Package org.apache.cassandra.service

Class AbstractWriteResponseHandler<T>

java.lang.Object

org.apache.cassandra.service.AbstractWriteResponseHandler<T>

All Implemented Interfaces:

RequestCallback<T>

Direct Known Subclasses:

BatchlogResponseHandler, DatacenterSyncWriteResponseHandler, WriteResponseHandler

public abstract class AbstractWriteResponseHandler<T>
extends java.lang.Object
implements RequestCallback<T>

Field Summary

Fields

Modifier and Type	Field	Description
protected java.lang.Runnable	callback
protected static org.slf4j.Logger	logger
protected ReplicaPlan.ForWrite	replicaPlan
protected WriteType	writeType

Constructor Summary

Constructors

Modifier	Constructor	Description
protected	AbstractWriteResponseHandler(ReplicaPlan.ForWrite replicaPlan, java.lang.Runnable callback, WriteType writeType, java.util.function.Supplier<Mutation> hintOnFailure, Dispatcher.RequestTime requestTime)

Method Summary

Modifier and Type	Method	Description
protected abstract int	ackCount()
protected int	blockFor()
protected int	candidateReplicaCount()	TODO: this method is brittle for its purpose of deciding when we should fail a query; this needs to be aware of which nodes are live/down
ConsistencyLevel	consistencyLevel()
long	currentTimeoutNanos()
void	expired()
void	get()
boolean	invokeOnFailure()	Returns true if the callback handles failure reporting - in which case the remove host will be asked to report failures to us in the event of a problem processing the request.
protected void	logResponseToIdealCLDelegate(Message<T> m)	This logs the response but doesn't do any further processing related to this write response handler on whether the CL was achieved.
void	maybeTryAdditionalReplicas(IMutation mutation, StorageProxy.WritePerformer writePerformer, java.lang.String localDC)	Cheap Quorum backup.
void	onFailure(InetAddressAndPort from, RequestFailureReason failureReason)	Called when there is an exception on the remote node or timeout happens
abstract void	onResponse(Message<T> msg)	null message means "response from local write"
void	setIdealCLResponseHandler(AbstractWriteResponseHandler handler)	Set a delegate ideal CL write response handler.
protected void	signal()
protected boolean	waitingFor(InetAddressAndPort from)

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.apache.cassandra.net.RequestCallback

trackLatencyForSnitch

Field Detail

logger

protected static final org.slf4j.Logger logger

replicaPlan

protected final ReplicaPlan.ForWrite replicaPlan

callback

protected final java.lang.Runnable callback

writeType

protected final WriteType writeType

Constructor Detail

AbstractWriteResponseHandler

protected AbstractWriteResponseHandler(ReplicaPlan.ForWrite replicaPlan,
                                       java.lang.Runnable callback,
                                       WriteType writeType,
                                       java.util.function.Supplier<Mutation> hintOnFailure,
                                       Dispatcher.RequestTime requestTime)

Parameters:

callback - A callback to be called when the write is successful.

hintOnFailure -

requestTime -

Method Detail

get

public void get()
         throws WriteTimeoutException,
                WriteFailureException

Throws:

WriteTimeoutException

WriteFailureException

currentTimeoutNanos

public final long currentTimeoutNanos()

setIdealCLResponseHandler

public void setIdealCLResponseHandler(AbstractWriteResponseHandler handler)

Set a delegate ideal CL write response handler. Note that this could be the same as this if the ideal CL and requested CL are the same.

logResponseToIdealCLDelegate

protected final void logResponseToIdealCLDelegate(Message<T> m)

This logs the response but doesn't do any further processing related to this write response handler on whether the CL was achieved. Only call this after the subclass has completed all it's processing since the subclass instance may be queried to find out if the CL was achieved.

expired

public final void expired()

blockFor

protected int blockFor()

Returns:

the minimum number of endpoints that must respond.

candidateReplicaCount

protected int candidateReplicaCount()

TODO: this method is brittle for its purpose of deciding when we should fail a query; this needs to be aware of which nodes are live/down

Returns:

the total number of endpoints the request can send to.

consistencyLevel

public ConsistencyLevel consistencyLevel()

waitingFor

protected boolean waitingFor(InetAddressAndPort from)

Returns:

true if the message counts towards the blockFor() threshold

ackCount

protected abstract int ackCount()

Returns:

number of responses received

onResponse

public abstract void onResponse(Message<T> msg)

null message means "response from local write"

Specified by:

onResponse in interface RequestCallback<T>

Parameters:

msg - response received.

signal

protected void signal()

onFailure

public void onFailure(InetAddressAndPort from,
                      RequestFailureReason failureReason)

Description copied from interface: RequestCallback

Called when there is an exception on the remote node or timeout happens

Specified by:

onFailure in interface RequestCallback<T>

invokeOnFailure

public boolean invokeOnFailure()

Description copied from interface: RequestCallback

Returns true if the callback handles failure reporting - in which case the remove host will be asked to report failures to us in the event of a problem processing the request. TODO: this is an error prone method, and we should be handling failures everywhere so we should probably just start doing that, and remove this method

Specified by:

invokeOnFailure in interface RequestCallback<T>

Returns:

true if the callback should be invoked on failure

maybeTryAdditionalReplicas

public void maybeTryAdditionalReplicas(IMutation mutation,
                                       StorageProxy.WritePerformer writePerformer,
                                       java.lang.String localDC)

Cheap Quorum backup. If we failed to reach quorum with our initial (full) nodes, reach out to other nodes.