Apache
/
hadoop
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

YARN-6959. RM may allocate wrong AM Container for new attempt. Contributed by Yuqi Wang

This commit is contained in:
Jian He 2017-08-18 13:08:54 -07:00
parent 2810e6ab9c
commit 3f735ad640
5 changed files with 63 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/AbstractYarnScheduler.java
View File

@ -288,6 +288,7 @@ public abstract class AbstractYarnScheduler
}
}
// TODO: Rename it to getCurrentApplicationAttempt
public T getApplicationAttempt(ApplicationAttemptId applicationAttemptId) {
SchedulerApplication<T> app =
applications.get(applicationAttemptId.getApplicationId());

13
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/capacity/CapacityScheduler.java
View File

@ -948,6 +948,19 @@ public class CapacityScheduler extends
FiCaSchedulerApp application = getApplicationAttempt(applicationAttemptId);
if (application == null) {
LOG.error("Calling allocate on removed or non existent application " +
applicationAttemptId.getApplicationId());
return EMPTY_ALLOCATION;
}
// The allocate may be the leftover from previous attempt, and it will
// impact current attempt, such as confuse the request and allocation for
// current attempt's AM container.
// Note outside precondition check for the attempt id may be
// outdated here, so double check it here is necessary.
if (!application.getApplicationAttemptId().equals(applicationAttemptId)) {
LOG.error("Calling allocate on previous or removed " +
"or non existent application attempt " + applicationAttemptId);
return EMPTY_ALLOCATION;
}

15
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
View File

@ -962,8 +962,19 @@ public class FairScheduler extends
// Make sure this application exists
FSAppAttempt application = getSchedulerApp(appAttemptId);
if (application == null) {
LOG.info("Calling allocate on removed " +
"or non existant application " + appAttemptId);
LOG.error("Calling allocate on removed or non existent application " +
appAttemptId.getApplicationId());
return EMPTY_ALLOCATION;
}
// The allocate may be the leftover from previous attempt, and it will
// impact current attempt, such as confuse the request and allocation for
// current attempt's AM container.
// Note outside precondition check for the attempt id may be
// outdated here, so double check it here is necessary.
if (!application.getApplicationAttemptId().equals(appAttemptId)) {
LOG.error("Calling allocate on previous or removed " +
"or non existent application attempt " + appAttemptId);
return EMPTY_ALLOCATION;
}

15
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/main/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fifo/FifoScheduler.java
View File

@ -332,8 +332,19 @@ public class FifoScheduler extends
List<UpdateContainerRequest> decreaseRequests) {
FiCaSchedulerApp application = getApplicationAttempt(applicationAttemptId);
if (application == null) {
LOG.error("Calling allocate on removed " +
"or non existant application " + applicationAttemptId);
LOG.error("Calling allocate on removed or non existent application " +
applicationAttemptId.getApplicationId());
return EMPTY_ALLOCATION;
}
// The allocate may be the leftover from previous attempt, and it will
// impact current attempt, such as confuse the request and allocation for
// current attempt's AM container.
// Note outside precondition check for the attempt id may be
// outdated here, so double check it here is necessary.
if (!application.getApplicationAttemptId().equals(applicationAttemptId)) {
LOG.error("Calling allocate on previous or removed " +
"or non existent application attempt " + applicationAttemptId);
return EMPTY_ALLOCATION;
}

46
hadoop-yarn-project/hadoop-yarn/hadoop-yarn-server/hadoop-yarn-server-resourcemanager/src/test/java/org/apache/hadoop/yarn/server/resourcemanager/scheduler/fair/TestFairScheduler.java
View File

@ -1680,45 +1680,45 @@ public class TestFairScheduler extends FairSchedulerTestBase {
scheduler.init(conf);
scheduler.start();
scheduler.reinitialize(conf, resourceManager.getRMContext());
int minReqSize =
FairSchedulerConfiguration.DEFAULT_RM_SCHEDULER_INCREMENT_ALLOCATION_MB;
// First ask, queue1 requests 1 large (minReqSize * 2).
ApplicationAttemptId id11 = createAppAttemptId(1, 1);
createMockRMApp(id11);
scheduler.addApplication(id11.getApplicationId(), "root.queue1", "user1", false);
scheduler.addApplication(id11.getApplicationId(),
"root.queue1", "user1", false);
scheduler.addApplicationAttempt(id11, false, false);
ApplicationAttemptId id21 = createAppAttemptId(2, 1);
createMockRMApp(id21);
scheduler.addApplication(id21.getApplicationId(), "root.queue2", "user1", false);
scheduler.addApplicationAttempt(id21, false, false);
ApplicationAttemptId id22 = createAppAttemptId(2, 2);
createMockRMApp(id22);
scheduler.addApplication(id22.getApplicationId(), "root.queue2", "user1", false);
scheduler.addApplicationAttempt(id22, false, false);
int minReqSize =
FairSchedulerConfiguration.DEFAULT_RM_SCHEDULER_INCREMENT_ALLOCATION_MB;
// First ask, queue1 requests 1 large (minReqSize * 2).
List<ResourceRequest> ask1 = new ArrayList<ResourceRequest>();
ResourceRequest request1 =
createResourceRequest(minReqSize * 2, ResourceRequest.ANY, 1, 1, true);
ResourceRequest request1 = createResourceRequest(minReqSize * 2,
ResourceRequest.ANY, 1, 1, true);
ask1.add(request1);
scheduler.allocate(id11, ask1, new ArrayList<ContainerId>(), null, null, null, null);
// Second ask, queue2 requests 1 large + (2 * minReqSize)
ApplicationAttemptId id21 = createAppAttemptId(2, 1);
createMockRMApp(id21);
scheduler.addApplication(id21.getApplicationId(),
"root.queue2", "user1", false);
scheduler.addApplicationAttempt(id21, false, false);
List<ResourceRequest> ask2 = new ArrayList<ResourceRequest>();
ResourceRequest request2 = createResourceRequest(2 * minReqSize, "foo", 1, 1,
false);
ResourceRequest request3 = createResourceRequest(minReqSize, "bar", 1, 2,
false);
ResourceRequest request2 = createResourceRequest(2 * minReqSize,
"foo", 1, 1, false);
ResourceRequest request3 = createResourceRequest(minReqSize,
ResourceRequest.ANY, 1, 2, false);
ask2.add(request2);
ask2.add(request3);
scheduler.allocate(id21, ask2, new ArrayList<ContainerId>(), null, null, null, null);
// Third ask, queue2 requests 1 large
ApplicationAttemptId id22 = createAppAttemptId(2, 2);
createMockRMApp(id22);
scheduler.addApplication(id22.getApplicationId(),
"root.queue2", "user1", false);
scheduler.addApplicationAttempt(id22, false, false);
List<ResourceRequest> ask3 = new ArrayList<ResourceRequest>();
ResourceRequest request4 =
createResourceRequest(2 * minReqSize, ResourceRequest.ANY, 1, 1, true);
ResourceRequest request4 = createResourceRequest(2 * minReqSize,
ResourceRequest.ANY, 1, 1, true);
ask3.add(request4);
scheduler.allocate(id22, ask3, new ArrayList<ContainerId>(), null, null, null, null);