[
https://issues.apache.org/jira/browse/HBASE-1868?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
stack updated HBASE-1868:
-------------------------
Attachment: 1868.patch
An ugly patch that notices the condition where the regions to assign is <=
number to assign to balance. When it sees this condition, it just hands out a
single region and returns.
This code is beyond my powers of comprehension. When I see this kind of stuff
normally, I rip it out and start over with test code to verify its operation.
I don't want to do this in 0.20.1 bug fix context. Hence this compromise patch.
Trying to test it actually does the right thing before commiting.
> Spew about rebalancing but none done....
> ----------------------------------------
>
> Key: HBASE-1868
> URL: https://issues.apache.org/jira/browse/HBASE-1868
> Project: Hadoop HBase
> Issue Type: Bug
> Environment: 0.20.0 RC2
> Reporter: stack
> Fix For: 0.20.1
>
> Attachments: 1868.patch
>
>
> I'm seeing loads of this in logs:
> {code}
> 2009-09-24 21:27:22,130 DEBUG org.apache.hadoop.hbase.master.RegionManager:
> Server XX.XX.XX.100,20020,1253219583523 will be unloaded for balance. Server
> load: 5 avg: 3.78, regions can be moved: 4
> {code}
> Its like balancer is coming up w/ wrong answer to question... I don't see
> subsequent stuff going on... It does it over and over for hours.
> Then a split comes in and its seems to shake things up. I see it do a bunch
> of assigning.
> {code}
> 2009-09-24 21:41:02,784 INFO org.apache.hadoop.hbase.master.ServerManager:
> Processing MSG_REPORT_SPLIT: locations,,1253657949707: Daughters;
> locations,,1253
> 828460677,
> locations,http:\x2F\x2Fen.wikipedia.org\x2Fwiki\x2FLarry_Lucchino,1253828460677
> from aa0-009-2.u.powerset.com,20020,1253219584971; 1 of 3
> 2009-09-24 21:41:02,784 DEBUG org.apache.hadoop.hbase.master.RegionManager:
> Assigning for address: XX.XX.XX.6:20020, startcode: 1253219584971, load:
> (reque
> sts=5213, regions=3, usedHeap=114, maxHeap=2031): total nregions to assign=2,
> nregions to reach balance=4, isMetaAssign=false
> 2009-09-24 21:41:02,820 DEBUG org.apache.hadoop.hbase.master.RegionManager:
> Assigning for address: XX.XX.XX.96:20020, startcode: 1253219584175, load:
> (requ
> ests=12, regions=4, usedHeap=404, maxHeap=2031): total nregions to assign=2,
> nregions to reach balance=4, isMetaAssign=false
> ...
> {code}
> Then back to the 'will be unloaded'... message.
> A new split comes in and then the assigning gets triggered again... a few
> regions are opened but not enough.
> Eventually it goes back to 'normal' (average load went to 3.85 from 3.8?)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
