You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "stack (JIRA)" <ji...@apache.org> on 2010/12/22 22:55:03 UTC
[jira] Created: (HBASE-3390) Double-close makes for double-assign
of a region
Double-close makes for double-assign of a region
------------------------------------------------
Key: HBASE-3390
URL: https://issues.apache.org/jira/browse/HBASE-3390
Project: HBase
Issue Type: Bug
Reporter: stack
I see double-assign during a rolling restart test. Its happening when master joins existing cluster... half the RSs have reported in but its taking on splits anyways (I have it splitting and balancing continuously). A split comes in immediately followed by a balance. What I think is happening is that the split comes in and clears a CLOSING from RIT as part of its cleanup of parent region references. This is messing up the run of balancers' close somehow; we're getting node created and children changed zk events and this makes somehow for the double closed.
I'm enabling zk logging and will try harder to hunt this one down.
Not creating as BLOCKER on 0.90.0 because I think its ok to disable balancer while rolling restart is going on -- and the current rolling restart is a horror anyways doing its maximmal region moving.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-3390) Double-close makes for double-assign
of a region
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-3390?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-3390:
-------------------------
Attachment: history_of_region_in_master.txt
Here is extract from Master. I checked RS log and its doing nice clean close.
> Double-close makes for double-assign of a region
> ------------------------------------------------
>
> Key: HBASE-3390
> URL: https://issues.apache.org/jira/browse/HBASE-3390
> Project: HBase
> Issue Type: Bug
> Reporter: stack
> Attachments: history_of_region_in_master.txt
>
>
> I see double-assign during a rolling restart test. Its happening when master joins existing cluster... half the RSs have reported in but its taking on splits anyways (I have it splitting and balancing continuously). A split comes in immediately followed by a balance. What I think is happening is that the split comes in and clears a CLOSING from RIT as part of its cleanup of parent region references. This is messing up the run of balancers' close somehow; we're getting node created and children changed zk events and this makes somehow for the double closed.
> I'm enabling zk logging and will try harder to hunt this one down.
> Not creating as BLOCKER on 0.90.0 because I think its ok to disable balancer while rolling restart is going on -- and the current rolling restart is a horror anyways doing its maximmal region moving.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.