You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by "ryan rawson (JIRA)" <ji...@apache.org> on 2009/04/27 22:21:30 UTC
[jira] Created: (HBASE-1351) need to be able to rolling restart
hbase
need to be able to rolling restart hbase
----------------------------------------
Key: HBASE-1351
URL: https://issues.apache.org/jira/browse/HBASE-1351
Project: Hadoop HBase
Issue Type: Bug
Affects Versions: 0.20.0
Reporter: ryan rawson
Fix For: 0.20.0
Once a hbase install goes live, unless you can take the readers offline, you cannot stop and start the cluster to issue config changes, and other generalized changes.
We need to do a rolling restart. My own experiments confirmed that:
- Restarting a regionserver works, you lose access to those regions, but comes back when that regionserver restarts moments later. Maybe there should be a 'reassign these regions because i am going to shut down' flag/code/command.
- Restarting .META. regionserver just does not work. The master becomes confused. I have seen it reassign the .META. but not update -ROOT-, now no one knows where it is, and the root/meta scanners in the master just get hung.
- Cant restart master yet, need multi-master election via ZK to make this happen.
This is a combo bug/feature request.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1351) need to be able to rolling restart
hbase
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12711798#action_12711798 ]
stack commented on HBASE-1351:
------------------------------
It looks like we have all tools in place to do this when hbase-1302 goes in. I'll test and if works, write a wiki page on it and close this issue.
> need to be able to rolling restart hbase
> ----------------------------------------
>
> Key: HBASE-1351
> URL: https://issues.apache.org/jira/browse/HBASE-1351
> Project: Hadoop HBase
> Issue Type: Bug
> Affects Versions: 0.20.0
> Reporter: ryan rawson
> Fix For: 0.20.0
>
>
> Once a hbase install goes live, unless you can take the readers offline, you cannot stop and start the cluster to issue config changes, and other generalized changes.
> We need to do a rolling restart. My own experiments confirmed that:
> - Restarting a regionserver works, you lose access to those regions, but comes back when that regionserver restarts moments later. Maybe there should be a 'reassign these regions because i am going to shut down' flag/code/command.
> - Restarting .META. regionserver just does not work. The master becomes confused. I have seen it reassign the .META. but not update -ROOT-, now no one knows where it is, and the root/meta scanners in the master just get hung.
> - Cant restart master yet, need multi-master election via ZK to make this happen.
> This is a combo bug/feature request.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Resolved: (HBASE-1351) need to be able to rolling restart
hbase
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack resolved HBASE-1351.
--------------------------
Resolution: Fixed
I talked to Ryan. I asked if ok to close and he said "Yes... probably". Taking that as a yes for now. We can open a new issue to cover improvements on rolling restart.
> need to be able to rolling restart hbase
> ----------------------------------------
>
> Key: HBASE-1351
> URL: https://issues.apache.org/jira/browse/HBASE-1351
> Project: Hadoop HBase
> Issue Type: Bug
> Affects Versions: 0.20.0
> Reporter: ryan rawson
> Assignee: stack
> Fix For: 0.20.0
>
>
> Once a hbase install goes live, unless you can take the readers offline, you cannot stop and start the cluster to issue config changes, and other generalized changes.
> We need to do a rolling restart. My own experiments confirmed that:
> - Restarting a regionserver works, you lose access to those regions, but comes back when that regionserver restarts moments later. Maybe there should be a 'reassign these regions because i am going to shut down' flag/code/command.
> - Restarting .META. regionserver just does not work. The master becomes confused. I have seen it reassign the .META. but not update -ROOT-, now no one knows where it is, and the root/meta scanners in the master just get hung.
> - Cant restart master yet, need multi-master election via ZK to make this happen.
> This is a combo bug/feature request.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1351) need to be able to rolling restart
hbase
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12711955#action_12711955 ]
stack commented on HBASE-1351:
------------------------------
With JD's patch, there's a basic rolling restart in place... Seems to work in testing. Can update at least point releases where RPC don't change and no data migration needed. I did up a wiki page: http://wiki.apache.org/hadoop/Hbase/RollingRestart. We probably need to make the whole operation smoother but I figure we can open new issues as we learn more practising rolling restarts. OK to close this one?
> need to be able to rolling restart hbase
> ----------------------------------------
>
> Key: HBASE-1351
> URL: https://issues.apache.org/jira/browse/HBASE-1351
> Project: Hadoop HBase
> Issue Type: Bug
> Affects Versions: 0.20.0
> Reporter: ryan rawson
> Assignee: stack
> Fix For: 0.20.0
>
>
> Once a hbase install goes live, unless you can take the readers offline, you cannot stop and start the cluster to issue config changes, and other generalized changes.
> We need to do a rolling restart. My own experiments confirmed that:
> - Restarting a regionserver works, you lose access to those regions, but comes back when that regionserver restarts moments later. Maybe there should be a 'reassign these regions because i am going to shut down' flag/code/command.
> - Restarting .META. regionserver just does not work. The master becomes confused. I have seen it reassign the .META. but not update -ROOT-, now no one knows where it is, and the root/meta scanners in the master just get hung.
> - Cant restart master yet, need multi-master election via ZK to make this happen.
> This is a combo bug/feature request.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Assigned: (HBASE-1351) need to be able to rolling restart
hbase
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1351?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack reassigned HBASE-1351:
----------------------------
Assignee: stack
> need to be able to rolling restart hbase
> ----------------------------------------
>
> Key: HBASE-1351
> URL: https://issues.apache.org/jira/browse/HBASE-1351
> Project: Hadoop HBase
> Issue Type: Bug
> Affects Versions: 0.20.0
> Reporter: ryan rawson
> Assignee: stack
> Fix For: 0.20.0
>
>
> Once a hbase install goes live, unless you can take the readers offline, you cannot stop and start the cluster to issue config changes, and other generalized changes.
> We need to do a rolling restart. My own experiments confirmed that:
> - Restarting a regionserver works, you lose access to those regions, but comes back when that regionserver restarts moments later. Maybe there should be a 'reassign these regions because i am going to shut down' flag/code/command.
> - Restarting .META. regionserver just does not work. The master becomes confused. I have seen it reassign the .META. but not update -ROOT-, now no one knows where it is, and the root/meta scanners in the master just get hung.
> - Cant restart master yet, need multi-master election via ZK to make this happen.
> This is a combo bug/feature request.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.