From 025c6565725c1819566377632753e8b9055617a6 Mon Sep 17 00:00:00 2001 From: Robert Kanter Date: Tue, 24 Oct 2017 10:21:44 -0700 Subject: [PATCH] YARN-7382. NoSuchElementException in FairScheduler after failover causes RM crash (rkanter) --- .../scheduler/fair/FSAppAttempt.java | 10 +++++++++ .../TestWorkPreservingRMRestart.java | 21 ++++++++++++++++--- 2 files changed, 28 insertions(+), 3 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/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 8ab6e136973..157d2640530 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 @@ -658,6 +658,16 @@ public class FSAppAttempt extends SchedulerApplicationAttempt if (!rmContainer.getState().equals(RMContainerState.COMPLETED)) { getQueue().incUsedResource(rmContainer.getContainer().getResource()); } + + // If not running unmanaged, the first container we recover is always + // the AM. Set the amResource for this app and update the leaf queue's AM + // usage + if (!isAmRunning() && !getUnmanagedAM()) { + Resource resource = rmContainer.getAllocatedResource(); + setAMResource(resource); + getQueue().addAMResourceUsage(resource); + setAmRunning(true); + } } finally { writeLock.unlock(); } diff --git a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestWorkPreservingRMRestart.java b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestWorkPreservingRMRestart.java index c1cb4c1b309..064e2174e2f 100644 --- a/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestWorkPreservingRMRestart.java +++ b/hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/TestWorkPreservingRMRestart.java @@ -66,6 +66,7 @@ import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.LeafQueu import org.apache.hadoop.yarn.server.resourcemanager.scheduler.capacity.ParentQueue; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FSAppAttempt; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FSParentQueue; +import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FSQueueMetrics; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairScheduler; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairSchedulerConfiguration; import org.apache.hadoop.yarn.server.resourcemanager.scheduler.fair.FairSchedulerTestBase; @@ -158,6 +159,7 @@ public class TestWorkPreservingRMRestart extends ParameterizedSchedulerTestBase new MockNM("127.0.0.1:1234", 8192, rm1.getResourceTrackerService()); nm1.registerNode(); RMApp app1 = rm1.submitApp(200); + Resource amResources = app1.getAMResourceRequests().get(0).getCapability(); MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1); // clear queue metrics @@ -240,7 +242,8 @@ public class TestWorkPreservingRMRestart extends ParameterizedSchedulerTestBase if (getSchedulerType() == SchedulerType.CAPACITY) { checkCSQueue(rm2, schedulerApp, nmResource, nmResource, usedResources, 2); } else { - checkFSQueue(rm2, schedulerApp, usedResources, availableResources); + checkFSQueue(rm2, schedulerApp, usedResources, availableResources, + amResources); } // *********** check scheduler attempt state.******** @@ -310,6 +313,7 @@ public class TestWorkPreservingRMRestart extends ParameterizedSchedulerTestBase RMApp app1 = rm1.submitApp(200, "dynamicQApp", UserGroupInformation.getCurrentUser().getShortUserName(), null, ReservationSystemTestUtil.getReservationQueueName()); + Resource amResources = app1.getAMResourceRequests().get(0).getCapability(); MockAM am1 = MockRM.launchAndRegisterAM(app1, rm1, nm1); // clear queue metrics @@ -384,7 +388,8 @@ public class TestWorkPreservingRMRestart extends ParameterizedSchedulerTestBase if (getSchedulerType() == SchedulerType.CAPACITY) { checkCSQueue(rm2, schedulerApp, nmResource, nmResource, usedResources, 2); } else { - checkFSQueue(rm2, schedulerApp, usedResources, availableResources); + checkFSQueue(rm2, schedulerApp, usedResources, availableResources, + amResources); } // *********** check scheduler attempt state.******** @@ -456,7 +461,7 @@ public class TestWorkPreservingRMRestart extends ParameterizedSchedulerTestBase private void checkFSQueue(ResourceManager rm, SchedulerApplication schedulerApp, Resource usedResources, - Resource availableResources) throws Exception { + Resource availableResources, Resource amResources) throws Exception { // waiting for RM's scheduling apps int retry = 0; Resource assumedFairShare = Resource.newInstance(8192, 8); @@ -488,6 +493,16 @@ public class TestWorkPreservingRMRestart extends ParameterizedSchedulerTestBase assertMetrics(queueMetrics, 1, 0, 1, 0, 2, availableResources.getMemorySize(), availableResources.getVirtualCores(), usedResources.getMemorySize(), usedResources.getVirtualCores()); + + // ************ check AM resources **************** + assertEquals(amResources, + schedulerApp.getCurrentAppAttempt().getAMResource()); + FSQueueMetrics fsQueueMetrics = + (FSQueueMetrics) schedulerApp.getQueue().getMetrics(); + assertEquals(amResources.getMemorySize(), + fsQueueMetrics.getAMResourceUsageMB()); + assertEquals(amResources.getVirtualCores(), + fsQueueMetrics.getAMResourceUsageVCores()); } // create 3 container reports for AM