[
https://issues.apache.org/jira/browse/HBASE-22256?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16820728#comment-16820728
]
Nikhil Bafna commented on HBASE-22256:
--------------------------------------
Skip un-assign is required for this to work & makes the test case pass while
not breaking other UTs.
If i un-assign, the MoveRegionProcedure procedure fails because it expects the
region to be in OPEN state and the UT fails.
{code:java}
2019-04-18 10:15:16,068 WARN
[RpcServer.default.FPBQ.Fifo.handler=2,queue=0,port=64869]
master.HMaster(1682): Failed balance plan:
hri=2df26fa6b12b15fb35e3ff89ea84dac7, source=192.168.1.4,64881,1555562532668,
destination=, just skip it
org.apache.hadoop.hbase.client.DoNotRetryRegionException:
2df26fa6b12b15fb35e3ff89ea84dac7 is not OPEN;
regionState={2df26fa6b12b15fb35e3ff89ea84dac7 state=CLOSED, ts=1555562631969,
server=null}
at
org.apache.hadoop.hbase.master.procedure.AbstractStateMachineTableProcedure.checkOnline(AbstractStateMachineTableProcedure.java:194)
at
org.apache.hadoop.hbase.master.assignment.MoveRegionProcedure.<init>(MoveRegionProcedure.java:78)
at
org.apache.hadoop.hbase.master.assignment.AssignmentManager.createMoveRegionProcedure(AssignmentManager.java:741)
at
org.apache.hadoop.hbase.master.assignment.AssignmentManager.moveAsync(AssignmentManager.java:555)
at org.apache.hadoop.hbase.master.HMaster.balance(HMaster.java:1678)
{code}
> Enabling FavoredStochasticBalancer on existing cluster leaves regions
> unassigned
> --------------------------------------------------------------------------------
>
> Key: HBASE-22256
> URL: https://issues.apache.org/jira/browse/HBASE-22256
> Project: HBase
> Issue Type: Bug
> Components: Balancer
> Affects Versions: 2.1.3
> Reporter: Nikhil Bafna
> Priority: Major
> Attachments: HBASE-22256.master.001.patch
>
>
> This is related to HBASE-18349.
> The test that fails corresponding to this is
> TestFavoredStochasticLoadBalancer#testMisplacedRegions. When a region is
> misplaced w.r.t to the favored nodes, this balancer unassigns the region and
> the new RegionPlan has the source server as null leading to NPE later. This
> leaves the affected regions to be unassigned after the balancer run.
> This is problematic especially when moving from a different balancer to the
> FavoredStochasticLoadBalancer because all regions would be "misplaced" in the
> favored balancer's run.
> The fix is along the lines of HBASE-18602.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
