[SPARK-19649] Spark YARN client throws exception if job succeeds and max-completed-applications=0

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Won't Fix
• Affects Version/s: 1.6.3
• Component/s: Spark Core, YARN

Environment

EMR release label 4.8.x

• Assignee: Unassigned
• Reporter: Joshua Caplan
• Created: 17/Feb/17 22:56
• Updated: 17/May/20 18:14
• Resolved: 12/Feb/19 21:32

Description

I believe the patch in SPARK-3877 created a new race condition between YARN and the Spark client.

I typically configure YARN not to keep any recent jobs in memory, as some of my jobs get pretty large.

yarn-site	yarn.resourcemanager.max-completed-applications	0

The once-per-second call to getApplicationReport may thus encounter a RUNNING application followed by a not found application, and report a false negative.

(typical) Executor log:

17/01/09 19:31:23 INFO ApplicationMaster: Final app status: SUCCEEDED, exitCode: 0
17/01/09 19:31:23 INFO SparkContext: Invoking stop() from shutdown hook
17/01/09 19:31:24 INFO SparkUI: Stopped Spark web UI at http://10.0.0.168:37046
17/01/09 19:31:24 INFO YarnClusterSchedulerBackend: Shutting down all executors
17/01/09 19:31:24 INFO YarnClusterSchedulerBackend: Asking each executor to shut down
17/01/09 19:31:24 INFO MapOutputTrackerMasterEndpoint: MapOutputTrackerMasterEndpoint stopped!
17/01/09 19:31:24 INFO MemoryStore: MemoryStore cleared
17/01/09 19:31:24 INFO BlockManager: BlockManager stopped
17/01/09 19:31:24 INFO BlockManagerMaster: BlockManagerMaster stopped
17/01/09 19:31:24 INFO OutputCommitCoordinator$OutputCommitCoordinatorEndpoint: OutputCommitCoordinator stopped!
17/01/09 19:31:24 INFO SparkContext: Successfully stopped SparkContext
17/01/09 19:31:24 INFO ApplicationMaster: Unregistering ApplicationMaster with SUCCEEDED
17/01/09 19:31:24 INFO RemoteActorRefProvider$RemotingTerminator: Shutting down remote daemon.
17/01/09 19:31:24 INFO RemoteActorRefProvider$RemotingTerminator: Remote daemon shut down; proceeding with flushing remote transports.
17/01/09 19:31:24 INFO AMRMClientImpl: Waiting for application to be successfully unregistered.
17/01/09 19:31:24 INFO RemoteActorRefProvider$RemotingTerminator: Remoting shut down.

Client log:

17/01/09 19:31:23 INFO Client: Application report for application_1483983939941_0056 (state: RUNNING)
17/01/09 19:31:24 ERROR Client: Application application_1483983939941_0056 not found.
Exception in thread "main" org.apache.spark.SparkException: Application application_1483983939941_0056 is killed
	at org.apache.spark.deploy.yarn.Client.run(Client.scala:1038)
	at org.apache.spark.deploy.yarn.Client$.main(Client.scala:1081)
	at org.apache.spark.deploy.yarn.Client.main(Client.scala)
	at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
	at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
	at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
	at java.lang.reflect.Method.invoke(Method.java:606)
	at org.apache.spark.deploy.SparkSubmit$.org$apache$spark$deploy$SparkSubmit$$runMain(SparkSubmit.scala:731)
	at org.apache.spark.deploy.SparkSubmit$.doRunMain$1(SparkSubmit.scala:181)
	at org.apache.spark.deploy.SparkSubmit$.submit(SparkSubmit.scala:206)
	at org.apache.spark.deploy.SparkSubmit$.main(SparkSubmit.scala:121)
	at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)

Issue Links

• is broken by: SPARK-3877 The exit code of spark-submit is still 0 when an yarn application fails
• is related to: MAPREDUCE-6091 YARNRunner.getJobStatus() fails with ApplicationNotFoundException if the job rolled off the RM view