ClusterManager we must: 1. Interrupt the Scheduler on whichever machine it is running on. We do this by interrupting the leadership. When leadership is interrupted it will shut down the Scheduler on that machine. 2. If the scheduler is running on this machine, close it noThrow() functions used here so that if an error occurs during execution of a certain step, the subsequent stops continue to execute.

stateChanged

public void stateChanged(org.apache.curator.framework.CuratorFramework client,
                         org.apache.curator.framework.state.ConnectionState newState)

Specified by:

stateChanged in interface org.apache.curator.framework.state.ConnectionStateListener

Overrides:

stateChanged in class org.apache.curator.framework.recipes.leader.LeaderSelectorListenerAdapter

takeLeadership

public void takeLeadership(org.apache.curator.framework.CuratorFramework client)
                    throws Exception

On leadership takeover, start a new Scheduler instance and wait for it to complete. The method should not return until leadership is relinquished.

Throws:

Exception

getScheduler

public Scheduler getScheduler()