You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hbase.apache.org by "Andrew Purtell (JIRA)" <ji...@apache.org> on 2008/12/18 14:19:45 UTC
[jira] Created: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Master should support close/open/reassignment/enable/disable operations on individual regions
---------------------------------------------------------------------------------------------
Key: HBASE-1066
URL: https://issues.apache.org/jira/browse/HBASE-1066
Project: Hadoop HBase
Issue Type: Improvement
Reporter: Andrew Purtell
The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-1066:
-------------------------
Attachment: 1066-v4.patch
Add shell help for these new 'tools'
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Fix For: 0.19.0
>
> Attachments: 1066-v3.patch, 1066-v4.patch, 1066.patch
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Assigned: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack reassigned HBASE-1066:
----------------------------
Assignee: stack
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Assignee: stack
> Fix For: 0.19.0
>
> Attachments: 1066-v3.patch, 1066-v4.patch, 1066.patch
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-1066:
-------------------------
Attachment: 1066-v3.patch
Adds enable/disable and close region with optional removal of STARTCODE and SERVER (so gets reassigned). Also added being able to specify servername you want to send the close to in case META is not in agreement with whats out on regionservers. Done as new commands in the shell: enable_region, disable_region, and close_region. Need to add in help and figure why removal of STARTCODE and SERVER does not always get a region reassigned.
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Fix For: 0.19.0
>
> Attachments: 1066-v3.patch, 1066.patch
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "Andrew Purtell (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12658396#action_12658396 ]
Andrew Purtell commented on HBASE-1066:
---------------------------------------
The HBaseAdmin.modifyTable API is perfect for that. Just add op constants and backing implementation in HMaster for MODIFY_TABLE_CLOSE, MODIFY_TABLE_OPEN, etc.
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Fix For: 0.19.0
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12658167#action_12658167 ]
stack commented on HBASE-1066:
------------------------------
Agreed.
You can do some of this now but its more awkward than it need be. To enable/disable individual region, you'd write a script that got info:regioninfo cell from .META., rendered it as HRI, set the particular flag, then wrote it back. If master has a region and no HRS does, then setting the server to null will force the master to reassign. Both of the above can be done in two or three lines of script but error-prone and for the congnescenti. I agree we need it fixing broke clusters and the facility should be more amenable to admins.
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-1066:
-------------------------
Fix Version/s: 0.19.0
Pulling into 0.19.0. We can punt on it later. HBASE-1073 makes for a case where we have a single region of many out of sync. The 'repair' is currently shutting down the regionserver which is very disruptive, especially in small cluster with each node carrying many regions. Need a way for the master to send a CLOSE region from shell/UI. See how hard to do.
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Fix For: 0.19.0
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Updated: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack updated HBASE-1066:
-------------------------
Attachment: 1066.patch
Not fully tested. Adds enable/disable to the shell. Adds close of a region. Others, open and reassign are tricky. How do I say open a region? The scanners are supposed to be looking at the .META. and opening unopened tables. Same for reassign. Just close it and then when scanner runs, it'll be assigned elsewhere.
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Fix For: 0.19.0
>
> Attachments: 1066.patch
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Commented: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12658980#action_12658980 ]
stack commented on HBASE-1066:
------------------------------
{code}
hbase(main):002:0> tools
HBASE SURGERY TOOLS:
enable_region Enable a single region. For example,
hbase> enable_region 'REGIONNAME'
disable_region Disable a single region
close_region Close a single region. Optionally specify regionserver and
whether to 'clean' STARTCODE and SERVER from '.META.' Do not
clean the latter if you do not want the region reassigned.
Examples:
hbase> close_region 'REGIONNAME'
hbase> close_region 'REGIONNAME', false
hbase> close_region 'REGIONNAME', 'REGIONSERVER_IP:PORT'
Above commands are for 'experts'-only as misuse can damage an install
{code}
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Assignee: stack
> Fix For: 0.19.0
>
> Attachments: 1066-v3.patch, 1066-v4.patch, 1066.patch
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
[jira] Resolved: (HBASE-1066) Master should support
close/open/reassignment/enable/disable operations on individual regions
Posted by "stack (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-1066?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
stack resolved HBASE-1066.
--------------------------
Resolution: Fixed
Committed after removing notion of clean. It doesn't add anything. Can not clean the SERVER and STARTCODE but hbase will assign the region again anyways in the region close handler. Can add more tools easy enough as we go. This should do as a start.
> Master should support close/open/reassignment/enable/disable operations on individual regions
> ---------------------------------------------------------------------------------------------
>
> Key: HBASE-1066
> URL: https://issues.apache.org/jira/browse/HBASE-1066
> Project: Hadoop HBase
> Issue Type: Improvement
> Reporter: Andrew Purtell
> Assignee: stack
> Fix For: 0.19.0
>
> Attachments: 1066-v3.patch, 1066-v4.patch, 1066.patch
>
>
> The master should support close/open/reassignment/enable/disable operations on individual regions by way of a client API and corresponding shell command(s), and maybe also controls on the master UI.
>
> If one region out of 1000s is closed or offline, for example due to transient DFS problems, currently the whole table must be disabled, then reenabled to trigger reassignment and, hopefully, successful (re)open of the offline regions. The flurry of compactions this causes can exacerbate the underlying problem and actually make cluster state worse, as more DFS errors accumulate.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.