GitHub user zsxwing opened a pull request:
https://github.com/apache/spark/pull/22816
[SPARK-25822][PySpark]Fix a race condition when releasing a Python worker
## What changes were proposed in this pull request?
There is a race condition when releasing a Python worker. If
`ReaderIterator.handleEndOfDataSection` is not running in the task thread, when
a task is early terminated (such as `take(N)`), the task completion listener
may close the worker but "handleEndOfDataSection" can still put the worker into
the worker pool to reuse.
https://github.com/zsxwing/spark/commit/0e07b483d2e7c68f3b5c3c118d0bf58c501041b7
is a patch to reproduce this issue.
I also found a user reported this in the mail list:
http://mail-archives.apache.org/mod_mbox/spark-user/201610.mbox/%3CCAAUq=h+yluepd23nwvq13ms5hostkhx3ao4f4zqv6sgo5zm...@mail.gmail.com%3E
This PR fixes the issue by using `compareAndSet` to make sure we will never
return a closed worker to the work pool.
## How was this patch tested?
Jenkins.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/zsxwing/spark fix-socket-closed
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/22816.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #22816
----
commit a22e38917b4f2893ce0d72febed4df1d3eb9fdd5
Author: Shixiong Zhu <zsxwing@...>
Date: 2018-10-24T07:51:26Z
Fix a race condition when releasing a Python worker
----
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
