[
https://issues.apache.org/jira/browse/DRILL-3967?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14971150#comment-14971150
]
Sudheesh Katkam commented on DRILL-3967:
----------------------------------------
Note that after planning is done, the Foreman thread exits; the state
transitions are handled by RPC threads.
The cancel signal is sent as soon as the three batches arrive (not ideal), at
this point the query state is either RUNNING or COMPLETED.
The test passes if the state transition from RUNNING --> CANCELLATION_REQUESTED
is handled by a BitServer thread. The resume signal from the UserServer thread
is ignored.
The test hangs if the state transition from RUNNING -->
CANCELLATION_REQUESTED --> CANCELLED is handled by the UserServer thread. The
thread is waiting (on itself) for a resume signal when closing.
The test fails if the state transition from RUNNING --> COMPLETED is
handled by a BitServer thread. The CANCELLATION_REQUESTED message is simply
ignored. And the resume signal allows the query to move to COMPLETED state.
> Broken Test: TestDrillbitResilience.cancelAfterEverythingIsCompleted()
> ----------------------------------------------------------------------
>
> Key: DRILL-3967
> URL: https://issues.apache.org/jira/browse/DRILL-3967
> Project: Apache Drill
> Issue Type: Test
> Components: Execution - Flow, Execution - RPC
> Affects Versions: 1.2.0
> Reporter: Andrew
> Assignee: Sudheesh Katkam
> Priority: Minor
>
> TestDrillbitResilience.cancelAfterEverythingIsCompleted() can sometimes fail.
> I've noticed that running this test on an m2.xlarge on AWS causes a
> reproducible failure when running against the patch for
> https://issues.apache.org/jira/browse/DRILL-3749 (Upgraded Hadoop and Curator
> libraries).
> When running this test with the same patch on my laptop, this test passes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
