Examples with RMContainerRecoverEvent org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainerRecoverEvent used on opensource projects

Example 1 with RMContainerRecoverEvent

use of org.apache.hadoop.yarn.server.resourcemanager.rmcontainer.RMContainerRecoverEvent in project hadoop by apache.

the class AbstractYarnScheduler method recoverContainersOnNode.

public void recoverContainersOnNode(List<NMContainerStatus> containerReports, RMNode nm) {
    try {
        writeLock.lock();
        if (!rmContext.isWorkPreservingRecoveryEnabled() || containerReports == null || (containerReports != null && containerReports.isEmpty())) {
            return;
        }
        for (NMContainerStatus container : containerReports) {
            ApplicationId appId = container.getContainerId().getApplicationAttemptId().getApplicationId();
            RMApp rmApp = rmContext.getRMApps().get(appId);
            if (rmApp == null) {
                LOG.error("Skip recovering container " + container + " for unknown application.");
                killOrphanContainerOnNode(nm, container);
                continue;
            }
            SchedulerApplication<T> schedulerApp = applications.get(appId);
            if (schedulerApp == null) {
                LOG.info("Skip recovering container  " + container + " for unknown SchedulerApplication. " + "Application current state is " + rmApp.getState());
                killOrphanContainerOnNode(nm, container);
                continue;
            }
            LOG.info("Recovering container " + container);
            SchedulerApplicationAttempt schedulerAttempt = schedulerApp.getCurrentAppAttempt();
            if (!rmApp.getApplicationSubmissionContext().getKeepContainersAcrossApplicationAttempts()) {
                // Do not recover containers for stopped attempt or previous attempt.
                if (schedulerAttempt.isStopped() || !schedulerAttempt.getApplicationAttemptId().equals(container.getContainerId().getApplicationAttemptId())) {
                    LOG.info("Skip recovering container " + container + " for already stopped attempt.");
                    killOrphanContainerOnNode(nm, container);
                    continue;
                }
            }
            // create container
            RMContainer rmContainer = recoverAndCreateContainer(container, nm);
            // recover RMContainer
            rmContainer.handle(new RMContainerRecoverEvent(container.getContainerId(), container));
            // recover scheduler node
            SchedulerNode schedulerNode = nodeTracker.getNode(nm.getNodeID());
            schedulerNode.recoverContainer(rmContainer);
            // recover queue: update headroom etc.
            Queue queue = schedulerAttempt.getQueue();
            queue.recoverContainer(getClusterResource(), schedulerAttempt, rmContainer);
            // recover scheduler attempt
            schedulerAttempt.recoverContainer(schedulerNode, rmContainer);
            // set master container for the current running AMContainer for this
            // attempt.
            RMAppAttempt appAttempt = rmApp.getCurrentAppAttempt();
            if (appAttempt != null) {
                Container masterContainer = appAttempt.getMasterContainer();
                // container ID stored in AppAttempt.
                if (masterContainer != null && masterContainer.getId().equals(rmContainer.getContainerId())) {
                    ((RMContainerImpl) rmContainer).setAMContainer(true);
                }
            }
            if (schedulerAttempt.getPendingRelease().remove(container.getContainerId())) {
                // release the container
                rmContainer.handle(new RMContainerFinishedEvent(container.getContainerId(), SchedulerUtils.createAbnormalContainerStatus(container.getContainerId(), SchedulerUtils.RELEASED_CONTAINER), RMContainerEventType.RELEASED));
                LOG.info(container.getContainerId() + " is released by application.");
            }
        }
    } finally {
        writeLock.unlock();
    }
}