//
// ========================================================================
// Copyright (c) 1995-2017 Mort Bay Consulting Pty. Ltd.
// ------------------------------------------------------------------------
// All rights reserved. This program and the accompanying materials
// are made available under the terms of the Eclipse Public License v1.0
// and Apache License v2.0 which accompanies this distribution.
//
// The Eclipse Public License is available at
// http://www.eclipse.org/legal/epl-v10.html
//
// The Apache License v2.0 is available at
// http://www.opensource.org/licenses/apache2.0.php
//
// You may elect to redistribute this code under either of these licenses.
// ========================================================================
//
package org.eclipse.jetty.util;
import java.nio.channels.ClosedChannelException;
import org.eclipse.jetty.util.thread.Locker;
/**
* This specialized callback implements a pattern that allows
* a large job to be broken into smaller tasks using iteration
* rather than recursion.
* <p>
* A typical example is the write of a large content to a socket,
* divided in chunks. Chunk C1 is written by thread T1, which
* also invokes the callback, which writes chunk C2, which invokes
* the callback again, which writes chunk C3, and so forth.
* </p>
* <p>
* The problem with the example is that if the callback thread
* is the same that performs the I/O operation, then the process
* is recursive and may result in a stack overflow.
* To avoid the stack overflow, a thread dispatch must be performed,
* causing context switching and cache misses, affecting performance.
* </p>
* <p>
* To avoid this issue, this callback uses an AtomicReference to
* record whether success callback has been called during the processing
* of a sub task, and if so then the processing iterates rather than
* recurring.
* </p>
* <p>
* Subclasses must implement method {@link #process()} where the sub
* task is executed and a suitable {@link IteratingCallback.Action} is
* returned to this callback to indicate the overall progress of the job.
* This callback is passed to the asynchronous execution of each sub
* task and a call the {@link #succeeded()} on this callback represents
* the completion of the sub task.
* </p>
*/
public abstract class IteratingCallback implements Callback
{
/**
* The internal states of this callback
*/
private enum State
{
/**
* This callback is IDLE, ready to iterate.
*/
IDLE,
/**
* This callback is iterating calls to {@link #process()} and is dealing with
* the returns. To get into processing state, it much of held the lock state
* and set iterating to true.
*/
PROCESSING,
/**
* Waiting for a schedule callback
*/
PENDING,
/**
* Called by a schedule callback
*/
CALLED,
/**
* The overall job has succeeded as indicated by a {@link Action#SUCCEEDED} return
* from {@link IteratingCallback#process()}
*/
SUCCEEDED,
/**
* The overall job has failed as indicated by a call to {@link IteratingCallback#failed(Throwable)}
*/
FAILED,
/**
* This callback has been closed and cannot be reset.
*/
CLOSED
}
/**
* The indication of the overall progress of the overall job that
* implementations of {@link #process()} must return.
*/
protected enum Action
{
/**
* Indicates that {@link #process()} has no more work to do,
* but the overall job is not completed yet, probably waiting
* for additional events to trigger more work.
*/
IDLE,
/**
* Indicates that {@link #process()} is executing asynchronously
* a sub task, where the execution has started but the callback
* may have not yet been invoked.
*/
SCHEDULED,
/**
* Indicates that {@link #process()} has completed the overall job.
*/
SUCCEEDED
}
private Locker _locker = new Locker();
private State _state;
private boolean _iterate;
protected IteratingCallback()
{
_state = State.IDLE;
}
protected IteratingCallback(boolean needReset)
{
_state = needReset ? State.SUCCEEDED : State.IDLE;
}
/**
* Method called by {@link #iterate()} to process the sub task.
* <p>
* Implementations must start the asynchronous execution of the sub task
* (if any) and return an appropriate action:
* </p>
* <ul>
* <li>{@link Action#IDLE} when no sub tasks are available for execution
* but the overall job is not completed yet</li>
* <li>{@link Action#SCHEDULED} when the sub task asynchronous execution
* has been started</li>
* <li>{@link Action#SUCCEEDED} when the overall job is completed</li>
* </ul>
*
* @return the appropriate Action
*
* @throws Throwable if the sub task processing throws
*/
protected abstract Action process() throws Throwable;
/**
* Invoked when the overall task has completed successfully.
*
* @see #onCompleteFailure(Throwable)
*/
protected void onCompleteSuccess()
{
}
/**
* Invoked when the overall task has completed with a failure.
* @param cause the throwable to indicate cause of failure
*
* @see #onCompleteSuccess()
*/
protected void onCompleteFailure(Throwable cause)
{
}
/**
* This method must be invoked by applications to start the processing
* of sub tasks. It can be called at any time by any thread, and it's
* contract is that when called, then the {@link #process()} method will
* be called during or soon after, either by the calling thread or by
* another thread.
*/
public void iterate()
{
boolean process=false;
loop: while (true)
{
try (Locker.Lock lock = _locker.lock())
{
switch (_state)
{
case PENDING:
case CALLED:
// process will be called when callback is handled
break loop;
case IDLE:
_state=State.PROCESSING;
process=true;
break loop;
case PROCESSING:
_iterate=true;
break loop;
case FAILED:
case SUCCEEDED:
break loop;
case CLOSED:
default:
throw new IllegalStateException(toString());
}
}
}
if (process)
processing();
}
private void processing()
{
// This should only ever be called when in processing state, however a failed or close call
// may happen concurrently, so state is not assumed.
boolean on_complete_success=false;
// While we are processing
processing: while (true)
{
// Call process to get the action that we have to take.
Action action;
try
{
action = process();
}
catch (Throwable x)
{
failed(x);
break processing;
}
// acted on the action we have just received
try(Locker.Lock lock = _locker.lock())
{
switch (_state)
{
case PROCESSING:
{
switch (action)
{
case IDLE:
{
// Has iterate been called while we were processing?
if (_iterate)
{
// yes, so skip idle and keep processing
_iterate=false;
_state=State.PROCESSING;
continue processing;
}
// No, so we can go idle
_state=State.IDLE;
break processing;
}
case SCHEDULED:
{
// we won the race against the callback, so the callback has to process and we can break processing
_state=State.PENDING;
break processing;
}
case SUCCEEDED:
{
// we lost the race against the callback,
_iterate=false;
_state=State.SUCCEEDED;
on_complete_success=true;
break processing;
}
default:
throw new IllegalStateException(String.format("%s[action=%s]", this, action));
}
}
case CALLED:
{
switch (action)
{
case SCHEDULED:
{
// we lost the race, so we have to keep processing
_state=State.PROCESSING;
continue processing;
}
default:
throw new IllegalStateException(String.format("%s[action=%s]", this, action));
}
}
case SUCCEEDED:
case FAILED:
case CLOSED:
break processing;
case IDLE:
case PENDING:
default:
throw new IllegalStateException(String.format("%s[action=%s]", this, action));
}
}
}
if (on_complete_success)
onCompleteSuccess();
}
/**
* Invoked when the sub task succeeds.
* Subclasses that override this method must always remember to call
* {@code super.succeeded()}.
*/
@Override
public void succeeded()
{
boolean process=false;
try(Locker.Lock lock = _locker.lock())
{
switch (_state)
{
case PROCESSING:
{
_state=State.CALLED;
break;
}
case PENDING:
{
_state=State.PROCESSING;
process=true;
break;
}
case CLOSED:
case FAILED:
{
// Too late!
break;
}
default:
{
throw new IllegalStateException(toString());
}
}
}
if (process)
processing();
}
/**
* Invoked when the sub task fails.
* Subclasses that override this method must always remember to call
* {@code super.failed(Throwable)}.
*/
@Override
public void failed(Throwable x)
{
boolean failure=false;
try(Locker.Lock lock = _locker.lock())
{
switch (_state)
{
case SUCCEEDED:
case FAILED:
case IDLE:
case CLOSED:
case CALLED:
// too late!.
break;
case PENDING:
case PROCESSING:
{
_state=State.FAILED;
failure=true;
break;
}
default:
throw new IllegalStateException(toString());
}
}
if (failure)
onCompleteFailure(x);
}
public void close()
{
boolean failure=false;
try(Locker.Lock lock = _locker.lock())
{
switch (_state)
{
case IDLE:
case SUCCEEDED:
case FAILED:
_state=State.CLOSED;
break;
case CLOSED:
break;
default:
_state=State.CLOSED;
failure=true;
}
}
if(failure)
onCompleteFailure(new ClosedChannelException());
}
/*
* only for testing
* @return whether this callback is idle and {@link #iterate()} needs to be called
*/
boolean isIdle()
{
try(Locker.Lock lock = _locker.lock())
{
return _state == State.IDLE;
}
}
public boolean isClosed()
{
try(Locker.Lock lock = _locker.lock())
{
return _state == State.CLOSED;
}
}
/**
* @return whether this callback has failed
*/
public boolean isFailed()
{
try(Locker.Lock lock = _locker.lock())
{
return _state == State.FAILED;
}
}
/**
* @return whether this callback has succeeded
*/
public boolean isSucceeded()
{
try(Locker.Lock lock = _locker.lock())
{
return _state == State.SUCCEEDED;
}
}
/**
* Resets this callback.
* <p>
* A callback can only be reset to IDLE from the
* SUCCEEDED or FAILED states or if it is already IDLE.
* </p>
*
* @return true if the reset was successful
*/
public boolean reset()
{
try(Locker.Lock lock = _locker.lock())
{
switch(_state)
{
case IDLE:
return true;
case SUCCEEDED:
case FAILED:
_iterate=false;
_state=State.IDLE;
return true;
default:
return false;
}
}
}
@Override
public String toString()
{
return String.format("%s[%s]", super.toString(), _state);
}
}