org.apache.solr.cloud.overseer

Class ZkStateWriter

java.lang.Object

org.apache.solr.cloud.overseer.ZkStateWriter

public class ZkStateWriter
extends Object

ZkStateWriter is responsible for writing updates to the cluster state stored in ZooKeeper for both stateFormat=1 collection (stored in shared /clusterstate.json in ZK) and stateFormat=2 collections each of which get their own individual state.json in ZK. Updates to the cluster state are specified using the enqueueUpdate(ClusterState, ZkWriteCommand, ZkWriteCallback) method. The class buffers updates to reduce the number of writes to ZK. The buffered updates are flushed during enqueueUpdate automatically if necessary. The writePendingUpdates() can be used to force flush any pending updates. If either enqueueUpdate(ClusterState, ZkWriteCommand, ZkWriteCallback) or writePendingUpdates() throws a KeeperException.BadVersionException then the internal buffered state of the class is suspect and the current instance of the class should be discarded and a new instance should be created and used for any future updates.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static interface	ZkStateWriter.ZkWriteCallback

Field Summary

Fields

Modifier and Type	Field and Description
protected ClusterState	clusterState
protected boolean	invalidState Set to true if we ever get a BadVersionException so that we can disallow future operations with this instance
protected boolean	isClusterStateModified
protected String	lastCollectionName
protected int	lastStateFormat
protected long	lastUpdatedTime
static ZkWriteCommand	NO_OP Represents a no-op ZkWriteCommand which will result in no modification to cluster state
protected ZkStateReader	reader
protected Overseer.Stats	stats
protected Map<String,DocCollection>	updates

Constructor Summary

Constructors

Constructor and Description
ZkStateWriter(ZkStateReader zkStateReader, Overseer.Stats stats)

Method Summary

Modifier and Type	Method and Description
ClusterState	enqueueUpdate(ClusterState prevState, ZkWriteCommand cmd, ZkStateWriter.ZkWriteCallback callback) Applies the given ZkWriteCommand on the prevState.
ClusterState	getClusterState()
long	getLastUpdatedTime()
boolean	hasPendingUpdates()
protected boolean	maybeFlushAfter(ZkWriteCommand cmd) Logic to decide a flush after processing a ZkWriteCommand
protected boolean	maybeFlushBefore(ZkWriteCommand cmd) Logic to decide a flush before processing a ZkWriteCommand
ClusterState	writePendingUpdates() Writes all pending updates to ZooKeeper and returns the modified cluster state

Methods inherited from class java.lang.Object

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

Field Detail

NO_OP

public static ZkWriteCommand NO_OP

Represents a no-op ZkWriteCommand which will result in no modification to cluster state

reader

protected final ZkStateReader reader

stats

protected final Overseer.Stats stats

updates

protected Map<String,DocCollection> updates

clusterState

protected ClusterState clusterState

isClusterStateModified

protected boolean isClusterStateModified

lastUpdatedTime

protected long lastUpdatedTime

lastStateFormat

protected int lastStateFormat

lastCollectionName

protected String lastCollectionName

invalidState

protected boolean invalidState

Set to true if we ever get a BadVersionException so that we can disallow future operations with this instance

Constructor Detail

ZkStateWriter

public ZkStateWriter(ZkStateReader zkStateReader,
                     Overseer.Stats stats)

Method Detail

enqueueUpdate

public ClusterState enqueueUpdate(ClusterState prevState,
                                  ZkWriteCommand cmd,
                                  ZkStateWriter.ZkWriteCallback callback)
                           throws IllegalStateException,
                                  Exception

Applies the given ZkWriteCommand on the prevState. The modified ClusterState is returned and it is expected that the caller will use the returned cluster state for the subsequent invocation of this method.

The modified state may be buffered or flushed to ZooKeeper depending on the internal buffering logic of this class. The hasPendingUpdates() method may be used to determine if the last enqueue operation resulted in buffered state. The method writePendingUpdates() can be used to force an immediate flush of pending cluster state changes.

Parameters:

prevState - the cluster state information on which the given cmd is applied

cmd - the ZkWriteCommand which specifies the change to be applied to cluster state

callback - a ZkStateWriter.ZkWriteCallback object to be used for any callbacks

Returns:

modified cluster state created after applying cmd to prevState. If cmd is a no-op (NO_OP) then the prevState is returned unmodified.

Throws:

IllegalStateException - if the current instance is no longer usable. The current instance must be discarded.

Exception - on an error in ZK operations or callback. If a flush to ZooKeeper results in a KeeperException.BadVersionException this instance becomes unusable and must be discarded

maybeFlushBefore

protected boolean maybeFlushBefore(ZkWriteCommand cmd)

Logic to decide a flush before processing a ZkWriteCommand

Parameters:

cmd - the ZkWriteCommand instance

Returns:

true if a flush is required, false otherwise

maybeFlushAfter

protected boolean maybeFlushAfter(ZkWriteCommand cmd)

Logic to decide a flush after processing a ZkWriteCommand

Parameters:

cmd - the ZkWriteCommand instance

Returns:

true if a flush to ZK is required, false otherwise

hasPendingUpdates

public boolean hasPendingUpdates()

writePendingUpdates

public ClusterState writePendingUpdates()
                                 throws IllegalStateException,
                                        org.apache.zookeeper.KeeperException,
                                        InterruptedException

Writes all pending updates to ZooKeeper and returns the modified cluster state

Returns:

the modified cluster state

Throws:

IllegalStateException - if the current instance is no longer usable and must be discarded

org.apache.zookeeper.KeeperException - if any ZooKeeper operation results in an error

InterruptedException - if the current thread is interrupted

getLastUpdatedTime

public long getLastUpdatedTime()

Returns:

time returned by System.nanoTime at which the main cluster state was last written to ZK or 0 if never

getClusterState

public ClusterState getClusterState()

Returns:

the most up-to-date cluster state until the last enqueueUpdate operation