Examples with SystemClock org.apache.hadoop.yarn.util.SystemClock used on opensource projects

Example 11 with SystemClock

use of org.apache.hadoop.yarn.util.SystemClock in project apex-core by apache.

the class CheckpointTest method testUpdateRecoveryCheckpoint.

@Test
public void testUpdateRecoveryCheckpoint() throws Exception {
    Clock clock = new SystemClock();
    dag.setAttribute(com.datatorrent.api.Context.OperatorContext.STORAGE_AGENT, new MemoryStorageAgent());
    GenericTestOperator o1 = dag.addOperator("o1", GenericTestOperator.class);
    GenericTestOperator o2 = dag.addOperator("o2", GenericTestOperator.class);
    GenericTestOperator o3SL = dag.addOperator("o3SL", StatelessOperator.class);
    dag.addStream("o1.output1", o1.outport1, o2.inport1);
    dag.addStream("o2.output1", o2.outport1, o3SL.inport1);
    StreamingContainerManager dnm = new StreamingContainerManager(dag);
    PhysicalPlan plan = dnm.getPhysicalPlan();
    for (PTOperator oper : plan.getAllOperators().values()) {
        Assert.assertEquals("activation windowId " + oper, Checkpoint.INITIAL_CHECKPOINT, oper.getRecoveryCheckpoint());
        Assert.assertEquals("checkpoints " + oper, Collections.emptyList(), oper.checkpoints);
    }
    List<PTOperator> nodes1 = plan.getOperators(dag.getMeta(o1));
    Assert.assertNotNull(nodes1);
    Assert.assertEquals(1, nodes1.size());
    PTOperator o1p1 = nodes1.get(0);
    PTOperator o2p1 = plan.getOperators(dag.getMeta(o2)).get(0);
    PTOperator o3SLp1 = plan.getOperators(dag.getMeta(o3SL)).get(0);
    // recovery checkpoint won't update in deploy state
    for (PTOperator oper : plan.getAllOperators().values()) {
        Assert.assertEquals("", PTOperator.State.PENDING_DEPLOY, oper.getState());
    }
    dnm.updateRecoveryCheckpoints(o2p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("no checkpoints " + o2p1, Checkpoint.INITIAL_CHECKPOINT, o2p1.getRecoveryCheckpoint());
    UpdateCheckpointsContext ctx = new UpdateCheckpointsContext(clock);
    dnm.updateRecoveryCheckpoints(o1p1, ctx, false);
    Assert.assertEquals("no checkpoints " + o1p1, Checkpoint.INITIAL_CHECKPOINT, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals("number dependencies " + ctx.visited, 3, ctx.visited.size());
    // adding checkpoints to upstream only does not move recovery checkpoint
    Checkpoint cp3 = new Checkpoint(3L, 0, 0);
    Checkpoint cp5 = new Checkpoint(5L, 0, 0);
    Checkpoint cp4 = new Checkpoint(4L, 0, 0);
    o1p1.checkpoints.add(cp3);
    o1p1.checkpoints.add(cp5);
    dnm.updateRecoveryCheckpoints(o1p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("checkpoint " + o1p1, Checkpoint.INITIAL_CHECKPOINT, o1p1.getRecoveryCheckpoint());
    o2p1.checkpoints.add(new Checkpoint(3L, 0, 0));
    dnm.updateRecoveryCheckpoints(o1p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("checkpoint " + o1p1, Checkpoint.INITIAL_CHECKPOINT, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals("checkpoint " + o2p1, Checkpoint.INITIAL_CHECKPOINT, o2p1.getRecoveryCheckpoint());
    // set leaf operator checkpoint
    dnm.addCheckpoint(o3SLp1, cp5);
    dnm.updateRecoveryCheckpoints(o1p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("checkpoint " + o1p1, Checkpoint.INITIAL_CHECKPOINT, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals("checkpoint " + o2p1, Checkpoint.INITIAL_CHECKPOINT, o2p1.getRecoveryCheckpoint());
    // set all operators as active to enable recovery window id update
    for (PTOperator oper : plan.getAllOperators().values()) {
        oper.setState(PTOperator.State.ACTIVE);
    }
    dnm.updateRecoveryCheckpoints(o1p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("checkpoint " + o1p1, cp3, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals("checkpoint " + o2p1, cp3, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals("checkpoint " + o3SLp1, cp5, o3SLp1.getRecoveryCheckpoint());
    Assert.assertNull("checkpoint null for stateless operator " + o3SLp1, o3SLp1.stats.checkpointStats);
    o2p1.checkpoints.add(cp4);
    dnm.updateRecoveryCheckpoints(o1p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("checkpoint " + o1p1, cp3, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals("checkpoint " + o2p1, cp4, o2p1.getRecoveryCheckpoint());
    o1p1.checkpoints.add(1, cp4);
    Assert.assertEquals(o1p1.checkpoints, getCheckpoints(3L, 4L, 5L));
    dnm.updateRecoveryCheckpoints(o1p1, new UpdateCheckpointsContext(clock), false);
    Assert.assertEquals("checkpoint " + o1p1, cp4, o1p1.getRecoveryCheckpoint());
    Assert.assertEquals(o1p1.checkpoints, getCheckpoints(4L, 5L));
    // out of sequence windowIds should be sorted
    dnm.addCheckpoint(o2p1, new Checkpoint(2L, 0, 0));
    Assert.assertEquals("add first", getCheckpoints(2L, 4L), o2p1.checkpoints);
    dnm.addCheckpoint(o2p1, new Checkpoint(3L, 0, 0));
    Assert.assertEquals("add middle", getCheckpoints(2L, 3L, 4L), o2p1.checkpoints);
    dnm.addCheckpoint(o2p1, new Checkpoint(4L, 0, 0));
    Assert.assertEquals("ignore duplicate", getCheckpoints(2L, 3L, 4L), o2p1.checkpoints);
    dnm.addCheckpoint(o2p1, new Checkpoint(5L, 0, 0));
    Assert.assertEquals("add latest", getCheckpoints(2L, 3L, 4L, 5L), o2p1.checkpoints);
}

Example 12 with SystemClock

use of org.apache.hadoop.yarn.util.SystemClock in project apex-core by apache.

the class CheckpointTest method testUpdateRecoveryCheckpointWithCycle.

@Test
public void testUpdateRecoveryCheckpointWithCycle() throws Exception {
    Clock clock = new SystemClock();
    dag.setAttribute(com.datatorrent.api.Context.OperatorContext.STORAGE_AGENT, new MemoryStorageAgent());
    // Simulate a DAG with a loop which has a unifier operator
    TestGeneratorInputOperator o1 = dag.addOperator("o1", TestGeneratorInputOperator.class);
    GenericTestOperator o2 = dag.addOperator("o2", GenericTestOperator.class);
    GenericTestOperator o3 = dag.addOperator("o3", GenericTestOperator.class);
    GenericTestOperator o4 = dag.addOperator("o4", GenericTestOperator.class);
    DefaultDelayOperator d = dag.addOperator("d", DefaultDelayOperator.class);
    dag.addStream("o1.output1", o1.outport, o2.inport1);
    dag.addStream("o2.output1", o2.outport1, o3.inport1);
    dag.addStream("o3.output1", o3.outport1, o4.inport1);
    dag.addStream("o4.output1", o4.outport1, d.input);
    dag.addStream("d.output", d.output, o2.inport2);
    dag.setOperatorAttribute(o3, Context.OperatorContext.PARTITIONER, new StatelessPartitioner<Operator>(2));
    dag.validate();
    StreamingContainerManager dnm = new StreamingContainerManager(dag);
    PhysicalPlan plan = dnm.getPhysicalPlan();
    for (PTOperator oper : plan.getAllOperators().values()) {
        Assert.assertEquals("Initial activation windowId" + oper, Checkpoint.INITIAL_CHECKPOINT, oper.getRecoveryCheckpoint());
        Assert.assertEquals("Checkpoints empty" + oper, Collections.emptyList(), oper.checkpoints);
    }
    Checkpoint cp1 = new Checkpoint(1L, 0, 0);
    Checkpoint cp2 = new Checkpoint(2L, 0, 0);
    Map<OperatorMeta, Set<OperatorMeta>> checkpointGroups = dnm.getCheckpointGroups();
    Map<Integer, PTOperator> allOperators = plan.getAllOperators();
    for (PTOperator operator : allOperators.values()) {
        operator.setState(PTOperator.State.ACTIVE);
        operator.checkpoints.add(cp1);
        dnm.updateRecoveryCheckpoints(operator, new UpdateCheckpointsContext(clock, false, checkpointGroups), false);
    }
    List<PTOperator> physicalO1 = plan.getOperators(dag.getOperatorMeta("o1"));
    physicalO1.get(0).checkpoints.add(cp2);
    dnm.updateRecoveryCheckpoints(physicalO1.get(0), new UpdateCheckpointsContext(clock, false, checkpointGroups), false);
    Assert.assertEquals("Recovery checkpoint updated ", physicalO1.get(0).getRecoveryCheckpoint(), cp1);
}

Example 13 with SystemClock

use of org.apache.hadoop.yarn.util.SystemClock in project hive by apache.

the class TestTaskExecutorService method runPreemptionGraceTest.

private void runPreemptionGraceTest(MockRequest victim1, MockRequest victim2, int time) throws InterruptedException {
    MockRequest preemptor = createMockRequest(3, 1, 100, 100, true, 20000l, false);
    victim1.setSleepAfterKill();
    victim2.setSleepAfterKill();
    ControlledClock clock = new ControlledClock(new SystemClock());
    clock.setTime(0);
    TaskExecutorServiceForTest taskExecutorService = new TaskExecutorServiceForTest(2, 3, ShortestJobFirstComparator.class.getName(), true, clock);
    taskExecutorService.init(new Configuration());
    taskExecutorService.start();
    try {
        taskExecutorService.schedule(victim1);
        awaitStartAndSchedulerRun(victim1, taskExecutorService);
        taskExecutorService.schedule(victim2);
        awaitStartAndSchedulerRun(victim2, taskExecutorService);
        taskExecutorService.schedule(preemptor);
        // Wait for scheduling to run a few times.
        taskExecutorService.waitForScheduleRuns(5);
        clock.setTime(time);
        // Wait for scheduling to run a few times.
        taskExecutorService.waitForScheduleRuns(5);
        victim1.unblockKill();
        victim2.unblockKill();
        preemptor.complete();
        preemptor.awaitEnd();
        TaskExecutorServiceForTest.InternalCompletionListenerForTest icl3 = taskExecutorService.getInternalCompletionListenerForTest(preemptor.getRequestId());
        icl3.awaitCompletion();
    } finally {
        taskExecutorService.shutDown(false);
    }
}

Example 14 with SystemClock

use of org.apache.hadoop.yarn.util.SystemClock in project hive by apache.

the class LlapSliderUtils method getAppReport.

public static ApplicationReport getAppReport(String appName, SliderClient sliderClient, long timeoutMs) throws LlapStatusServiceDriver.LlapStatusCliException {
    Clock clock = new SystemClock();
    long startTime = clock.getTime();
    long timeoutTime = timeoutMs < 0 ? Long.MAX_VALUE : (startTime + timeoutMs);
    ApplicationReport appReport = null;
    while (appReport == null) {
        try {
            appReport = sliderClient.getYarnAppListClient().findInstance(appName);
            if (timeoutMs == 0) {
                // break immediately if timeout is 0
                break;
            }
            // Otherwise sleep, and try again.
            if (appReport == null) {
                long remainingTime = Math.min(timeoutTime - clock.getTime(), 500l);
                if (remainingTime > 0) {
                    Thread.sleep(remainingTime);
                } else {
                    break;
                }
            }
        } catch (Exception e) {
            // No point separating IOException vs YarnException vs others
            throw new LlapStatusServiceDriver.LlapStatusCliException(LlapStatusServiceDriver.ExitCode.YARN_ERROR, "Failed to get Yarn AppReport", e);
        }
    }
    return appReport;
}

Example 15 with SystemClock

use of org.apache.hadoop.yarn.util.SystemClock in project hive by apache.

the class LlapStatusServiceDriver method main.

public static void main(String[] args) {
    LOG.info("LLAP status invoked with arguments = {}", Arrays.toString(args));
    int ret = ExitCode.SUCCESS.getInt();
    Clock clock = new SystemClock();
    long startTime = clock.getTime();
    long lastSummaryLogTime = -1;
    LlapStatusServiceDriver statusServiceDriver = null;
    LlapStatusOptions options = null;
    try {
        statusServiceDriver = new LlapStatusServiceDriver();
        options = statusServiceDriver.parseOptions(args);
    } catch (Throwable t) {
        statusServiceDriver.close();
        logError(t);
        if (t instanceof LlapStatusCliException) {
            LlapStatusCliException ce = (LlapStatusCliException) t;
            ret = ce.getExitCode().getInt();
        } else {
            ret = ExitCode.INTERNAL_ERROR.getInt();
        }
    }
    if (ret != 0 || options == null) {
        // Failure / help
        if (statusServiceDriver != null) {
            statusServiceDriver.close();
        }
        System.exit(ret);
    }
    boolean firstAttempt = true;
    final long refreshInterval = options.getRefreshIntervalMs();
    final boolean watchMode = options.isWatchMode();
    final long watchTimeout = options.getWatchTimeoutMs();
    long numAttempts = watchTimeout / refreshInterval;
    // Break out of the loop fast if watchMode is disabled.
    numAttempts = watchMode ? numAttempts : 1;
    LlapStatusHelpers.State launchingState = null;
    LlapStatusHelpers.State currentState = null;
    boolean desiredStateAttained = false;
    final float runningNodesThreshold = options.getRunningNodesThreshold();
    try (OutputStream os = options.getOutputFile() == null ? System.out : new BufferedOutputStream(new FileOutputStream(options.getOutputFile()));
        PrintWriter pw = new PrintWriter(os)) {
        LOG.info("Configured refresh interval: {}s. Watch timeout: {}s. Attempts remaining: {}." + " Watch mode: {}. Running nodes threshold: {}.", TimeUnit.SECONDS.convert(refreshInterval, TimeUnit.MILLISECONDS), TimeUnit.SECONDS.convert(watchTimeout, TimeUnit.MILLISECONDS), numAttempts, watchMode, new DecimalFormat("#.###").format(runningNodesThreshold));
        while (numAttempts > 0) {
            if (!firstAttempt) {
                if (watchMode) {
                    try {
                        Thread.sleep(refreshInterval);
                    } catch (InterruptedException e) {
                    // ignore
                    }
                } else {
                    // reported once, so break
                    break;
                }
            } else {
                firstAttempt = false;
            }
            ret = statusServiceDriver.run(options, watchMode ? watchTimeout : 0);
            currentState = statusServiceDriver.appStatusBuilder.getState();
            try {
                lastSummaryLogTime = LlapStatusServiceDriver.maybeLogSummary(clock, lastSummaryLogTime, statusServiceDriver, watchMode, watchTimeout, launchingState);
            } catch (Exception e) {
                LOG.warn("Failed to log summary", e);
            }
            if (ret == ExitCode.SUCCESS.getInt()) {
                if (watchMode) {
                    // slider has started llap application, now if for some reason state changes to COMPLETE then fail fast
                    if (launchingState == null && LAUNCHING_STATES.contains(currentState)) {
                        launchingState = currentState;
                    }
                    if (currentState.equals(State.COMPLETE)) {
                        if (launchingState != null || options.isLaunched()) {
                            LOG.warn("COMPLETE state reached while waiting for RUNNING state. Failing.");
                            System.err.println("Final diagnostics: " + statusServiceDriver.appStatusBuilder.getDiagnostics());
                            break;
                        } else {
                            LOG.info("Found a stopped application; assuming it was a previous attempt " + "and waiting for the next one. Omit the -l flag to avoid this.");
                        }
                    }
                    if (!(currentState.equals(State.RUNNING_PARTIAL) || currentState.equals(State.RUNNING_ALL))) {
                        if (LOG.isDebugEnabled()) {
                            LOG.debug("Current state: {}. Desired state: {}. {}/{} instances.", currentState, runningNodesThreshold == 1.0f ? State.RUNNING_ALL : State.RUNNING_PARTIAL, statusServiceDriver.appStatusBuilder.getLiveInstances(), statusServiceDriver.appStatusBuilder.getDesiredInstances());
                        }
                        numAttempts--;
                        continue;
                    }
                    // we have reached RUNNING state, now check if running nodes threshold is met
                    final int liveInstances = statusServiceDriver.appStatusBuilder.getLiveInstances();
                    final int desiredInstances = statusServiceDriver.appStatusBuilder.getDesiredInstances();
                    if (desiredInstances > 0) {
                        final float ratio = (float) liveInstances / (float) desiredInstances;
                        if (ratio < runningNodesThreshold) {
                            if (LOG.isDebugEnabled()) {
                                LOG.debug("Waiting until running nodes threshold is reached. Current: {} Desired: {}." + " {}/{} instances.", new DecimalFormat("#.###").format(ratio), new DecimalFormat("#.###").format(runningNodesThreshold), statusServiceDriver.appStatusBuilder.getLiveInstances(), statusServiceDriver.appStatusBuilder.getDesiredInstances());
                            }
                            numAttempts--;
                            continue;
                        } else {
                            desiredStateAttained = true;
                            statusServiceDriver.appStatusBuilder.setRunningThresholdAchieved(true);
                        }
                    } else {
                        numAttempts--;
                        continue;
                    }
                }
            } else if (ret == ExitCode.YARN_ERROR.getInt() && watchMode) {
                LOG.warn("Watch mode enabled and got YARN error. Retrying..");
                numAttempts--;
                continue;
            } else if (ret == ExitCode.SLIDER_CLIENT_ERROR_CREATE_FAILED.getInt() && watchMode) {
                LOG.warn("Watch mode enabled and slider client creation failed. Retrying..");
                numAttempts--;
                continue;
            } else if (ret == ExitCode.SLIDER_CLIENT_ERROR_OTHER.getInt() && watchMode) {
                LOG.warn("Watch mode enabled and got slider client error. Retrying..");
                numAttempts--;
                continue;
            } else if (ret == ExitCode.LLAP_REGISTRY_ERROR.getInt() && watchMode) {
                LOG.warn("Watch mode enabled and got LLAP registry error. Retrying..");
                numAttempts--;
                continue;
            }
            break;
        }
        // Log final state to CONSOLE_LOGGER
        LlapStatusServiceDriver.maybeLogSummary(clock, 0L, statusServiceDriver, watchMode, watchTimeout, launchingState);
        CONSOLE_LOGGER.info("\n\n\n");
        // print current state before exiting
        statusServiceDriver.outputJson(pw);
        os.flush();
        pw.flush();
        if (numAttempts == 0 && watchMode && !desiredStateAttained) {
            LOG.warn("Watch timeout {}s exhausted before desired state RUNNING is attained.", TimeUnit.SECONDS.convert(watchTimeout, TimeUnit.MILLISECONDS));
        }
    } catch (Throwable t) {
        logError(t);
        if (t instanceof LlapStatusCliException) {
            LlapStatusCliException ce = (LlapStatusCliException) t;
            ret = ce.getExitCode().getInt();
        } else {
            ret = ExitCode.INTERNAL_ERROR.getInt();
        }
    } finally {
        LOG.info("LLAP status finished");
        statusServiceDriver.close();
    }
    if (LOG.isDebugEnabled()) {
        LOG.debug("Completed processing - exiting with " + ret);
    }
    System.exit(ret);
}