[
https://issues.apache.org/jira/browse/YARN-5272?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wilfred Spiegelenburg updated YARN-5272:
----------------------------------------
Attachment: YARN-5272.4.patch
OK the assessment was correct: it is the encoding, however not on read but on
write. The XML parser defaults to UTF-8 but the FileWriter does not. The
FileWriter picks up the standard file encoding of the OS and if that is not a
UTF-8 based encoding the char written will be corrupted in the file on write.
Changing the file writing to make sure a UTF-8 encoding is used even if the OS
is not using it.
> FairScheduler handles "invalid" queue names inconsistently even after
> YARN-3241
> -------------------------------------------------------------------------------
>
> Key: YARN-5272
> URL: https://issues.apache.org/jira/browse/YARN-5272
> Project: Hadoop YARN
> Issue Type: Bug
> Components: fairscheduler
> Affects Versions: 2.8.0
> Reporter: Wilfred Spiegelenburg
> Assignee: Wilfred Spiegelenburg
> Attachments: YARN-5272.1.patch, YARN-5272.3.patch, YARN-5272.4.patch
>
>
> The fix used in YARN-3214 uses a the JDK trim() method to remove leading and
> trailing spaces. The QueueMetrics uses a guava based trim when it splits the
> queues.
> The guava based trim uses the unicode definition of a white space which is
> different than the java trim as can be seen
> [here|https://docs.google.com/a/cloudera.com/spreadsheets/d/1kq4ECwPjHX9B8QUCTPclgsDCXYaj7T-FlT4tB5q3ahk/pub]
> A queue name with a non-breaking white space will thus still cause the same
> "Metrics source XXX already exists!" MetricsException.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
