[
https://issues.apache.org/jira/browse/FLINK-9270?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16457520#comment-16457520
]
ASF GitHub Bot commented on FLINK-9270:
---------------------------------------
Github user StefanRRichter commented on the issue:
https://github.com/apache/flink/pull/5937
@bowenli86 I already wanted to do this for some time but unfortunately we
currently cannot upgrade RocksDB to any higher version than what is used in
master. It seems like there is again a performance regression for the merge
operator for all newer versions. If you check your Travis runs, you will see
that `RocksDBPerformanceTest.testRocksDbMergePerformance` fails all the time. I
have commented about this problem in the RocksDB issue tracker, but no reaction
so far.
If you agree, please close this PR.
> Upgrade RocksDB to 5.11.3, and resolve concurrent test invocation problem of
> @RetryOnFailure
> --------------------------------------------------------------------------------------------
>
> Key: FLINK-9270
> URL: https://issues.apache.org/jira/browse/FLINK-9270
> Project: Flink
> Issue Type: Improvement
> Components: State Backends, Checkpointing
> Affects Versions: 1.5.0
> Reporter: Bowen Li
> Assignee: Bowen Li
> Priority: Major
> Fix For: 1.6.0
>
>
> Upgrade RocksDB to 5.11.3 to take latest bug fixes
> Besides, I found that unit tests annotated with {{@RetryOnFailure}} will be
> run concurrently if there's only {{try}} clause without a {{catch}}
> following. For example, sometimes,
> {{RocksDBPerformanceTest.testRocksDbMergePerformance()}} will actually be
> running in 3 concurrent invocations, and multiple concurrent write to RocksDB
> result in errors.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
