org.apache.hadoop.hbase.regionserver
Class SplitLogWorker

java.lang.Object
  org.apache.hadoop.hbase.zookeeper.ZooKeeperListener
      org.apache.hadoop.hbase.regionserver.SplitLogWorker

All Implemented Interfaces:

Runnable

@InterfaceAudience.Private
public class SplitLogWorker
extends ZooKeeperListener
implements Runnable

This worker is spawned in every regionserver (should we also spawn one in the master?). The Worker waits for log splitting tasks to be put up by the SplitLogManager running in the master and races with other workers in other serves to acquire those tasks. The coordination is done via zookeeper. All the action takes place at /hbase/splitlog znode.

If a worker has successfully moved the task from state UNASSIGNED to OWNED then it owns the task. It keeps heart beating the manager by periodically moving the task from UNASSIGNED to OWNED state. On success it moves the task to TASK_DONE. On unrecoverable error it moves task state to ERR. If it cannot continue but wants the master to retry the task then it moves the task state to RESIGNED.

The manager can take a task away from a worker by moving the task from OWNED to UNASSIGNED. In the absence of a global lock there is a unavoidable race here - a worker might have just finished its task when it is stripped of its ownership. Here we rely on the idempotency of the log splitting task for correctness

Nested Class Summary

static interface

SplitLogWorker.TaskExecutor Objects implementing this interface actually do the task that has been acquired by a SplitLogWorker.

Field Summary

static int	DEFAULT_MAX_SPLITTERS
protected AtomicInteger	tasksInProgress

Fields inherited from class org.apache.hadoop.hbase.zookeeper.ZooKeeperListener

watcher

Constructor Summary

SplitLogWorker(ZooKeeperWatcher watcher, org.apache.hadoop.conf.Configuration conf, RegionServerServices server, LastSequenceId sequenceIdChecker)

SplitLogWorker(ZooKeeperWatcher watcher, org.apache.hadoop.conf.Configuration conf, RegionServerServices server, SplitLogWorker.TaskExecutor splitTaskExecutor)

Method Summary

protected static int	attemptToOwnTask(boolean isFirstTime, ZooKeeperWatcher zkw, ServerName server, String task, org.apache.hadoop.hbase.protobuf.generated.ZooKeeperProtos.SplitLogTask.RecoveryMode mode, int taskZKVersion) Try to own the task by transitioning the zk node data from UNASSIGNED to OWNED.
void	nodeChildrenChanged(String path)
void	nodeDataChanged(String path)
void	run()
void	start() start the SplitLogWorker thread
void	stop() stop the SplitLogWorker thread

Methods inherited from class org.apache.hadoop.hbase.zookeeper.ZooKeeperListener

getWatcher, nodeCreated, nodeDeleted

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_MAX_SPLITTERS

public static final int DEFAULT_MAX_SPLITTERS

See Also:

Constant Field Values

tasksInProgress

protected final AtomicInteger tasksInProgress

Constructor Detail

SplitLogWorker

public SplitLogWorker(ZooKeeperWatcher watcher,
                      org.apache.hadoop.conf.Configuration conf,
                      RegionServerServices server,
                      SplitLogWorker.TaskExecutor splitTaskExecutor)

SplitLogWorker

public SplitLogWorker(ZooKeeperWatcher watcher,
                      org.apache.hadoop.conf.Configuration conf,
                      RegionServerServices server,
                      LastSequenceId sequenceIdChecker)

Method Detail

run

public void run()

Specified by:

run in interface Runnable

attemptToOwnTask

protected static int attemptToOwnTask(boolean isFirstTime,
                                      ZooKeeperWatcher zkw,
                                      ServerName server,
                                      String task,
                                      org.apache.hadoop.hbase.protobuf.generated.ZooKeeperProtos.SplitLogTask.RecoveryMode mode,
                                      int taskZKVersion)

Try to own the task by transitioning the zk node data from UNASSIGNED to OWNED.

This method is also used to periodically heartbeat the task progress by transitioning the node from OWNED to OWNED.

Parameters:

isFirstTime -

zkw -

server -

task -

taskZKVersion -

Returns:

non-negative integer value when task can be owned by current region server otherwise -1

nodeDataChanged

public void nodeDataChanged(String path)

Overrides:

nodeDataChanged in class ZooKeeperListener

nodeChildrenChanged

public void nodeChildrenChanged(String path)

Overrides:

nodeChildrenChanged in class ZooKeeperListener

start

public void start()

start the SplitLogWorker thread

stop

public void stop()

stop the SplitLogWorker thread