You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "Karthik Ranganathan (Created) (JIRA)" <ji...@apache.org> on 2011/11/11 21:28:51 UTC
[jira] [Created] (HBASE-4774) Retain znodes for all regions in
UNASSIGNED (zookeeper)
Retain znodes for all regions in UNASSIGNED (zookeeper)
-------------------------------------------------------
Key: HBASE-4774
URL: https://issues.apache.org/jira/browse/HBASE-4774
Project: HBase
Issue Type: Brainstorming
Components: master
Reporter: Karthik Ranganathan
We delete znodes from UNASSIGNED once a region is opened and the open is entered into META.
A lot of multiple assignment issues arise from the fact that we delete this znode and dont have state information after the region is opened. So one proposal is to always retain the znodes in META and never delete the regions from UNASSIGNED. That way we can have a simple, deterministic state machine for each region. We could bucket the regions into sub-directories using a deterministic hash or the region prefixes.
Note that what we put in zk is not permanent state but just the state of the cluster that is running. This state is invalid upon cluster restart - so we are still storing temporary data in zk.
Thoughts?
--
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
[jira] [Commented] (HBASE-4774) Retain znodes for all regions in
UNASSIGNED (zookeeper)
Posted by "Todd Lipcon (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-4774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13148742#comment-13148742 ]
Todd Lipcon commented on HBASE-4774:
------------------------------------
I agree. I also would really like to define a proper FSM class for region state transitions -- one for RegionOnMaster and another for RegionOnRS (the state machines are a bit different between the two). The inputs into the state machines would be ZK state changes as well as lifecycle events in the server. I've done a bit of hacking in this direction, but it's not a simple project.
> Retain znodes for all regions in UNASSIGNED (zookeeper)
> -------------------------------------------------------
>
> Key: HBASE-4774
> URL: https://issues.apache.org/jira/browse/HBASE-4774
> Project: HBase
> Issue Type: Brainstorming
> Components: master
> Reporter: Karthik Ranganathan
>
> We delete znodes from UNASSIGNED once a region is opened and the open is entered into META.
> A lot of multiple assignment issues arise from the fact that we delete this znode and dont have state information after the region is opened. So one proposal is to always retain the znodes in META and never delete the regions from UNASSIGNED. That way we can have a simple, deterministic state machine for each region. We could bucket the regions into sub-directories using a deterministic hash or the region prefixes.
> Note that what we put in zk is not permanent state but just the state of the cluster that is running. This state is invalid upon cluster restart - so we are still storing temporary data in zk.
> Thoughts?
--
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
[jira] [Commented] (HBASE-4774) Retain znodes for all regions in
UNASSIGNED (zookeeper)
Posted by "stack (Commented) (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-4774?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13148817#comment-13148817 ]
stack commented on HBASE-4774:
------------------------------
I like this idea. I think it would help determinacy. We'd have to rename UNASSIGNED to REGIONS. Split would hoist new regions up into zk and delete of parent region would have to remove from zk (which is a little ugly -- delete from .meta., fs, and zk).
> Retain znodes for all regions in UNASSIGNED (zookeeper)
> -------------------------------------------------------
>
> Key: HBASE-4774
> URL: https://issues.apache.org/jira/browse/HBASE-4774
> Project: HBase
> Issue Type: Brainstorming
> Components: master
> Reporter: Karthik Ranganathan
>
> We delete znodes from UNASSIGNED once a region is opened and the open is entered into META.
> A lot of multiple assignment issues arise from the fact that we delete this znode and dont have state information after the region is opened. So one proposal is to always retain the znodes in META and never delete the regions from UNASSIGNED. That way we can have a simple, deterministic state machine for each region. We could bucket the regions into sub-directories using a deterministic hash or the region prefixes.
> Note that what we put in zk is not permanent state but just the state of the cluster that is running. This state is invalid upon cluster restart - so we are still storing temporary data in zk.
> Thoughts?
--
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