[
https://issues.apache.org/jira/browse/HADOOP-18993?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17808654#comment-17808654
]
ASF GitHub Bot commented on HADOOP-18993:
-----------------------------------------
tmnd1991 commented on PR #6301:
URL: https://github.com/apache/hadoop/pull/6301#issuecomment-1900429593
So finally I go the final "push" on this.
I was able to verify my tests pass through running locally: `mvn -pl
hadoop-tools/hadoop-aws verify -Dit.test=ITestS3AFileSystemIsolatedClassloader
-Dtest=none` while running `mvn -pl hadoop-tools/hadoop-aws clean verify` lead
to 40ish errors/failures that to my eyes look completely unrelated to this
patch.
I will wait for yetus results in order to see if I've left anything behind :)
Maybe the last thing that I might do to improve this is to make it a unit
test instead of an integration one, but given [your
comment](https://github.com/apache/hadoop/pull/6301#discussion_r1407732777) I
guess given the fact that instantiates and fs it must be an IT test and I can't
think a way of testing this without instantiating the fs.
> Allow to not isolate S3AFileSystem classloader when needed
> ----------------------------------------------------------
>
> Key: HADOOP-18993
> URL: https://issues.apache.org/jira/browse/HADOOP-18993
> Project: Hadoop Common
> Issue Type: Improvement
> Components: hadoop-thirdparty
> Affects Versions: 3.3.6
> Reporter: Antonio Murgia
> Priority: Minor
> Labels: pull-request-available
>
> In HADOOP-17372 the S3AFileSystem forces the configuration classloader to be
> the same as the one that loaded S3AFileSystem. This leads to the
> impossibility in Spark applications to load third party credentials providers
> as user jars.
> I propose to add a configuration key
> {{fs.s3a.extensions.isolated.classloader}} with a default value of {{true}}
> that if set to {{false}} will not perform the classloader set.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
