[
https://issues.apache.org/jira/browse/IMPALA-14679?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18051744#comment-18051744
]
ASF subversion and git services commented on IMPALA-14679:
--
Commit 7fbb83749b1aeb3e359313fa5f084a450f85f083 in impala's branch
refs/heads/master from Michael Smith
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=7fbb83749 ]
IMPALA-14679: Avoid authentication hash reload on partial write
Removes IN_MODIFY from the list of inotify triggers for reloading
authentication hashes. IN_MODIFY appears to trigger when a `write`
system call begins, which resulted in tests running into
Authentication hash from /tmp/junit...tmp did not contain 32 bytes
This would delay loading the hash because ReloadMonitor waits 1000ms
before trying again. IN_CLOSE_WRITE is sufficient to cover the same
cases, as it receives a notification when a file opened for writing is
closed.
Adds a test that failed with IN_MODIFY, and now passes 30+ runs.
Change-Id: I9549c7a4ec0e8ebfc8a49444f2f217ffd9311525
Reviewed-on: http://gerrit.cloudera.org:8080/23862
Reviewed-by: Impala Public Jenkins
Tested-by: Impala Public Jenkins
> Authentication hash reload sometimes fails with "did not contain 32 bytes"
> --
>
> Key: IMPALA-14679
> URL: https://issues.apache.org/jira/browse/IMPALA-14679
> Project: IMPALA
> Issue Type: Bug
> Components: Backend
>Affects Versions: Impala 5.0.0
>Reporter: Michael Smith
>Assignee: Michael Smith
>Priority: Major
> Fix For: Impala 5.0.0
>
>
> In diagnosing IMPALA-14678, I noticed that sometimes reload will fail the 1st
> time with
> {code}
> E20260112 15:49:12.255777 338029 openssl-util.cc:385] GENERAL: Authentication
> hash from /tmp/junit713748083974798515/junit3924236808116170714.tmp did not
> contain 32 bytes
> {code}
> It would get successfully reloaded on retry after a 1s sleep.
> This appears to be a result of using {{IN_MODIFY}}, even though writing out
> the 32 bytes should happen via a single {{write}} system call. We should
> instead rely on {{IN_CLOSE_WRITE}} for these cases, which will occur after
> all {{write}} system calls and should let us read the full file.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
-
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
