[
https://issues.apache.org/jira/browse/HBASE-5344?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13209128#comment-13209128
]
Phabricator commented on HBASE-5344:
------------------------------------
Karthik has commented on the revision "[jira] [HBASE-5344] [89-fb] Scan
unassigned region directory on master failover".
@stack -
<< So, the '/unassigned' dir will change to be named '/regions' or some such?
On region open, we'll update its state in zk to be OPENED? And leave it there
(and update .META. too). We don't really need .META. then? Smile. >>
Not totally... the idea is to change /unassigned to /regions/<tablename> and
it will not be deleted on the region being opened. This will only track
assignment of regions though. We still need meta to figure out start/stop keys
and other information like preferred regionservers for the regions. In order to
truly eliminate meta, we should be willing to store more permanent data in ZK,
but thats probably not needed unless we are going all the way to rip out root
and meta special-casing, which would be a huge change.
<< "We would like to rely on ZK and (for now) on META instead to recover the
region assignment on master startup/failure." The hard part in here is on
failover, what if the .META. is on a crashed server? You'll need to process its
logs and get .META. back online before you can proceed w/ failover. To get it
online, you'll need to listen for events (though I suppose you could filter and
only process .META. events). Or what if the the server carrrying .META. crashes
during onlining. >>
The idea here is that the master can read unassigned (while continuing to
queue events). It will split all logs (via distributed log splitting) and
process opening of the meta region out of band (not go though the standard
heartbeat based assignment - we do this for root upon startup, we need
something similar for meta).
REVISION DETAIL
https://reviews.facebook.net/D1605
> [89-fb] Scan unassigned region directory on master failover
> -----------------------------------------------------------
>
> Key: HBASE-5344
> URL: https://issues.apache.org/jira/browse/HBASE-5344
> Project: HBase
> Issue Type: Bug
> Reporter: Mikhail Bautin
> Assignee: Mikhail Bautin
> Attachments: D1605.1.patch
>
>
> In case the master dies after a regionserver writes region state as OPENED or
> CLOSED in ZK but before the update is received by master and written to meta,
> the new master that comes up has to pick up the region state from ZK and
> write it to meta. Otherwise we can get multiply-assigned regions.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
