[
https://issues.apache.org/jira/browse/HBASE-21863?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sergey Shelukhin updated HBASE-21863:
-------------------------------------
Status: Patch Available (was: Open)
This adds two changes, one I like and one I don't like as much, but I think it
might be useful even after we fix HBASE-21862, to speed up the master-RS
coordination after a failure with a timeout, where master has to close the
region and confirm close if we open it.
Change #1 is to kill RS that opens the region we don't expect it to open.
Before fixing other races I'm not adding it back to RS report, but during the
assignment it's more likely something is screwed up and we'll double assign,
that that RS is just sending an innocent "I opened it" due to a benign race.
Change #2 is to add the timeout to procedures RPC, so it's not executed after
master already gave up on it and we failed to respond in time. This is not
bulletproof; I added a setting to turn it off or add a safety buffer.
Still need to test on the cluster.
[~Apache9] [~busbey] can you take a look? thnx
> narrow down the double-assignment race window
> ---------------------------------------------
>
> Key: HBASE-21863
> URL: https://issues.apache.org/jira/browse/HBASE-21863
> Project: HBase
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Priority: Major
> Attachments: HBASE-21863.patch
>
>
> See HBASE-21862.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
