[
https://issues.apache.org/jira/browse/HBASE-3107?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12920931#action_12920931
]
Alex Newman commented on HBASE-3107:
------------------------------------
I know right, it seems like a lot. So I am very openminded to a change like
that but here is what ran through my head. Distributed logsplitting now just
enqueues zk nodes when a new split comes in. It's then up to a thread running
in the regionserver to do the actual splitting. The choice to skipErrors or
not to skipErrors happens by checking the (hbase.hlog.split.skip.errors)
variable whenever the real split log happens. However, once I startup the
regionserver in the test (@BeforeClass) with skip errors or not skip errors, I
worry we are stuck with it, for the length of the test. As an alternative:
- We could embed into the zk data whether or not it's a skip errors log I
suppose it could be done, but that seems like a weird place to do it *shrug*
- Alternatively we could change a property on the running regionserver? I don't
know of any code which does something like that?
Anyway, reducing unit test run time is a goal of mine as well, as I am lazy ,
any suggestions are welcome.
> Breakup HLogSplitTest unit tests.
> ---------------------------------
>
> Key: HBASE-3107
> URL: https://issues.apache.org/jira/browse/HBASE-3107
> Project: HBase
> Issue Type: Sub-task
> Reporter: Alex Newman
> Assignee: Alex Newman
>
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
