From d5b71e4175c13679d451710be150fc461a661263 Mon Sep 17 00:00:00 2001 From: Daniel Templeton Date: Wed, 31 May 2017 15:48:04 -0700 Subject: [PATCH] YARN-6246. Identifying starved apps does not need the scheduler writelock (Contributed by Karthik Kambatla via Daniel Templeton) --- .../scheduler/fair/FSLeafQueue.java | 9 +++---- .../scheduler/fair/FSParentQueue.java | 4 +-- .../scheduler/fair/FSQueue.java | 19 ++++++++----- .../scheduler/fair/FairScheduler.java | 27 +++++++++++++------ 4 files changed, 38 insertions(+), 21 deletions(-) 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/FSLeafQueue.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/FSLeafQueue.java index 10f1e287f4..1de0e302de 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/FSLeafQueue.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/FSLeafQueue.java @@ -198,13 +198,10 @@ public void collectSchedulerApplications( } @Override - public void updateInternal(boolean checkStarvation) { + void updateInternal() { readLock.lock(); try { policy.computeShares(runnableApps, getFairShare()); - if (checkStarvation) { - updateStarvedApps(); - } } finally { readLock.unlock(); } @@ -283,8 +280,10 @@ private void updateStarvedAppsMinshare( * If this queue is starving due to fairshare, there must be at least * one application that is starved. And, even if the queue is not * starved due to fairshare, there might still be starved applications. + * + * Caller does not need read/write lock on the leaf queue. */ - private void updateStarvedApps() { + void updateStarvedApps() { // Fetch apps with pending demand TreeSet appsWithDemand = fetchAppsWithDemand(false); 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/FSParentQueue.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/FSParentQueue.java index b062c586d0..5b4e4dc9e5 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/FSParentQueue.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/FSParentQueue.java @@ -79,13 +79,13 @@ void removeChildQueue(FSQueue child) { } @Override - public void updateInternal(boolean checkStarvation) { + void updateInternal() { readLock.lock(); try { policy.computeShares(childQueues, getFairShare()); for (FSQueue childQueue : childQueues) { childQueue.getMetrics().setFairShare(childQueue.getFairShare()); - childQueue.updateInternal(checkStarvation); + childQueue.updateInternal(); } } finally { readLock.unlock(); 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/FSQueue.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/FSQueue.java index e131140702..12b1b83f1a 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/FSQueue.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/FSQueue.java @@ -326,16 +326,23 @@ public boolean isPreemptable() { /** * Recomputes the shares for all child queues and applications based on this - * queue's current share, and checks for starvation. + * queue's current share. * - * @param checkStarvation whether to check for fairshare or minshare - * starvation on update + * To be called holding the scheduler writelock. */ - abstract void updateInternal(boolean checkStarvation); + abstract void updateInternal(); - public void update(Resource fairShare, boolean checkStarvation) { + /** + * Set the queue's fairshare and update the demand/fairshare of child + * queues/applications. + * + * To be called holding the scheduler writelock. + * + * @param fairShare + */ + public void update(Resource fairShare) { setFairShare(fairShare); - updateInternal(checkStarvation); + updateInternal(); } /** 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 d1a237ada9..d779159084 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 @@ -366,20 +366,31 @@ private void dumpSchedulerState() { */ @VisibleForTesting public void update() { + FSQueue rootQueue = queueMgr.getRootQueue(); + + // Update demands and fairshares + writeLock.lock(); try { - writeLock.lock(); - - FSQueue rootQueue = queueMgr.getRootQueue(); - // Recursively update demands for all queues rootQueue.updateDemand(); - - Resource clusterResource = getClusterResource(); - rootQueue.update(clusterResource, shouldAttemptPreemption()); + rootQueue.update(getClusterResource()); // Update metrics updateRootQueueMetrics(); + } finally { + writeLock.unlock(); + } + readLock.lock(); + try { + // Update starvation stats and identify starved applications + if (shouldAttemptPreemption()) { + for (FSLeafQueue queue : queueMgr.getLeafQueues()) { + queue.updateStarvedApps(); + } + } + + // Log debug information if (LOG.isDebugEnabled()) { if (--updatesToSkipForDebug < 0) { updatesToSkipForDebug = UPDATE_DEBUG_FREQUENCY; @@ -387,7 +398,7 @@ public void update() { } } } finally { - writeLock.unlock(); + readLock.unlock(); } }