Package org.apache.hadoop.yarn.api.records

Examples of org.apache.hadoop.yarn.api.records.AMResponse


  }
 
  @SuppressWarnings("unchecked")
  private List<Container> getResources() throws Exception {
    int headRoom = getAvailableResources() != null ? getAvailableResources().getMemory() : 0;//first time it would be null
    AMResponse response;
    /*
     * If contact with RM is lost, the AM will wait MR_AM_TO_RM_WAIT_INTERVAL_MS
     * milliseconds before aborting. During this interval, AM will still try
     * to contact the RM.
     */
    try {
      response = makeRemoteRequest();
      // Reset retry count if no exception occurred.
      retrystartTime = System.currentTimeMillis();
    } catch (Exception e) {
      // This can happen when the connection to the RM has gone down. Keep
      // re-trying until the retryInterval has expired.
      if (System.currentTimeMillis() - retrystartTime >= retryInterval) {
        LOG.error("Could not contact RM after " + retryInterval + " milliseconds.");
        eventHandler.handle(new JobEvent(this.getJob().getID(),
                                         JobEventType.INTERNAL_ERROR));
        throw new YarnException("Could not contact RM after " +
                                retryInterval + " milliseconds.");
      }
      // Throw this up to the caller, which may decide to ignore it and
      // continue to attempt to contact the RM.
      throw e;
    }
    if (response.getReboot()) {
      // This can happen if the RM has been restarted. If it is in that state,
      // this application must clean itself up.
      eventHandler.handle(new JobEvent(this.getJob().getID(),
                                       JobEventType.INTERNAL_ERROR));
      throw new YarnException("Resource Manager doesn't recognize AttemptId: " +
                               this.getContext().getApplicationID());
    }
    int newHeadRoom = getAvailableResources() != null ? getAvailableResources().getMemory() : 0;
    List<Container> newContainers = response.getAllocatedContainers();
    List<ContainerStatus> finishedContainers = response.getCompletedContainersStatuses();
    if (newContainers.size() + finishedContainers.size() > 0 || headRoom != newHeadRoom) {
      //something changed
      recalculateReduceSchedule = true;
      if (LOG.isDebugEnabled() && headRoom != newHeadRoom) {
        LOG.debug("headroom=" + newHeadRoom);
View Full Code Here


        nm2.getNodeId());
    Assert.assertEquals(2 * GB, report_nm2.getUsedResource().getMemory());

    // add request for containers
    am1.addRequests(new String[] { "h1", "h2" }, GB, 1, 1);
    AMResponse am1Response = am1.schedule(); // send the request
    // add request for containers
    am2.addRequests(new String[] { "h1", "h2" }, 3 * GB, 0, 1);
    AMResponse am2Response = am2.schedule(); // send the request

    // kick the scheduler, 1 GB and 3 GB given to AM1 and AM2, remaining 0
    nm1.nodeHeartbeat(true);
    while (am1Response.getAllocatedContainers().size() < 1) {
      LOG.info("Waiting for containers to be created for app 1...");
      Thread.sleep(1000);
      am1Response = am1.schedule();
    }
    while (am2Response.getAllocatedContainers().size() < 1) {
      LOG.info("Waiting for containers to be created for app 2...");
      Thread.sleep(1000);
      am2Response = am2.schedule();
    }
    // kick the scheduler, nothing given remaining 2 GB.
    nm2.nodeHeartbeat(true);

    List<Container> allocated1 = am1Response.getAllocatedContainers();
    Assert.assertEquals(1, allocated1.size());
    Assert.assertEquals(1 * GB, allocated1.get(0).getResource().getMemory());
    Assert.assertEquals(nm1.getNodeId(), allocated1.get(0).getNodeId());

    List<Container> allocated2 = am2Response.getAllocatedContainers();
    Assert.assertEquals(1, allocated2.size());
    Assert.assertEquals(3 * GB, allocated2.get(0).getResource().getMemory());
    Assert.assertEquals(nm1.getNodeId(), allocated2.get(0).getNodeId());
   
    report_nm1 = rm.getResourceScheduler().getNodeReport(nm1.getNodeId());
View Full Code Here

      int containers) throws Exception {
    requests.addAll(createReq(hosts, memory, priority, containers));
  }

  public AMResponse schedule() throws Exception {
    AMResponse response = allocate(requests, releases);
    requests.clear();
    releases.clear();
    return response;
  }
View Full Code Here

    ApplicationAttemptId applicationAttemptId = request
        .getApplicationAttemptId();
    authorizeRequest(applicationAttemptId);

    ApplicationId appID = applicationAttemptId.getApplicationId();
    AMResponse lastResponse = responseMap.get(applicationAttemptId);
    if (lastResponse == null) {
      String message = "Application doesn't exist in cache "
          + applicationAttemptId;
      LOG.error(message);
      RMAuditLogger.logFailure(this.rmContext.getRMApps().get(appID).getUser(),
View Full Code Here

    ApplicationAttemptId applicationAttemptId = request
        .getApplicationAttemptId();
    authorizeRequest(applicationAttemptId);

    AMResponse lastResponse = responseMap.get(applicationAttemptId);
    if (lastResponse == null) {
      String message = "Application doesn't exist in cache "
          + applicationAttemptId;
      LOG.error(message);
      throw RPCUtil.getRemoteException(message);
View Full Code Here

    this.amLivelinessMonitor.receivedPing(appAttemptId);

    /* check if its in cache */
    AllocateResponse allocateResponse = recordFactory
        .newRecordInstance(AllocateResponse.class);
    AMResponse lastResponse = responseMap.get(appAttemptId);
    if (lastResponse == null) {
      LOG.error("AppAttemptId doesnt exist in cache " + appAttemptId);
      allocateResponse.setAMResponse(reboot);
      return allocateResponse;
    }
    if ((request.getResponseId() + 1) == lastResponse.getResponseId()) {
      /* old heartbeat */
      allocateResponse.setAMResponse(lastResponse);
      return allocateResponse;
    } else if (request.getResponseId() + 1 < lastResponse.getResponseId()) {
      LOG.error("Invalid responseid from appAttemptId " + appAttemptId);
      // Oh damn! Sending reboot isn't enough. RM state is corrupted. TODO:
      allocateResponse.setAMResponse(reboot);
      return allocateResponse;
    }

    // Allow only one thread in AM to do heartbeat at a time.
    synchronized (lastResponse) { // BUG TODO: Locking order is screwed.

      // Send the status update to the appAttempt.
      this.rmContext.getDispatcher().getEventHandler().handle(
          new RMAppAttemptStatusupdateEvent(appAttemptId, request
              .getProgress()));

      List<ResourceRequest> ask = request.getAskList();
      List<ContainerId> release = request.getReleaseList();

      // Send new requests to appAttempt.
      Allocation allocation =
          this.rScheduler.allocate(appAttemptId, ask, release);

      RMApp app = this.rmContext.getRMApps().get(appAttemptId.getApplicationId());
      RMAppAttempt appAttempt = app.getRMAppAttempt(appAttemptId);

      AMResponse response = recordFactory.newRecordInstance(AMResponse.class);
      response.setAllocatedContainers(allocation.getContainers());
      response.setCompletedContainersStatuses(appAttempt
          .pullJustFinishedContainers());
      response.setResponseId(lastResponse.getResponseId() + 1);
      response.setAvailableResources(allocation.getResourceLimit());
      responseMap.put(appAttemptId, response);
      allocateResponse.setAMResponse(response);
      return allocateResponse;
    }
  }
View Full Code Here

      return allocateResponse;
    }
  }

  public void registerAppAttempt(ApplicationAttemptId attemptId) {
    AMResponse response = recordFactory.newRecordInstance(AMResponse.class);
    response.setResponseId(0);
    LOG.info("Registering " + attemptId);
    responseMap.put(attemptId, response);
  }
View Full Code Here

    LOG.info("Registering " + attemptId);
    responseMap.put(attemptId, response);
  }

  public void unregisterAttempt(ApplicationAttemptId attemptId) {
    AMResponse lastResponse = responseMap.get(attemptId);
    if (lastResponse != null) {
      synchronized (lastResponse) {
        responseMap.remove(attemptId);
      }
    }
View Full Code Here

    MockAM am1 = rm.sendAMLaunched(attempt1.getAppAttemptId());
    am1.registerAppAttempt();

    LOG.info("sending container requests ");
    am1.addRequests(new String[] { "*" }, 3 * GB, 1, 1);
    AMResponse alloc1Response = am1.schedule(); // send the request

    // kick the scheduler
    nm1.nodeHeartbeat(true);
    int waitCounter = 20;
    LOG.info("heartbeating nm1");
    while (alloc1Response.getAllocatedContainers().size() < 1
        && waitCounter-- > 0) {
      LOG.info("Waiting for containers to be created for app 1...");
      Thread.sleep(500);
      alloc1Response = am1.schedule();
    }
    LOG.info("received container : "
        + alloc1Response.getAllocatedContainers().size());

    // No container should be allocated.
    // Internally it should not been reserved.
    Assert.assertTrue(alloc1Response.getAllocatedContainers().size() == 0);

    LOG.info("heartbeating nm2");
    waitCounter = 20;
    nm2.nodeHeartbeat(true);
    while (alloc1Response.getAllocatedContainers().size() < 1
        && waitCounter-- > 0) {
      LOG.info("Waiting for containers to be created for app 1...");
      Thread.sleep(500);
      alloc1Response = am1.schedule();
    }
    LOG.info("received container : "
        + alloc1Response.getAllocatedContainers().size());
    Assert.assertTrue(alloc1Response.getAllocatedContainers().size() == 1);

    rm.stop();
  }
View Full Code Here

  }
 
  @SuppressWarnings("unchecked")
  private List<Container> getResources() throws Exception {
    int headRoom = getAvailableResources() != null ? getAvailableResources().getMemory() : 0;//first time it would be null
    AMResponse response;
    /*
     * If contact with RM is lost, the AM will wait MR_AM_TO_RM_WAIT_INTERVAL_MS
     * milliseconds before aborting. During this interval, AM will still try
     * to contact the RM.
     */
    try {
      response = makeRemoteRequest();
      // Reset retry count if no exception occurred.
      retrystartTime = System.currentTimeMillis();
    } catch (Exception e) {
      // This can happen when the connection to the RM has gone down. Keep
      // re-trying until the retryInterval has expired.
      if (System.currentTimeMillis() - retrystartTime >= retryInterval) {
        LOG.error("Could not contact RM after " + retryInterval + " milliseconds.");
        eventHandler.handle(new JobEvent(this.getJob().getID(),
                                         JobEventType.INTERNAL_ERROR));
        throw new YarnException("Could not contact RM after " +
                                retryInterval + " milliseconds.");
      }
      // Throw this up to the caller, which may decide to ignore it and
      // continue to attempt to contact the RM.
      throw e;
    }
    if (response.getReboot()) {
      // This can happen if the RM has been restarted. If it is in that state,
      // this application must clean itself up.
      eventHandler.handle(new JobEvent(this.getJob().getID(),
                                       JobEventType.INTERNAL_ERROR));
      throw new YarnException("Resource Manager doesn't recognize AttemptId: " +
                               this.getContext().getApplicationID());
    }
    int newHeadRoom = getAvailableResources() != null ? getAvailableResources().getMemory() : 0;
    List<Container> newContainers = response.getAllocatedContainers();
    List<ContainerStatus> finishedContainers = response.getCompletedContainersStatuses();
    if (newContainers.size() + finishedContainers.size() > 0 || headRoom != newHeadRoom) {
      //something changed
      recalculateReduceSchedule = true;
      if (LOG.isDebugEnabled() && headRoom != newHeadRoom) {
        LOG.debug("headroom=" + newHeadRoom);
View Full Code Here

TOP

Related Classes of org.apache.hadoop.yarn.api.records.AMResponse

Copyright © 2018 www.massapicom. All rights reserved.
All source code are property of their respective owners. Java is a trademark of Sun Microsystems, Inc and owned by ORACLE Inc. Contact coftware#gmail.com.