[
https://issues.apache.org/jira/browse/HADOOP-14158?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16138143#comment-16138143
]
Steve Loughran commented on HADOOP-14158:
-----------------------------------------
What's probably happening in the test is the config is being cached. The
configuration needs to mark the FS as non-cached/create a unique one.
> Possible for modified configuration to leak into metadatastore in S3GuardTool
> -----------------------------------------------------------------------------
>
> Key: HADOOP-14158
> URL: https://issues.apache.org/jira/browse/HADOOP-14158
> Project: Hadoop Common
> Issue Type: Sub-task
> Components: fs/s3
> Reporter: Sean Mackrory
>
> It doesn't appear to do it when run from the command-line, but when running
> the S3GuardTool.run (i.e. the parent function of most of the functions used
> in the unit tests) from a unit test, you end up with a NullMetadataStore,
> regardless of what else was configured.
> We create an instance of S3AFileSystem with the metadata store implementation
> overridden to NullMetadataStore so that we have distinct interfaces to S3 and
> the metadata store. S3Guard can later be called using this filesystem,
> causing it to pick up the filesystem's configuration, which instructs it to
> use the NullMetadataStore implementation. This shouldn't be possible.
> It is unknown if this happens in any real-world scenario - I've been unable
> to reproduce the problem from the command-line. But it definitely happens in
> a test, it shouldn't, and fixing this will at least allow HADOOP-14145 to
> have an automated test.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org
