tez-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sergey Shelukhin (JIRA)" <j...@apache.org>
Subject [jira] [Created] (TEZ-3879) potential abort propagation issue (race?)
Date Tue, 19 Dec 2017 23:18:00 GMT
Sergey Shelukhin created TEZ-3879:
-------------------------------------

             Summary: potential abort propagation issue (race?)
                 Key: TEZ-3879
                 URL: https://issues.apache.org/jira/browse/TEZ-3879
             Project: Apache Tez
          Issue Type: Bug
            Reporter: Sergey Shelukhin


I'm looking at a Hive query where AM aborts some tasks for whatever reason (AM preemption).
On the nodes, most tasks receive abort normally, like so (note that TezProcessor is part of
Hive and the main class that Tez code calls; the first thing it does on any abort is log "Received
abort").
{noformat}
2017-12-18T14:44:26,616 INFO  [TaskHeartbeatThread ()] org.apache.tez.runtime.task.TezTaskRunner2:
Attempting to abort attempt_1513367667720_3619_1_02_000012_0 due to an invocation of shutdownRequested
2017-12-18T14:44:26,621 INFO  [TaskHeartbeatThread ()] org.apache.hadoop.hive.ql.exec.tez.TezProcessor:
Received abort
2017-12-18T14:44:26,622 INFO  [TaskHeartbeatThread ()] org.apache.hadoop.hive.ql.exec.tez.TezProcessor:
Forwarding abort to RecordProcessor
2017-12-18T14:44:26,622 INFO  [TaskHeartbeatThread ()] org.apache.hadoop.hive.ql.exec.tez.MapRecordProcessor:
Forwarding abort to mapOp: {} MAP
{noformat}

However on some tasks that are terminated shortly after init, TezProcessor is never called.
Moreover, when AM tries to kill the task again (when it's already running, having ignored
the abort) Tez says the task is already aborted and doesn't propagate this either.

{noformat}
2017-12-18T14:47:22,995  INFO [TezTR-667720_3619_3_2_12_0 (1513367667720_3619_3_02_000012_0)]
reducesink.VectorReduceSinkCommonOperator: Using tag = -1
(this is the end of Hive init)
...
2017-12-18T14:47:23,133 INFO  [TaskHeartbeatThread ()] org.apache.hadoop.hive.llap.daemon.impl.LlapTaskReporter:
Asked to die via task heartbeat: attempt_1513367667720_3619_3_02_000012_0
2017-12-18T14:47:23,133 INFO  [TaskHeartbeatThread ()] org.apache.tez.runtime.task.TezTaskRunner2:
Attempting to abort attempt_1513367667720_3619_3_02_000012_0 due to an invocation of shutdownRequested
(no TezProcessor log statements)
{noformat}
The task keeps running and the next kill is ignored
{noformat}
2017-12-18T14:47:23,575 INFO  [IPC Server handler 2 on 40617 ()] org.apache.hadoop.hive.llap.daemon.impl.ContainerRunnerImpl:
DBG: Received terminateFragment request for attempt_1513367667720_3619_3_02_000012_0
...
2017-12-18T14:47:23,575 INFO  [IPC Server handler 2 on 40617 ()] org.apache.tez.runtime.task.TezTaskRunner2:
Ignoring killTask request since the task with id attempt_1513367667720_3619_3_02_000012_0
has ended for reason: CONTAINER_STOP_REQUESTED. IgnoredError:  
{noformat}



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message