[
https://issues.apache.org/jira/browse/HBASE-4042?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13057020#comment-13057020
]
stack commented on HBASE-4042:
------------------------------
More. More interesting deadlocks when servers carrying both -ROOT- and .META.
are carried by the servers that are killed.
This test also uses a timeout of 30 seconds but default rpc setup timeout is 60
seconds.
> TestDistributedLogSplitting broke on trunk
> ------------------------------------------
>
> Key: HBASE-4042
> URL: https://issues.apache.org/jira/browse/HBASE-4042
> Project: HBase
> Issue Type: Bug
> Reporter: stack
>
> There's a couple of issues going on here. Its taken me a while to figure
> whats up and am still not done. Here's what I found so far:
> + The last test testWorkerAbort has been hanging because there is no one to
> process the log split the final assert is expecting completed; we've killed
> the lone RS that this test put up. This test passes most of the time for me
> locally; luck has the log processed before the RS with its splitLogWorker
> processes the log before we go out. Putting up a new RS with a
> splitLogWorker makes this test pass for me most of the time now but I've seen
> an error in testing so need to dig in still.
> + The first test, testThreeRSAbort is a good test. Its turning up a issue
> that has nothing to do w/ log splitting. If .META. is on one of the RSs that
> goes down -- we launch w/ 6 RSs -- then we get stuck in catalog tracker
> waiting on meta to be up again (though it deploys fin -- we don't notice its
> deploy in master). I'm on this one at mo.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
