Maheedhar Reddy Chappidi created SPARK-19226:
------------------------------------------------
Summary: Report failure reason from Reporter Thread
Key: SPARK-19226
URL: https://issues.apache.org/jira/browse/SPARK-19226
Project: Spark
Issue Type: Improvement
Components: YARN
Affects Versions: 2.0.2
Environment: emr-5.2.1 with Zeppelin 0.6.2/Spark2.0.2 and 10 r3.xl
core nodes
Reporter: Maheedhar Reddy Chappidi
Priority: Minor
With the exponential[1] increase in executor count the Reporter thread [2]
fails without proper message.
==
17/01/12 09:33:44 INFO YarnAllocator: Driver requested a total number of 32767
executor(s).
17/01/12 09:33:44 INFO YarnAllocator: Will request 24576 executor containers,
each with 2 cores and 5632 MB memory including 512 MB overhead
17/01/12 09:33:44 INFO YarnAllocator: Canceled 0 container requests (locality
no longer needed)
17/01/12 09:33:52 INFO YarnAllocator: Driver requested a total number of 34419
executor(s).
17/01/12 09:33:52 INFO ApplicationMaster: Final app status: FAILED, exitCode:
12, (reason: Exception was thrown 1 time(s) from Reporter thread.)
17/01/12 09:33:52 INFO YarnAllocator: Driver requested a total number of 34410
executor(s).
17/01/12 09:33:52 INFO YarnAllocator: Driver requested a total number of 34409
executor(s).
17/01/12 09:33:52 INFO ShutdownHookManager: Shutdown hook called
==
We were able to run the workflows by setting/limiting the maxExecutor count
(spark.dynamicAllocation.maxExecutors) to avoid more requests(35k->65k).
Added I don't see any issues with ApplicationMaster's container memory/compute.
[1]
https://github.com/apache/spark/blob/6ee28423ad1b2e6089b82af64a31d77d3552bb38/core/src/main/scala/org/apache/spark/ExecutorAllocationManager.scala
[2]
https://github.com/apache/spark/blob/01e14bf303e61a5726f3b1418357a50c1bf8b16f/resource-managers/yarn/src/main/scala/org/apache/spark/deploy/yarn/ApplicationMaster.scala#L446-L480
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
