diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FSAppAttempt.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FSAppAttempt.java index 28c599ebdfb..8ff688e0f2d 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FSAppAttempt.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FSAppAttempt.java @@ -82,8 +82,10 @@ public class FSAppAttempt extends SchedulerApplicationAttempt private Resource fairShare = Resources.createResource(0, 0); // Preemption related variables + private final Object preemptionVariablesLock = new Object(); private final Resource preemptedResources = Resources.clone(Resources.none()); private final Set containersToPreempt = new HashSet<>(); + private Resource fairshareStarvation = Resources.none(); private long lastTimeAtFairShare; private long nextStarvationCheck; @@ -551,29 +553,29 @@ public class FSAppAttempt extends SchedulerApplicationAttempt } void trackContainerForPreemption(RMContainer container) { - if (containersToPreempt.add(container)) { - synchronized (preemptedResources) { + synchronized (preemptionVariablesLock) { + if (containersToPreempt.add(container)) { Resources.addTo(preemptedResources, container.getAllocatedResource()); } } } private void untrackContainerForPreemption(RMContainer container) { - if (containersToPreempt.remove(container)) { - synchronized (preemptedResources) { + synchronized (preemptionVariablesLock) { + if (containersToPreempt.remove(container)) { Resources.subtractFrom(preemptedResources, container.getAllocatedResource()); } } } - Set getPreemptionContainers() { - return containersToPreempt; - } - - private Resource getPreemptedResources() { - synchronized (preemptedResources) { - return preemptedResources; + Set getPreemptionContainerIds() { + synchronized (preemptionVariablesLock) { + Set preemptionContainerIds = new HashSet<>(); + for (RMContainer container : containersToPreempt) { + preemptionContainerIds.add(container.getContainerId()); + } + return preemptionContainerIds; } } @@ -590,9 +592,11 @@ public class FSAppAttempt extends SchedulerApplicationAttempt return false; } - if (containersToPreempt.contains(container)) { - // The container is already under consideration for preemption - return false; + synchronized (preemptionVariablesLock) { + if (containersToPreempt.contains(container)) { + // The container is already under consideration for preemption + return false; + } } // Check if the app's allocation will be over its fairshare even @@ -964,7 +968,8 @@ public class FSAppAttempt extends SchedulerApplicationAttempt if (LOG.isTraceEnabled()) { LOG.trace("Assign container on " + node.getNodeName() + " node, assignType: OFF_SWITCH" + ", allowedLocality: " - + allowedLocality + ", priority: " + schedulerKey.getPriority() + + allowedLocality + ", priority: " + + schedulerKey.getPriority() + ", app attempt id: " + this.attemptId); } return assignContainer(node, offSwitchRequest, NodeType.OFF_SWITCH, @@ -1219,13 +1224,13 @@ public class FSAppAttempt extends SchedulerApplicationAttempt @Override public Resource getResourceUsage() { - /* - * getResourcesToPreempt() returns zero, except when there are containers - * to preempt. Avoid creating an object in the common case. - */ - return getPreemptedResources().equals(Resources.none()) - ? getCurrentConsumption() - : Resources.subtract(getCurrentConsumption(), getPreemptedResources()); + // Subtract copies the object, so that we have a snapshot, + // in case usage changes, while the caller is using the value + synchronized (preemptionVariablesLock) { + return containersToPreempt.isEmpty() + ? getCurrentConsumption() + : Resources.subtract(getCurrentConsumption(), preemptedResources); + } } @Override diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java index f21b5f1542d..ae13b236719 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/FairScheduler.java @@ -32,6 +32,7 @@ import java.util.Map; import java.util.Map.Entry; import java.util.Set; import java.util.concurrent.ConcurrentHashMap; +import java.util.concurrent.locks.ReentrantReadWriteLock; import org.apache.commons.logging.Log; import org.apache.commons.logging.LogFactory; @@ -831,8 +832,9 @@ public class FairScheduler extends // Release containers releaseContainers(release, application); + ReentrantReadWriteLock.WriteLock lock = application.getWriteLock(); + lock.lock(); try { - application.getWriteLock().lock(); if (!ask.isEmpty()) { if (LOG.isDebugEnabled()) { LOG.debug( @@ -847,24 +849,21 @@ public class FairScheduler extends application.showRequests(); } } finally { - application.getWriteLock().unlock(); + lock.unlock(); } + Set preemptionContainerIds = + application.getPreemptionContainerIds(); if (LOG.isDebugEnabled()) { LOG.debug( "allocate: post-update" + " applicationAttemptId=" + appAttemptId + " #ask=" + ask.size() + " reservation= " + application .getCurrentReservation()); - LOG.debug("Preempting " + application.getPreemptionContainers().size() + LOG.debug("Preempting " + preemptionContainerIds.size() + " container(s)"); } - Set preemptionContainerIds = new HashSet(); - for (RMContainer container : application.getPreemptionContainers()) { - preemptionContainerIds.add(container.getContainerId()); - } - application.updateBlacklist(blacklistAdditions, blacklistRemovals); List newlyAllocatedContainers =