[
https://issues.apache.org/jira/browse/HIVE-15184?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15947497#comment-15947497
]
Sunitha Beeram commented on HIVE-15184:
---------------------------------------
Thanks for the clarification [~pvary].
The memory leak would be a concern for MemoryTokenStore right? In case of a
persistent store based TokenStore, specifically in the db case, we'll
accumulate rows infinitely, which may ulltimately degrade the service. Let me
know if I am missing something. For the latter (db/zk) case a metric/alert
might be an option to consider?
If we do go down path of shutting the service down on failure to cleanup, would
be great if we can make the documentation clearer as to why we think its one of
the fatal errors.
For this final point, I don't have a clear answer either yet, but did want to
bring it up: is there a better way than using getRuntime.exit()? For ex, if we
are writing unit-tests for the token-remover, it could result in the test jvm
exiting unexpectedly if for some reason we end up tracing that block, right?
Can we signal the main application thread to exit using a different approach?
> Add the possibility to separate recoverable and not recoverable errors in
> DeletagionTokenStore
> ----------------------------------------------------------------------------------------------
>
> Key: HIVE-15184
> URL: https://issues.apache.org/jira/browse/HIVE-15184
> Project: Hive
> Issue Type: Bug
> Components: HiveServer2, Metastore
> Affects Versions: 1.3.0, 2.0.1, 2.1.0, 2.2.0
> Reporter: Peter Vary
> Assignee: Peter Vary
> Attachments: HIVE-15184.patch
>
>
> After HIVE-15090 committed we discussed it with [~thejas] and agreed, that it
> would be even better if the DelegateTokenStore implementation could decide if
> an error is recoverable or not. Since the DelegationTokenStore is not
> mentioned as a Hive API it is possible to change the interface, so we could
> change the implementations shipped with Hive to use the new functionality,
> and if someone uses it's own implementation of DelegationTokenStore, then he
> should do the matching changes himself
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
