diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-api/src/main/java/org/apache/hadoop/yarn/conf/YarnConfiguration.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-api/src/main/java/org/apache/hadoop/yarn/conf/YarnConfiguration.java index 5291ff2..e806cc8 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-api/src/main/java/org/apache/hadoop/yarn/conf/YarnConfiguration.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-api/src/main/java/org/apache/hadoop/yarn/conf/YarnConfiguration.java @@ -288,6 +288,15 @@ private static void addDeprecatedKeys() { RM_PREFIX + "am.max-attempts"; public static final int DEFAULT_RM_AM_MAX_ATTEMPTS = 2; + /** + * The minimum number of application attempts to keep in RMApp + * and RMStateStore. + * It's a global setting for all application masters. + */ + public static final String RM_AM_MIN_ATTEMPTS_TO_KEEP = + RM_PREFIX + "am.min-attempts-to-keep"; + public static final int DEFAULT_RM_AM_MIN_ATTEMPTS_TO_KEEP = 3; + /** The keytab for the resource manager.*/ public static final String RM_KEYTAB = RM_PREFIX + "keytab"; diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/ResourceManager.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/ResourceManager.java index 130cfd4..20d1bd9 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/ResourceManager.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/ResourceManager.java @@ -838,6 +838,21 @@ public void handle(RMAppAttemptEvent event) { LOG.error("Error in handling event type " + event.getType() + " for applicationAttempt " + appAttemptId, t); } + } else { + RMAppAttempt dummyAppAttempt = rmApp.getDummyAppAttempt(); + if (dummyAppAttempt != null) { + try { + LOG.info("Event handled by dummyAppAttempt: " + event); + dummyAppAttempt.handle(event); + } catch (Throwable t) { + LOG.error("Error in handling event type " + event.getType() + + " for applicationAttempt " + appAttemptId + + " with dummyAttempt", t); + } + } else { + LOG.error("Could not handle event " + event.getType() + + " for applicationAttempt " + appAttemptId); + } } } } diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/FileSystemRMStateStore.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/FileSystemRMStateStore.java index 68d26bb..6e2eb4a 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/FileSystemRMStateStore.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/FileSystemRMStateStore.java @@ -462,6 +462,18 @@ public synchronized void updateApplicationAttemptStateInternal( } @Override + public synchronized void removeApplicationAttemptInternal( + ApplicationAttemptId appAttemptId) + throws Exception { + Path appDirPath = + getAppDir(rmAppRoot, appAttemptId.getApplicationId()); + Path nodeRemovePath = getNodePath(appDirPath, appAttemptId.toString()); + LOG.info("Removing info for attempt: " + appAttemptId + " at: " + + nodeRemovePath); + deleteFileWithRetries(nodeRemovePath); + } + + @Override public synchronized void removeApplicationStateInternal( ApplicationStateData appState) throws Exception { diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/LeveldbRMStateStore.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/LeveldbRMStateStore.java index 46a3459..c9b5644 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/LeveldbRMStateStore.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/LeveldbRMStateStore.java @@ -519,6 +519,28 @@ protected void updateApplicationAttemptStateInternal( } @Override + public synchronized void removeApplicationAttemptInternal( + ApplicationAttemptId attemptId) + throws IOException { + String attemptKey = getApplicationAttemptNodeKey(attemptId); + try { + WriteBatch batch = db.createWriteBatch(); + try { + batch.delete(bytes(attemptKey)); + if (LOG.isDebugEnabled()) { + LOG.debug("Removing state for attempt " + attemptId + " at " + + attemptKey); + } + db.write(batch); + } finally { + batch.close(); + } + } catch (DBException e) { + throw new IOException(e); + } + } + + @Override protected void removeApplicationStateInternal(ApplicationStateData appState) throws IOException { ApplicationId appId = diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/MemoryRMStateStore.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/MemoryRMStateStore.java index 609f403..c85b9e8 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/MemoryRMStateStore.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/MemoryRMStateStore.java @@ -139,6 +139,19 @@ public synchronized void updateApplicationAttemptStateInternal( } @Override + public synchronized void removeApplicationAttemptInternal( + ApplicationAttemptId appAttemptId) throws Exception { + ApplicationStateData appState = + state.getApplicationState().get(appAttemptId.getApplicationId()); + ApplicationAttemptStateData attemptState = + appState.attempts.remove(appAttemptId); + LOG.info("Removing state for attempt: " + appAttemptId); + if (attemptState == null) { + throw new YarnRuntimeException("Application doesn't exist"); + } + } + + @Override public synchronized void removeApplicationStateInternal( ApplicationStateData appState) throws Exception { ApplicationId appId = diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/NullRMStateStore.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/NullRMStateStore.java index 92c07cd..d81bf50 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/NullRMStateStore.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/NullRMStateStore.java @@ -118,6 +118,12 @@ protected void updateApplicationAttemptStateInternal(ApplicationAttemptId attemp } @Override + public synchronized void removeApplicationAttemptInternal( + ApplicationAttemptId attemptId) throws Exception { + // Do nothing + } + + @Override public void checkVersion() throws Exception { // Do nothing } diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStore.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStore.java index cc4edd7..2589015 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStore.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStore.java @@ -119,6 +119,9 @@ .addTransition(RMStateStoreState.ACTIVE, RMStateStoreState.ACTIVE, RMStateStoreEventType.UPDATE_APP_ATTEMPT, new UpdateAppAttemptTransition()) .addTransition(RMStateStoreState.ACTIVE, RMStateStoreState.ACTIVE, + RMStateStoreEventType.REMOVE_APP_ATTEMPT, + new RemoveAppAttemptTransition()) + .addTransition(RMStateStoreState.ACTIVE, RMStateStoreState.ACTIVE, RMStateStoreEventType.STORE_MASTERKEY, new StoreRMDTMasterKeyTransition()) .addTransition(RMStateStoreState.ACTIVE, RMStateStoreState.ACTIVE, @@ -414,6 +417,28 @@ public void transition(RMStateStore store, RMStateStoreEvent event) { } } + private static class RemoveAppAttemptTransition implements + SingleArcTransition { + @Override + public void transition(RMStateStore store, RMStateStoreEvent event) { + if (!(event instanceof RMStateStoreRemoveAppAttemptEvent)) { + // should never happen + LOG.error("Illegal event type: " + event.getClass()); + return; + } + ApplicationAttemptId attemptId = + ((RMStateStoreRemoveAppAttemptEvent) event).getApplicationAttemptId(); + ApplicationId appId = attemptId.getApplicationId(); + LOG.info("Removing attempt " + attemptId + " from app: " + appId); + try { + store.removeApplicationAttemptInternal(attemptId); + } catch (Exception e) { + LOG.error("Error removing attempt: " + attemptId, e); + store.notifyStoreOperationFailed(e); + } + }; + } + public RMStateStore() { super(RMStateStore.class.getName()); ReentrantReadWriteLock lock = new ReentrantReadWriteLock(); @@ -798,6 +823,29 @@ public synchronized void removeApplication(RMApp app) { protected abstract void removeApplicationStateInternal( ApplicationStateData appState) throws Exception; + /** + * Non-blocking API + * ResourceManager services call this to remove an attempt from the state + * store + * This does not block the dispatcher threads + * There is no notification of completion for this operation. + */ + @SuppressWarnings("unchecked") + public synchronized void removeApplicationAttempt( + ApplicationAttemptId applicationAttemptId) { + dispatcher.getEventHandler().handle( + new RMStateStoreRemoveAppAttemptEvent(applicationAttemptId)); + } + + /** + * Blocking API + * Derived classes must implement this method to remove the state of specified + * attempt. + */ + protected abstract void removeApplicationAttemptInternal( + ApplicationAttemptId attemptId) throws Exception; + + // TODO: This should eventually become cluster-Id + "AM_RM_TOKEN_SERVICE". See // YARN-1779 public static final Text AM_RM_TOKEN_SERVICE = new Text( diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java index beba5eb..09d18a2 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java @@ -24,6 +24,7 @@ UPDATE_APP, UPDATE_APP_ATTEMPT, REMOVE_APP, + REMOVE_APP_ATTEMPT, FENCED, // Below events should be called synchronously diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreRemoveAppAttemptEvent.java similarity index 62% copy from hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java copy to hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreRemoveAppAttemptEvent.java index beba5eb..4f09f24 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreEventType.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/RMStateStoreRemoveAppAttemptEvent.java @@ -18,19 +18,20 @@ package org.apache.hadoop.yarn.server.resourcemanager.recovery; -public enum RMStateStoreEventType { - STORE_APP_ATTEMPT, - STORE_APP, - UPDATE_APP, - UPDATE_APP_ATTEMPT, - REMOVE_APP, - FENCED, +import org.apache.hadoop.yarn.api.records.ApplicationAttemptId; - // Below events should be called synchronously - STORE_MASTERKEY, - REMOVE_MASTERKEY, - STORE_DELEGATION_TOKEN, - REMOVE_DELEGATION_TOKEN, - UPDATE_DELEGATION_TOKEN, - UPDATE_AMRM_TOKEN -} +/** + * A event used to remove an attempt. + */ +public class RMStateStoreRemoveAppAttemptEvent extends RMStateStoreEvent { + private ApplicationAttemptId applicationAttemptId; + + RMStateStoreRemoveAppAttemptEvent(ApplicationAttemptId applicationAttemptId) { + super(RMStateStoreEventType.REMOVE_APP_ATTEMPT); + this.applicationAttemptId = applicationAttemptId; + } + + public ApplicationAttemptId getApplicationAttemptId() { + return applicationAttemptId; + } +} \ No newline at end of file diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/ZKRMStateStore.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/ZKRMStateStore.java index 364c970..16e20d9 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/ZKRMStateStore.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/ZKRMStateStore.java @@ -676,6 +676,21 @@ public synchronized void updateApplicationAttemptStateInternal( } @Override + public synchronized void removeApplicationAttemptInternal( + ApplicationAttemptId appAttemptId) throws Exception { + String appId = appAttemptId.getApplicationId().toString(); + String appIdRemovePath = getNodePath(rmAppRoot, appId); + String attemptIdRemovePath = getNodePath(appIdRemovePath, + appAttemptId.toString()); + + if (LOG.isDebugEnabled()) { + LOG.debug("Removing attempt info: " + appAttemptId + " at: " + + attemptIdRemovePath); + } + doDeleteMultiWithRetries(Op.delete(attemptIdRemovePath, -1)); + } + + @Override public synchronized void removeApplicationStateInternal( ApplicationStateData appState) throws Exception { diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/records/ApplicationStateData.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/records/ApplicationStateData.java index 43046a9..aeb5e0c 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/records/ApplicationStateData.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/recovery/records/ApplicationStateData.java @@ -70,6 +70,16 @@ public ApplicationAttemptStateData getAttempt( return attempts.get(attemptId); } + public int getAttemptStartId() { + int min = Integer.MAX_VALUE; + for(ApplicationAttemptId attemptId : attempts.keySet()) { + if (attemptId.getAttemptId() < min) { + min = attemptId.getAttemptId(); + } + } + return min == Integer.MAX_VALUE ? 0 : min - 1; + } + public abstract ApplicationStateDataProto getProto(); /** diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMApp.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMApp.java index be9dfaf..f56f96d 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMApp.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMApp.java @@ -114,6 +114,12 @@ RMAppAttempt getCurrentAppAttempt(); /** + * A dummy appAttempt used to handle event for those removed from attempts. + * @return a dummy appAttempt to handle event for those removed attempts. + */ + RMAppAttempt getDummyAppAttempt(); + + /** * {@link RMApp} can have multiple application attempts {@link RMAppAttempt}. * This method returns the all {@link RMAppAttempt}s for the RMApp. * @return all {@link RMAppAttempt}s for the RMApp. diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMAppImpl.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMAppImpl.java index 8abc478..5837f8a 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMAppImpl.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/RMAppImpl.java @@ -119,6 +119,8 @@ private final ApplicationMasterService masterService; private final StringBuilder diagnostics = new StringBuilder(); private final int maxAppAttempts; + // Minimum attempts to be kept in RMApp and RMStateStore + private final int minAppAttemptsToKeep; private final ReadLock readLock; private final WriteLock writeLock; private final Map attempts @@ -138,8 +140,16 @@ private long startTime; private long finishTime = 0; private long storedFinishTime = 0; + private int currentAttemptId = 0; // This field isn't protected by readlock now. private volatile RMAppAttempt currentAttempt; + /** + * For work-preserving AM restart, failed attempt are still capturing + * CONTAINER_FINISHED event and record the finished containers for the + * use by the next new attempt. It is used to handle CONTAINER_FINISHED + * event for those removed attempts. + */ + private RMAppAttempt dummyAttempt; private String queue; private EventHandler handler; private static final AppFinishedTransition FINISHED_TRANSITION = @@ -408,6 +418,9 @@ public RMAppImpl(ApplicationId applicationId, RMContext rmContext, this.maxAppAttempts = individualMaxAppAttempts; } + this.minAppAttemptsToKeep = conf.getInt( + YarnConfiguration.RM_AM_MIN_ATTEMPTS_TO_KEEP, + YarnConfiguration.DEFAULT_RM_AM_MIN_ATTEMPTS_TO_KEEP); this.attemptFailuresValidityInterval = submissionContext.getAttemptFailuresValidityInterval(); if (this.attemptFailuresValidityInterval > 0) { @@ -524,6 +537,16 @@ public RMAppAttempt getCurrentAppAttempt() { } @Override + public RMAppAttempt getDummyAppAttempt() { + this.readLock.lock(); + try { + return this.dummyAttempt; + } finally { + this.readLock.unlock(); + } + } + + @Override public Map getAppAttempts() { this.readLock.lock(); @@ -765,17 +788,21 @@ public void recover(RMState state) { this.diagnostics.append(appState.getDiagnostics()); this.storedFinishTime = appState.getFinishTime(); this.startTime = appState.getStartTime(); + this.currentAttemptId = appState.getAttemptStartId(); for(int i=0; i endTime - - this.attemptFailuresValidityInterval)) { - completedAttempts++; - } + if (isFailedAppAttempt(attempt)) { + completedAttempts++; } } return completedAttempts; } + private boolean isFailedAppAttempt(RMAppAttempt attempt) { + long endTime = this.systemClock.getTime(); + if (attempt.shouldCountTowardsMaxAttemptRetry()) { + if (this.attemptFailuresValidityInterval <= 0 + || (attempt.getFinishTime() > endTime + - this.attemptFailuresValidityInterval)) { + return true; + } + } + return false; + } + private static final class AttemptFailedTransition implements MultipleArcTransition { @@ -1274,6 +1306,9 @@ public RMAppState transition(RMAppImpl app, RMAppEvent event) { // but when pulling finished container we will check this flag again. ((RMAppAttemptImpl) app.currentAttempt) .transferStateFromAttempt(oldAttempt); + + removeAttemptsBeyondValidityInterval(app); + return initialState; } else { if (numberOfFailure >= app.maxAppAttempts) { @@ -1285,6 +1320,29 @@ public RMAppState transition(RMAppImpl app, RMAppEvent event) { return RMAppState.FINAL_SAVING; } } + + private void removeAttemptsBeyondValidityInterval(RMAppImpl app) { + while (app.attempts.size() > app.minAppAttemptsToKeep) { + // attempts' first element is oldest attempt because it is a + // LinkedHashMap + RMAppAttempt oldestAttempt = app.attempts.values().iterator().next(); + if (app.isFailedAppAttempt(oldestAttempt)) { + break; + } + removeAppAttempt(app, oldestAttempt); + } + } + + // Remove the attempt in app.attempts, and set dummyAttempt if needed. + private void removeAppAttempt(RMAppImpl app, RMAppAttempt appAttempt) { + ApplicationAttemptId attemptId = appAttempt.getAppAttemptId(); + LOG.info("Remove oldest attempt in app.attempts : " + attemptId); + if (app.dummyAttempt == null) { + app.dummyAttempt = app.attempts.get(attemptId); + } + app.attempts.remove(attemptId); + app.rmContext.getStateStore().removeApplicationAttempt(attemptId); + } } @Override diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestRMRestart.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestRMRestart.java index c889446..8e28ddc 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestRMRestart.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestRMRestart.java @@ -491,6 +491,7 @@ public void testRMRestartWaitForPreviousAMToFinish() throws Exception { // be started immediately. YarnConfiguration conf = new YarnConfiguration(this.conf); conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 40); + conf.setInt(YarnConfiguration.RM_AM_MIN_ATTEMPTS_TO_KEEP, 40); MemoryRMStateStore memStore = new MemoryRMStateStore(); memStore.init(conf); RMState rmState = memStore.getState(); diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/MockAsm.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/MockAsm.java index a23c789..0cc58f7 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/MockAsm.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/MockAsm.java @@ -102,6 +102,10 @@ public RMAppAttempt getCurrentAppAttempt() { throw new UnsupportedOperationException("Not supported yet."); } @Override + public RMAppAttempt getDummyAppAttempt() { + throw new UnsupportedOperationException("Not supported yet."); + } + @Override public Map getAppAttempts() { throw new UnsupportedOperationException("Not supported yet."); } diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/TestAMRestart.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/TestAMRestart.java index 7befba4..b2826a5 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/TestAMRestart.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/applicationsmanager/TestAMRestart.java @@ -364,6 +364,7 @@ public void testShouldNotCountFailureToMaxAttemptRetry() throws Exception { ResourceScheduler.class); // explicitly set max-am-retry count as 1. conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 1); + conf.setInt(YarnConfiguration.RM_AM_MIN_ATTEMPTS_TO_KEEP, 3); conf.setBoolean(YarnConfiguration.RECOVERY_ENABLED, true); conf.set(YarnConfiguration.RM_STORE, MemoryRMStateStore.class.getName()); MemoryRMStateStore memStore = new MemoryRMStateStore(); @@ -427,7 +428,7 @@ public void testShouldNotCountFailureToMaxAttemptRetry() throws Exception { rm1.waitForState(app1.getApplicationId(), RMAppState.ACCEPTED); MockAM am4 = - rm1.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 4, nm1); + rm1.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 3, nm1); RMAppAttempt attempt4 = app1.getCurrentAppAttempt(); Assert.assertTrue(((RMAppAttemptImpl) attempt4).mayBeLastAttempt()); @@ -446,7 +447,7 @@ public void testShouldNotCountFailureToMaxAttemptRetry() throws Exception { // launch next AM in nm2 nm2.nodeHeartbeat(true); MockAM am5 = - rm1.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 5, nm2); + rm1.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 3, nm2); RMAppAttempt attempt5 = app1.getCurrentAppAttempt(); Assert.assertTrue(((RMAppAttemptImpl) attempt5).mayBeLastAttempt()); // fail the AM normally @@ -457,7 +458,7 @@ public void testShouldNotCountFailureToMaxAttemptRetry() throws Exception { // AM should not be restarted. rm1.waitForState(app1.getApplicationId(), RMAppState.FAILED); - Assert.assertEquals(5, app1.getAppAttempts().size()); + Assert.assertEquals(3, app1.getAppAttempts().size()); rm1.stop(); } @@ -603,6 +604,7 @@ public void testRMAppAttemptFailuresValidityInterval() throws Exception { conf.set(YarnConfiguration.RM_STORE, MemoryRMStateStore.class.getName()); // explicitly set max-am-retry count as 2. conf.setInt(YarnConfiguration.RM_AM_MAX_ATTEMPTS, 2); + conf.setInt(YarnConfiguration.RM_AM_MIN_ATTEMPTS_TO_KEEP, 3); MemoryRMStateStore memStore = new MemoryRMStateStore(); memStore.init(conf); @@ -691,7 +693,7 @@ public void testRMAppAttemptFailuresValidityInterval() throws Exception { // Lauch Attempt 4 MockAM am4 = - rm2.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 4, nm1); + rm2.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 3, nm1); // wait for 6 seconds clock.setTime(System.currentTimeMillis() + 6*1000); @@ -704,7 +706,7 @@ public void testRMAppAttemptFailuresValidityInterval() throws Exception { rm2.waitForState(app1.getApplicationId(), RMAppState.ACCEPTED); MockAM am5 = - rm2.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 5, nm1); + rm2.waitForNewAMToLaunchAndRegister(app1.getApplicationId(), 3, nm1); clock.reset(); am5.waitForState(RMAppAttemptState.RUNNING); diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/MockRMApp.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/MockRMApp.java index c6ee3ba..0387f69 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/MockRMApp.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/rmapp/MockRMApp.java @@ -147,6 +147,11 @@ public RMAppAttempt getCurrentAppAttempt() { return attempt; } + @Override + public RMAppAttempt getDummyAppAttempt() { + throw new UnsupportedOperationException("Not supported yet."); + } + public void setCurrentAppAttempt(RMAppAttempt attempt) { this.attempt = attempt; }