Examples with PartitionNotFoundException org.apache.flink.runtime.io.network.partition.PartitionNotFoundException used on opensource projects

Example 1 with PartitionNotFoundException

use of org.apache.flink.runtime.io.network.partition.PartitionNotFoundException in project flink by apache.

the class LocalInputChannelTest method testConcurrentReleaseAndRetriggerPartitionRequest.

/**
	 * Verifies that concurrent release via the SingleInputGate and re-triggering
	 * of a partition request works smoothly.
	 *
	 * - SingleInputGate acquires its request lock and tries to release all
	 * registered channels. When releasing a channel, it needs to acquire
	 * the channel's shared request-release lock.
	 * - If a LocalInputChannel concurrently retriggers a partition request via
	 * a Timer Thread it acquires the channel's request-release lock and calls
	 * the retrigger callback on the SingleInputGate, which again tries to
	 * acquire the gate's request lock.
	 *
	 * For certain timings this obviously leads to a deadlock. This test reliably
	 * reproduced such a timing (reported in FLINK-5228). This test is pretty much
	 * testing the buggy implementation and has not much more general value. If it
	 * becomes obsolete at some point (future greatness ;)), feel free to remove it.
	 *
	 * The fix in the end was to to not acquire the channels lock when releasing it
	 * and/or not doing any input gate callbacks while holding the channel's lock.
	 * I decided to do both.
	 */
@Test
public void testConcurrentReleaseAndRetriggerPartitionRequest() throws Exception {
    final SingleInputGate gate = new SingleInputGate("test task name", new JobID(), new IntermediateDataSetID(), ResultPartitionType.PIPELINED, 0, 1, mock(TaskActions.class), new UnregisteredTaskMetricsGroup.DummyTaskIOMetricGroup());
    ResultPartitionManager partitionManager = mock(ResultPartitionManager.class);
    when(partitionManager.createSubpartitionView(any(ResultPartitionID.class), anyInt(), any(BufferProvider.class), any(BufferAvailabilityListener.class))).thenAnswer(new Answer<ResultSubpartitionView>() {

        @Override
        public ResultSubpartitionView answer(InvocationOnMock invocationOnMock) throws Throwable {
            // Sleep here a little to give the releaser Thread
            // time to acquire the input gate lock. We throw
            // the Exception to retrigger the request.
            Thread.sleep(100);
            throw new PartitionNotFoundException(new ResultPartitionID());
        }
    });
    final LocalInputChannel channel = new LocalInputChannel(gate, 0, new ResultPartitionID(), partitionManager, new TaskEventDispatcher(), 1, 1, new UnregisteredTaskMetricsGroup.DummyTaskIOMetricGroup());
    gate.setInputChannel(new IntermediateResultPartitionID(), channel);
    Thread releaser = new Thread() {

        @Override
        public void run() {
            try {
                gate.releaseAllResources();
            } catch (IOException ignored) {
            }
        }
    };
    Thread requester = new Thread() {

        @Override
        public void run() {
            try {
                channel.requestSubpartition(0);
            } catch (IOException | InterruptedException ignored) {
            }
        }
    };
    requester.start();
    releaser.start();
    releaser.join();
    requester.join();
}

Example 2 with PartitionNotFoundException

use of org.apache.flink.runtime.io.network.partition.PartitionNotFoundException in project flink by apache.

the class LocalInputChannelTest method testPartitionRequestExponentialBackoff.

@Test
public void testPartitionRequestExponentialBackoff() throws Exception {
    // Config
    Tuple2<Integer, Integer> backoff = new Tuple2<>(500, 3000);
    // Start with initial backoff, then keep doubling, and cap at max.
    int[] expectedDelays = { backoff._1(), 1000, 2000, backoff._2() };
    // Setup
    SingleInputGate inputGate = mock(SingleInputGate.class);
    BufferProvider bufferProvider = mock(BufferProvider.class);
    when(inputGate.getBufferProvider()).thenReturn(bufferProvider);
    ResultPartitionManager partitionManager = mock(ResultPartitionManager.class);
    LocalInputChannel ch = createLocalInputChannel(inputGate, partitionManager, backoff);
    when(partitionManager.createSubpartitionView(eq(ch.partitionId), eq(0), eq(bufferProvider), any(BufferAvailabilityListener.class))).thenThrow(new PartitionNotFoundException(ch.partitionId));
    Timer timer = mock(Timer.class);
    doAnswer(new Answer<Void>() {

        @Override
        public Void answer(InvocationOnMock invocation) throws Throwable {
            ((TimerTask) invocation.getArguments()[0]).run();
            return null;
        }
    }).when(timer).schedule(any(TimerTask.class), anyLong());
    // Initial request
    ch.requestSubpartition(0);
    verify(partitionManager).createSubpartitionView(eq(ch.partitionId), eq(0), eq(bufferProvider), any(BufferAvailabilityListener.class));
    // Request subpartition and verify that the actual requests are delayed.
    for (long expected : expectedDelays) {
        ch.retriggerSubpartitionRequest(timer, 0);
        verify(timer).schedule(any(TimerTask.class), eq(expected));
    }
    // Exception after backoff is greater than the maximum backoff.
    try {
        ch.retriggerSubpartitionRequest(timer, 0);
        ch.getNextBuffer();
        fail("Did not throw expected exception.");
    } catch (Exception expected) {
    }
}

Example 3 with PartitionNotFoundException

use of org.apache.flink.runtime.io.network.partition.PartitionNotFoundException in project flink by apache.

the class TaskManagerTest method testRemotePartitionNotFound.

/**
	 * Tests that repeated remote {@link PartitionNotFoundException}s ultimately fail the receiver.
	 */
@Test
public void testRemotePartitionNotFound() throws Exception {
    new JavaTestKit(system) {

        {
            ActorGateway jobManager = null;
            ActorGateway taskManager = null;
            final ActorGateway testActorGateway = new AkkaActorGateway(getTestActor(), leaderSessionID);
            try {
                final IntermediateDataSetID resultId = new IntermediateDataSetID();
                // Create the JM
                ActorRef jm = system.actorOf(Props.create(new SimplePartitionStateLookupJobManagerCreator(leaderSessionID, getTestActor())));
                jobManager = new AkkaActorGateway(jm, leaderSessionID);
                final int dataPort = NetUtils.getAvailablePort();
                Configuration config = new Configuration();
                config.setInteger(ConfigConstants.TASK_MANAGER_DATA_PORT_KEY, dataPort);
                config.setInteger(TaskManagerOptions.NETWORK_REQUEST_BACKOFF_INITIAL, 100);
                config.setInteger(TaskManagerOptions.NETWORK_REQUEST_BACKOFF_MAX, 200);
                taskManager = TestingUtils.createTaskManager(system, jobManager, config, false, true);
                // ---------------------------------------------------------------------------------
                final ActorGateway tm = taskManager;
                final JobID jid = new JobID();
                final JobVertexID vid = new JobVertexID();
                final ExecutionAttemptID eid = new ExecutionAttemptID();
                final ResultPartitionID partitionId = new ResultPartitionID();
                // Remote location (on the same TM though) for the partition
                final ResultPartitionLocation loc = ResultPartitionLocation.createRemote(new ConnectionID(new InetSocketAddress("localhost", dataPort), 0));
                final InputChannelDeploymentDescriptor[] icdd = new InputChannelDeploymentDescriptor[] { new InputChannelDeploymentDescriptor(partitionId, loc) };
                final InputGateDeploymentDescriptor igdd = new InputGateDeploymentDescriptor(resultId, ResultPartitionType.PIPELINED, 0, icdd);
                final TaskDeploymentDescriptor tdd = createTaskDeploymentDescriptor(jid, "TestJob", vid, eid, new SerializedValue<>(new ExecutionConfig()), "Receiver", 1, 0, 1, 0, new Configuration(), new Configuration(), Tasks.AgnosticReceiver.class.getName(), Collections.<ResultPartitionDeploymentDescriptor>emptyList(), Collections.singletonList(igdd), Collections.<BlobKey>emptyList(), Collections.<URL>emptyList(), 0);
                new Within(d) {

                    @Override
                    protected void run() {
                        // Submit the task
                        tm.tell(new SubmitTask(tdd), testActorGateway);
                        expectMsgClass(Acknowledge.get().getClass());
                        // Wait to be notified about the final execution state by the mock JM
                        TaskExecutionState msg = expectMsgClass(TaskExecutionState.class);
                        // The task should fail after repeated requests
                        assertEquals(ExecutionState.FAILED, msg.getExecutionState());
                        Throwable t = msg.getError(ClassLoader.getSystemClassLoader());
                        assertEquals("Thrown exception was not a PartitionNotFoundException: " + t.getMessage(), PartitionNotFoundException.class, t.getClass());
                    }
                };
            } catch (Exception e) {
                e.printStackTrace();
                fail(e.getMessage());
            } finally {
                TestingUtils.stopActor(taskManager);
                TestingUtils.stopActor(jobManager);
            }
        }
    };
}

Example 4 with PartitionNotFoundException

use of org.apache.flink.runtime.io.network.partition.PartitionNotFoundException in project flink by apache.

the class PartitionRequestServerHandler method channelRead0.

@Override
protected void channelRead0(ChannelHandlerContext ctx, NettyMessage msg) throws Exception {
    try {
        Class<?> msgClazz = msg.getClass();
        // ----------------------------------------------------------------
        if (msgClazz == PartitionRequest.class) {
            PartitionRequest request = (PartitionRequest) msg;
            LOG.debug("Read channel on {}: {}.", ctx.channel().localAddress(), request);
            try {
                SequenceNumberingViewReader reader = new SequenceNumberingViewReader(request.receiverId, outboundQueue);
                reader.requestSubpartitionView(partitionProvider, request.partitionId, request.queueIndex, bufferPool);
            } catch (PartitionNotFoundException notFound) {
                respondWithError(ctx, notFound, request.receiverId);
            }
        } else // ----------------------------------------------------------------
        if (msgClazz == TaskEventRequest.class) {
            TaskEventRequest request = (TaskEventRequest) msg;
            if (!taskEventDispatcher.publish(request.partitionId, request.event)) {
                respondWithError(ctx, new IllegalArgumentException("Task event receiver not found."), request.receiverId);
            }
        } else if (msgClazz == CancelPartitionRequest.class) {
            CancelPartitionRequest request = (CancelPartitionRequest) msg;
            outboundQueue.cancel(request.receiverId);
        } else if (msgClazz == CloseRequest.class) {
            outboundQueue.close();
        } else {
            LOG.warn("Received unexpected client request: {}", msg);
        }
    } catch (Throwable t) {
        respondWithError(ctx, t);
    }
}

Example 5 with PartitionNotFoundException

use of org.apache.flink.runtime.io.network.partition.PartitionNotFoundException in project flink by apache.

the class LocalInputChannel method requestSubpartition.

// ------------------------------------------------------------------------
// Consume
// ------------------------------------------------------------------------
@Override
void requestSubpartition(int subpartitionIndex) throws IOException, InterruptedException {
    boolean retriggerRequest = false;
    // The lock is required to request only once in the presence of retriggered requests.
    synchronized (requestLock) {
        checkState(!isReleased, "LocalInputChannel has been released already");
        if (subpartitionView == null) {
            LOG.debug("{}: Requesting LOCAL subpartition {} of partition {}.", this, subpartitionIndex, partitionId);
            try {
                ResultSubpartitionView subpartitionView = partitionManager.createSubpartitionView(partitionId, subpartitionIndex, inputGate.getBufferProvider(), this);
                if (subpartitionView == null) {
                    throw new IOException("Error requesting subpartition.");
                }
                // make the subpartition view visible
                this.subpartitionView = subpartitionView;
                // check if the channel was released in the meantime
                if (isReleased) {
                    subpartitionView.releaseAllResources();
                    this.subpartitionView = null;
                }
            } catch (PartitionNotFoundException notFound) {
                if (increaseBackoff()) {
                    retriggerRequest = true;
                } else {
                    throw notFound;
                }
            }
        }
    }
    // input gate.
    if (retriggerRequest) {
        inputGate.retriggerPartitionRequest(partitionId.getPartitionId());
    }
}