You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by "Jean-Daniel Cryans (JIRA)" <ji...@apache.org> on 2009/06/23 21:48:07 UTC
[jira] Created: (HBASE-1575) HMaster does not handle ZK session
expiration
HMaster does not handle ZK session expiration
---------------------------------------------
Key: HBASE-1575
URL: https://issues.apache.org/jira/browse/HBASE-1575
Project: Hadoop HBase
Issue Type: Bug
Components: master
Reporter: Jean-Daniel Cryans
Assignee: Jean-Daniel Cryans
Priority: Critical
Fix For: 0.20.0
Currently the master does nothing when he gets a KeeperState.Expired event from ZK. That means that a new master will be elected but the other will still be there doing weird stuff.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1575) HMaster does not handle ZK session
expiration
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12728169#action_12728169 ]
stack commented on HBASE-1575:
------------------------------
Patch looks good.
So, if cluster has only one master and master loses its lease in zk, previous, it just stuck around as a zombie. Now it aborts. I suppose that'll drive home the many-masters in a pretty effective way.
> HMaster does not handle ZK session expiration
> ---------------------------------------------
>
> Key: HBASE-1575
> URL: https://issues.apache.org/jira/browse/HBASE-1575
> Project: Hadoop HBase
> Issue Type: Bug
> Components: master
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Priority: Critical
> Fix For: 0.20.0
>
> Attachments: hbase-1575.patch
>
>
> Currently the master does nothing when he gets a KeeperState.Expired event from ZK. That means that a new master will be elected but the other will still be there doing weird stuff.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1575) HMaster does not handle ZK session
expiration
Posted by "Jean-Daniel Cryans (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12728256#action_12728256 ]
Jean-Daniel Cryans commented on HBASE-1575:
-------------------------------------------
I forgot to handle the shutdown request, it was a test missing in process(). It is now committed.
> HMaster does not handle ZK session expiration
> ---------------------------------------------
>
> Key: HBASE-1575
> URL: https://issues.apache.org/jira/browse/HBASE-1575
> Project: Hadoop HBase
> Issue Type: Bug
> Components: master
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Priority: Critical
> Fix For: 0.20.0
>
> Attachments: hbase-1575.patch
>
>
> Currently the master does nothing when he gets a KeeperState.Expired event from ZK. That means that a new master will be elected but the other will still be there doing weird stuff.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-1575) HMaster does not handle ZK session
expiration
Posted by "Jean-Daniel Cryans (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1575?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jean-Daniel Cryans updated HBASE-1575:
--------------------------------------
Attachment: hbase-1575.patch
While playing with Master restart, I found that it was very difficult to do since stopping the threads always seemed to stop the region servers. Also the visibility in RegionManager isn't great. So, since it's critical that a Master who lost its znode should stop managing the cluster, I make it kill itself. Someone who really needs a restart can still do it via cluster management, or we can do it cleanly for 0.21, or both.
> HMaster does not handle ZK session expiration
> ---------------------------------------------
>
> Key: HBASE-1575
> URL: https://issues.apache.org/jira/browse/HBASE-1575
> Project: Hadoop HBase
> Issue Type: Bug
> Components: master
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Priority: Critical
> Fix For: 0.20.0
>
> Attachments: hbase-1575.patch
>
>
> Currently the master does nothing when he gets a KeeperState.Expired event from ZK. That means that a new master will be elected but the other will still be there doing weird stuff.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Resolved: (HBASE-1575) HMaster does not handle ZK session
expiration
Posted by "Jean-Daniel Cryans (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1575?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jean-Daniel Cryans resolved HBASE-1575.
---------------------------------------
Resolution: Fixed
Hadoop Flags: [Reviewed]
Committed to trunk.
> HMaster does not handle ZK session expiration
> ---------------------------------------------
>
> Key: HBASE-1575
> URL: https://issues.apache.org/jira/browse/HBASE-1575
> Project: Hadoop HBase
> Issue Type: Bug
> Components: master
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Priority: Critical
> Fix For: 0.20.0
>
> Attachments: hbase-1575.patch
>
>
> Currently the master does nothing when he gets a KeeperState.Expired event from ZK. That means that a new master will be elected but the other will still be there doing weird stuff.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.