Examples with InputAttemptIdentifier org.apache.tez.runtime.library.common.InputAttemptIdentifier used on opensource projects

Example 46 with InputAttemptIdentifier

use of org.apache.tez.runtime.library.common.InputAttemptIdentifier in project tez by apache.

the class TestShuffleScheduler method testReducerHealth_5.

@Test(timeout = 60000)
public /**
 * Scenario
 *    - Shuffle has progressed enough
 *    - Last event is yet to arrive
 *    - Failures start happening after Shuffle has progressed enough
 *    - no of attempts failing does not exceed maxFailedUniqueFetches (5)
 *    - Stalled
 * Expected result
 *    - Do not throw errors, as Shuffle is yet to receive inputs
 */
void testReducerHealth_5() throws IOException {
    long startTime = System.currentTimeMillis() - 500000;
    Shuffle shuffle = mock(Shuffle.class);
    final ShuffleSchedulerForTest scheduler = createScheduler(startTime, 320, shuffle);
    int totalProducerNodes = 20;
    // Generate 319 events (last event has not arrived)
    for (int i = 0; i < 319; i++) {
        CompositeInputAttemptIdentifier inputAttemptIdentifier = new CompositeInputAttemptIdentifier(i, 0, "attempt_", 1);
        scheduler.addKnownMapOutput("host" + (i % totalProducerNodes), 10000, i, inputAttemptIdentifier);
    }
    // 318 succeeds
    for (int i = 0; i < 319; i++) {
        InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(i, 0, "attempt_");
        MapOutput mapOutput = MapOutput.createMemoryMapOutput(inputAttemptIdentifier, mock(FetchedInputAllocatorOrderedGrouped.class), 100, false);
        scheduler.copySucceeded(inputAttemptIdentifier, new MapHost("host" + (i % totalProducerNodes), 10000, i, 1), 100, 200, startTime + (i * 100), mapOutput, false);
    }
    // 1 fails (last fetch)
    InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(318, 0, "attempt_");
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (318 % totalProducerNodes), 10000, 318, 1), false, true, false);
    // stall the shuffle
    scheduler.lastProgressTime = System.currentTimeMillis() - 1000000;
    assertEquals(scheduler.remainingMaps.get(), 1);
    // Retry for 3 more times
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (318 % totalProducerNodes), 10000, 318, 1), false, true, false);
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (318 % totalProducerNodes), 10000, 318, 1), false, true, false);
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (318 % totalProducerNodes), 10000, 318, 1), false, true, false);
    // Shuffle has not received the events completely. So do not bail out yet.
    verify(shuffle, times(0)).reportException(any(Throwable.class));
}

Example 47 with InputAttemptIdentifier

use of org.apache.tez.runtime.library.common.InputAttemptIdentifier in project tez by apache.

the class TestShuffleScheduler method testSimpleFlow.

@Test(timeout = 5000)
public void testSimpleFlow() throws Exception {
    InputContext inputContext = createTezInputContext();
    Configuration conf = new TezConfiguration();
    int numInputs = 10;
    Shuffle shuffle = mock(Shuffle.class);
    MergeManager mergeManager = mock(MergeManager.class);
    final ShuffleSchedulerForTest scheduler = new ShuffleSchedulerForTest(inputContext, conf, numInputs, shuffle, mergeManager, mergeManager, System.currentTimeMillis(), null, false, 0, "srcName");
    ExecutorService executor = Executors.newFixedThreadPool(1);
    try {
        Future<Void> executorFuture = executor.submit(new Callable<Void>() {

            @Override
            public Void call() throws Exception {
                scheduler.start();
                return null;
            }
        });
        InputAttemptIdentifier[] identifiers = new InputAttemptIdentifier[numInputs];
        for (int i = 0; i < numInputs; i++) {
            CompositeInputAttemptIdentifier inputAttemptIdentifier = new CompositeInputAttemptIdentifier(i, 0, "attempt_", 1);
            scheduler.addKnownMapOutput("host" + i, 10000, 1, inputAttemptIdentifier);
            identifiers[i] = inputAttemptIdentifier;
        }
        MapHost[] mapHosts = new MapHost[numInputs];
        int count = 0;
        for (MapHost mh : scheduler.mapLocations.values()) {
            mapHosts[count++] = mh;
        }
        for (int i = 0; i < numInputs; i++) {
            MapOutput mapOutput = MapOutput.createMemoryMapOutput(identifiers[i], mock(FetchedInputAllocatorOrderedGrouped.class), 100, false);
            scheduler.copySucceeded(identifiers[i], mapHosts[i], 20, 25, 100, mapOutput, false);
            scheduler.freeHost(mapHosts[i]);
        }
        verify(inputContext, atLeast(numInputs)).notifyProgress();
        // Ensure the executor exits, and without an error.
        executorFuture.get();
    } finally {
        scheduler.close();
        executor.shutdownNow();
    }
}

Example 48 with InputAttemptIdentifier

use of org.apache.tez.runtime.library.common.InputAttemptIdentifier in project tez by apache.

the class TestShuffleScheduler method testReducerHealth_3.

@Test(timeout = 60000)
public /**
 * Scenario
 *    - reducer has progressed enough
 *    - failures start happening after that in last fetch
 *    - no of attempts failing does not exceed maxFailedUniqueFetches (5)
 *    - Stalled
 * Expected result
 *    - Since reducer is stalled and if failures haven't happened across nodes,
 *    it should be fine to proceed. AM would restart source task eventually.
 */
void testReducerHealth_3() throws IOException {
    long startTime = System.currentTimeMillis() - 500000;
    Shuffle shuffle = mock(Shuffle.class);
    final ShuffleSchedulerForTest scheduler = createScheduler(startTime, 320, shuffle);
    int totalProducerNodes = 20;
    // Generate 320 events
    for (int i = 0; i < 320; i++) {
        CompositeInputAttemptIdentifier inputAttemptIdentifier = new CompositeInputAttemptIdentifier(i, 0, "attempt_", 1);
        scheduler.addKnownMapOutput("host" + (i % totalProducerNodes), 10000, i, inputAttemptIdentifier);
    }
    // 319 succeeds
    for (int i = 0; i < 319; i++) {
        InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(i, 0, "attempt_");
        MapOutput mapOutput = MapOutput.createMemoryMapOutput(inputAttemptIdentifier, mock(FetchedInputAllocatorOrderedGrouped.class), 100, false);
        scheduler.copySucceeded(inputAttemptIdentifier, new MapHost("host" + (i % totalProducerNodes), 10000, i, 1), 100, 200, startTime + (i * 100), mapOutput, false);
    }
    // 1 fails (last fetch)
    InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(319, 0, "attempt_");
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (319 % totalProducerNodes), 10000, 319, 1), false, true, false);
    // stall the shuffle
    scheduler.lastProgressTime = System.currentTimeMillis() - 1000000;
    assertEquals(scheduler.remainingMaps.get(), 1);
    // Retry for 3 more times
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (319 % totalProducerNodes), 10000, 319, 1), false, true, false);
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (319 % totalProducerNodes), 10000, 310, 1), false, true, false);
    scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (319 % totalProducerNodes), 10000, 310, 1), false, true, false);
    // failedShufflesSinceLastCompletion has crossed the limits. Throw error
    verify(shuffle, times(0)).reportException(any(Throwable.class));
}

Example 49 with InputAttemptIdentifier

use of org.apache.tez.runtime.library.common.InputAttemptIdentifier in project tez by apache.

the class TestShuffleScheduler method testShutdownWithInterrupt.

@Test(timeout = 30000)
public void testShutdownWithInterrupt() throws Exception {
    InputContext inputContext = createTezInputContext();
    Configuration conf = new TezConfiguration();
    int numInputs = 10;
    Shuffle shuffle = mock(Shuffle.class);
    MergeManager mergeManager = mock(MergeManager.class);
    final ShuffleSchedulerForTest scheduler = new ShuffleSchedulerForTest(inputContext, conf, numInputs, shuffle, mergeManager, mergeManager, System.currentTimeMillis(), null, false, 0, "srcName");
    ExecutorService executor = Executors.newFixedThreadPool(1);
    Future<Void> executorFuture = executor.submit(new Callable<Void>() {

        @Override
        public Void call() throws Exception {
            scheduler.start();
            return null;
        }
    });
    InputAttemptIdentifier[] identifiers = new InputAttemptIdentifier[numInputs];
    for (int i = 0; i < numInputs; i++) {
        CompositeInputAttemptIdentifier inputAttemptIdentifier = new CompositeInputAttemptIdentifier(i, 0, "attempt_", 1);
        scheduler.addKnownMapOutput("host" + i, 10000, 1, inputAttemptIdentifier);
        identifiers[i] = inputAttemptIdentifier;
    }
    MapHost[] mapHosts = new MapHost[numInputs];
    int count = 0;
    for (MapHost mh : scheduler.mapLocations.values()) {
        mapHosts[count++] = mh;
    }
    // Copy succeeded for 1 less host
    for (int i = 0; i < numInputs - 1; i++) {
        MapOutput mapOutput = MapOutput.createMemoryMapOutput(identifiers[i], mock(FetchedInputAllocatorOrderedGrouped.class), 100, false);
        scheduler.copySucceeded(identifiers[i], mapHosts[i], 20, 25, 100, mapOutput, false);
        scheduler.freeHost(mapHosts[i]);
    }
    try {
        // Close the scheduler on different thread to trigger interrupt
        Thread thread = new Thread(new Runnable() {

            @Override
            public void run() {
                scheduler.close();
            }
        });
        thread.start();
        thread.join();
    } finally {
        assertTrue("Fetcher executor should be shutdown, but still running", scheduler.hasFetcherExecutorStopped());
        executor.shutdownNow();
    }
}

Example 50 with InputAttemptIdentifier

use of org.apache.tez.runtime.library.common.InputAttemptIdentifier in project tez by apache.

the class TestShuffleScheduler method _testReducerHealth_6.

public void _testReducerHealth_6(Configuration conf) throws IOException {
    long startTime = System.currentTimeMillis() - 500000;
    Shuffle shuffle = mock(Shuffle.class);
    final ShuffleSchedulerForTest scheduler = createScheduler(startTime, 320, shuffle, conf);
    int totalProducerNodes = 20;
    // Generate 320 events (last event has not arrived)
    for (int i = 0; i < 320; i++) {
        CompositeInputAttemptIdentifier inputAttemptIdentifier = new CompositeInputAttemptIdentifier(i, 0, "attempt_", 1);
        scheduler.addKnownMapOutput("host" + (i % totalProducerNodes), 10000, i, inputAttemptIdentifier);
    }
    // 10 succeeds
    for (int i = 0; i < 10; i++) {
        InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(i, 0, "attempt_");
        MapOutput mapOutput = MapOutput.createMemoryMapOutput(inputAttemptIdentifier, mock(FetchedInputAllocatorOrderedGrouped.class), 100, false);
        scheduler.copySucceeded(inputAttemptIdentifier, new MapHost("host" + (i % totalProducerNodes), 10000, i, 1), 100, 200, startTime + (i * 100), mapOutput, false);
    }
    // 5 fetches fail once
    for (int i = 10; i < 15; i++) {
        InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(i, 0, "attempt_");
        scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (i % totalProducerNodes), 10000, i, 1), false, true, false);
    }
    assertTrue(scheduler.failureCounts.size() >= 5);
    assertEquals(scheduler.remainingMaps.get(), 310);
    // Do not bail out (number of failures is just 5)
    verify(scheduler.reporter, times(0)).reportException(any(Throwable.class));
    // 5 fetches fail repeatedly
    for (int i = 10; i < 15; i++) {
        InputAttemptIdentifier inputAttemptIdentifier = new InputAttemptIdentifier(i, 0, "attempt_");
        scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (i % totalProducerNodes), 10000, i, 1), false, true, false);
        scheduler.copyFailed(inputAttemptIdentifier, new MapHost("host" + (i % totalProducerNodes), 10000, i, 1), false, true, false);
    }
    boolean checkFailedFetchSinceLastCompletion = conf.getBoolean(TezRuntimeConfiguration.TEZ_RUNTIME_SHUFFLE_FAILED_CHECK_SINCE_LAST_COMPLETION, TezRuntimeConfiguration.TEZ_RUNTIME_SHUFFLE_FAILED_CHECK_SINCE_LAST_COMPLETION_DEFAULT);
    if (checkFailedFetchSinceLastCompletion) {
        // Now bail out, as Shuffle has crossed the
        // failedShufflesSinceLastCompletion limits. (even
        // though reducerHeathly is
        verify(shuffle, atLeast(1)).reportException(any(Throwable.class));
    } else {
        // Do not bail out yet.
        verify(shuffle, atLeast(0)).reportException(any(Throwable.class));
    }
}