You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hbase.apache.org by "David S. Wang (JIRA)" <ji...@apache.org> on 2012/06/22 19:28:43 UTC
[jira] [Created] (HBASE-6258) Backport some region splitting fixes
into 0.90.7
David S. Wang created HBASE-6258:
------------------------------------
Summary: Backport some region splitting fixes into 0.90.7
Key: HBASE-6258
URL: https://issues.apache.org/jira/browse/HBASE-6258
Project: HBase
Issue Type: Bug
Components: regionserver
Affects Versions: 0.90.6
Reporter: David S. Wang
Assignee: David S. Wang
Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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] [Updated] (HBASE-6258) Backport some region splitting fixes
into 0.90.7
Posted by "David S. Wang (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
David S. Wang updated HBASE-6258:
---------------------------------
Status: Patch Available (was: Open)
Patch for the backport into 0.90.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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] [Updated] (HBASE-6258) Backport some region splitting fixes
into 0.90.7
Posted by "David S. Wang (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
David S. Wang updated HBASE-6258:
---------------------------------
Attachment: HBASE-4816+4881+5189+6158.patch
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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-6258) Backport some region splitting
fixes into 0.90.7
Posted by "Jonathan Hsieh (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13402882#comment-13402882 ]
Jonathan Hsieh commented on HBASE-6258:
---------------------------------------
I think part of HBASE-4816 seems low risk (extra closing checks on splitrequest) but the part eventually gets rooted to HBASE-1502 (removing heartbeats) so sounds like it is best to probably avoid.
In discussion you mentioned that HBASE-4881 was pointless because the exception signature changed between 0.90 and 0.92 and suggested not including it. I'd be fine with that.
HBASE-6158 is embarrassing and a simple fix, but it sounds best to avoid porting the fix to avoid the potential for compatibility problems. The good news is that this one has an easy work around (don't use merges or splits as colfams). Instead can we add a patch to warn if a user tries to create/alter tables to have 'merges' or 'splits' as column families names?
HBASE-5189: I think metrics are always useful and seem low risk (I don't think we have a strict contract staying that adding metrics breaks compatibility). Since this is 0.94, ff we backport this to 0.90 we'd need to get into 0.92 as well.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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-6258) Backport some region splitting
fixes into 0.90.7
Posted by "David S. Wang (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13399934#comment-13399934 ]
David S. Wang commented on HBASE-6258:
--------------------------------------
JD,
Thanks for the timely comments.
Yes, including SplitRequest is an error on my end. I will remove it and apply the few changes that went into it accordingly.
Changing the SPLITDIR is from HBASE-6158, which went into 0.92.2 and 0.94.1. I think you bring up a good point, and it sounds like we will have the same problem in those releases as well. Perhaps the original patch needs to be changed to handle this case for the aforementioned releases.
I'll take a look at what region splitting tests we have today and add as necessary. It looks like we need to add at least a test for when SPLITDIR is changed.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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-6258) Backport some region splitting
fixes into 0.90.7
Posted by "Hadoop QA (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13399460#comment-13399460 ]
Hadoop QA commented on HBASE-6258:
----------------------------------
-1 overall. Here are the results of testing the latest attachment
http://issues.apache.org/jira/secure/attachment/12533075/HBASE-4816%2B4881%2B5189%2B6158.patch
against trunk revision .
+1 @author. The patch does not contain any @author tags.
-1 tests included. The patch doesn't appear to include any new or modified tests.
Please justify why no new tests are needed for this patch.
Also please list what manual steps were performed to verify this patch.
-1 patch. The patch command could not apply the patch.
Console output: https://builds.apache.org/job/PreCommit-HBASE-Build/2224//console
This message is automatically generated.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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-6258) Backport some region splitting
fixes into 0.90.7
Posted by "Jean-Daniel Cryans (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13399470#comment-13399470 ]
Jean-Daniel Cryans commented on HBASE-6258:
-------------------------------------------
Is the inclusion of SplitRequest an error? It's not being used.
I wonder what changing the SPLITDIR does in the context of a rolling restart. It seems that by calling SplitTransaction.cleanupAnySplitDetritus on an upgraded server you'd miss the old folder.
I haven't followed the 4 jiras closely but I don't see any unit tests in there and this worries me considering that we want to include this in a branch that has been through many bug fix releases eg it should be really stable.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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] [Updated] (HBASE-6258) Backport some region splitting fixes
into 0.90.7
Posted by "David S. Wang (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
David S. Wang updated HBASE-6258:
---------------------------------
Resolution: Won't Fix
Status: Resolved (was: Patch Available)
There's not enough value here to do the backports into 0.90.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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-6258) Backport some region splitting
fixes into 0.90.7
Posted by "David S. Wang (JIRA)" <ji...@apache.org>.
[ https://issues.apache.org/jira/browse/HBASE-6258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13399946#comment-13399946 ]
David S. Wang commented on HBASE-6258:
--------------------------------------
I see this comment from HBASE-6158: https://issues.apache.org/jira/browse/HBASE-6158?focusedCommentId=13288890&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13288890
> Is this change incompatible for rolling upgrade?
I don't think so. These are temporary work folders which get cleaned up routinely and particularly at Region Server (re)start for each region through HRegion.initialize().
However, a possible side effect could be that "merges" and "splits" folders created by pre-upgrade code (which somehow did not get cleaned up during a shutdown) may continue to exist on the file system as the cleanup code would no longer be looking for them. But this is better fixed in the upgrade scripts by deleting these folders if they are found to be a work folder instead of a CF container folder.
I'll continue the conversation over on HBASE-6158.
> Backport some region splitting fixes into 0.90.7
> ------------------------------------------------
>
> Key: HBASE-6258
> URL: https://issues.apache.org/jira/browse/HBASE-6258
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.90.6
> Reporter: David S. Wang
> Assignee: David S. Wang
> Attachments: HBASE-4816+4881+5189+6158.patch
>
>
> Issue tracking backport of some relatively small region splitting fixes into 0.90.7:
> HBASE-4816: Regionserver wouldn't go down because split happened exactly at same time we issued bulk user region close call on our way out - fixed in 0.92
> HBASE-4881: Unhealthy region is on service caused by rollback of region splitting - fixed in 0.92
> HBASE-5189: Add metrics to keep track of region-splits in RS - fixed in 0.94
> HBASE-6158: Data loss if the words 'merges' or 'splits' are used as Column Family name - fixed in 0.92 and 0.94
--
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