[
https://issues.apache.org/jira/browse/HBASE-1737?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jonathan Gray updated HBASE-1737:
---------------------------------
Attachment: HBASE-1737-v1.patch
Fixes small bug in RegionManager.
When looking at the most loaded server, we check to see if there is another
server that is underloaded. When we determine we should unassign from it
(numRegionsToClose = 0) then we determine how many to unassign. However, we
don't re-set numRegionsToClose to the number determine for reassignment (so it
stays at 0, and thus 0 are reassigned).
Also has a few small formatting changes and an extra variable in log line.
> Regions unbalanced when adding new node
> ---------------------------------------
>
> Key: HBASE-1737
> URL: https://issues.apache.org/jira/browse/HBASE-1737
> Project: Hadoop HBase
> Issue Type: Bug
> Affects Versions: 0.20.0
> Reporter: Jonathan Gray
> Priority: Blocker
> Fix For: 0.20.0
>
> Attachments: HBASE-1737-v1.patch
>
>
> When adding a new RegionServer to a cluster, the new RS will receive some
> regions but not enough to actually be considered balanced.
> To recreate, just take an RS offline, allow regions to be reassigned, and
> then bring it back up.
> Master will get itself into a broken, stuck state where it continuously
> outputs a line like this:
> {noformat}
> 2009-08-03 12:54:57,812 DEBUG org.apache.hadoop.hbase.master.RegionManager:
> Server dn4,60020,1249329081079 will be unloaded for balance. Server load: 341
> avg: 318.0, regions can be moved: 55
> {noformat}
> This line is output every 3 seconds and never stops until another RS
> joins/leaves the cluster.
> Making this a blocker because when your new RS only gets some regions (in my
> case, about half as many as it should have), then all new regions will be
> assigned to that RS. This basically destroys any possibility for good load
> distribution with new data.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
