[
https://issues.apache.org/jira/browse/HBASE-27046?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17552140#comment-17552140
]
Duo Zhang commented on HBASE-27046:
-----------------------------------
On branch-2, TestClusterQuotaThrottle will produce a bunch of error logs as it
reset the EnvironmentEdge, which make the filenum related logic in
AbstractFSWAL broken. In this way, our flaky test job can not finish properly
so I disabled it.
After fixing the AbstractFSWAL issue, I think it can quit normally, so even if
it is still flaky, we will add it to the flaky list, then it will not fail our
pre commit build, and we can check the flaky test job output to find out the
root cause and finally make it stable.
So the problem for now, is it will fail the pre commit, but not the nightly
jobs, so it is still not in the flaky list?
For me, I prefer we do not ignore any UTs, just let it go into the flaky tests
list and finally get fixed.
Thanks.
> The filenum in AbstractFSWAL should be monotone increasing
> ----------------------------------------------------------
>
> Key: HBASE-27046
> URL: https://issues.apache.org/jira/browse/HBASE-27046
> Project: HBase
> Issue Type: Improvement
> Reporter: Duo Zhang
> Assignee: Duo Zhang
> Priority: Major
>
> This is the current code
> {code}
> /**
> * retrieve the next path to use for writing. Increments the internal
> filenum.
> */
> private Path getNewPath() throws IOException {
> this.filenum.set(EnvironmentEdgeManager.currentTime());
> Path newPath = getCurrentFileName();
> while (fs.exists(newPath)) {
> this.filenum.incrementAndGet();
> newPath = getCurrentFileName();
> }
> return newPath;
> }
> {code}
> In some tests, we inject our own EnvironmentEdge, it may return the same ts
> always or even go backwards, the logic here is not rnough to keep the filenum
> monotone increasing, as we may have already archive the old file.
--
This message was sent by Atlassian Jira
(v8.20.7#820007)
