[
https://issues.apache.org/jira/browse/HIVE-19892?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16513214#comment-16513214
]
Jason Dere commented on HIVE-19892:
-----------------------------------
It looks like this was the 0-rows case - the temp directory for the query
doesn't actually exist (thus the "dummy_zero_rows" path in the results cache).
So there were no actual file operations that occurred, which is why there was
no error when trying to save the results to the results cache directory.
This path won't occur with the patch - if caching is disabled due to this fix,
then neither lookup or saving to the cache will occur.
That said, if you want a way to guarantee we don't hit this situation, maybe in
the 0-rows case we can do a FS.exists() check on the dummy_zero_rows path - I
guess that should hit an error and so Hive won't be able to save the results to
the results cache. How does that sound?
> Disable query results cache for for HiveServer2 doAs=true
> ---------------------------------------------------------
>
> Key: HIVE-19892
> URL: https://issues.apache.org/jira/browse/HIVE-19892
> Project: Hive
> Issue Type: Sub-task
> Reporter: Jason Dere
> Assignee: Jason Dere
> Priority: Major
> Fix For: 3.1.0
>
> Attachments: HIVE-19892.1.patch
>
>
> If running HS2 with doAs=true, the temp query results directory will have
> ownership/permissions based on the doAs user. A subsequent query running as a
> different user may not be able to access this query results directory.
> Results caching will have to be disabled in this case.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
