[
https://issues.apache.org/jira/browse/HBASE-17942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16753406#comment-16753406
]
Nihal Jain commented on HBASE-17942:
------------------------------------
I discovered a strange behavior while writing tests in the above patch.
It is easily reproducible: Just add following statement at the end of testcase:
{{TestSplitTableRegionProcedure.testSplitTableRegionEmptyDaughter()}}
{code:java}
UTIL.getAdmin().modifyTable(UTIL.getAdmin().getDescriptor(tableName));
{code}
The modify procedure succeeds but the future never returns and the test-case
keeps running.
See the tail of the log:
{noformat}
2019-01-27 20:23:25,302 DEBUG [PEWorker-1] procedure2.RootProcedureState(153):
Add procedure pid=14, state=SUCCESS, hasLock=true; ModifyTableProcedure
table=testSplitTableRegionDeletedRowsDaughter as the 27th rollback step
2019-01-27 20:23:25,319 DEBUG [PEWorker-1]
procedure.MasterProcedureScheduler(351): Add
TableQueue(testSplitTableRegionDeletedRowsDaughter, xlock=false sharedLock=0
size=0) to run queue because: pid=14, state=SUCCESS, hasLock=false;
ModifyTableProcedure table=testSplitTableRegionDeletedRowsDaughter released the
exclusive lock
2019-01-27 20:23:25,320 INFO [PEWorker-1] procedure2.ProcedureExecutor(1410):
Finished pid=14, state=SUCCESS, hasLock=false; ModifyTableProcedure
table=testSplitTableRegionDeletedRowsDaughter in 1.0010sec
2019-01-27 20:23:25,328 DEBUG
[RpcServer.default.FPBQ.Fifo.handler=3,queue=0,port=40235]
master.MasterRpcServices(1182): Checking to see if procedure is done pid=14
2019-01-27 20:23:27,168 WARN [HBase-Metrics2-1] impl.MetricsConfig(125):
Cannot locate configuration: tried
hadoop-metrics2-hbase.properties,hadoop-metrics2.properties
2019-01-27 20:23:27,503 INFO
[org.apache.hadoop.hdfs.server.blockmanagement.BlockManager$ReplicationMonitor@4e87e6e6]
blockmanagement.BlockManager(3480): BLOCK* BlockManager: ask 127.0.0.1:36701
to delete [blk_1073741834_1010]
2019-01-27 20:23:30,504 INFO
[org.apache.hadoop.hdfs.server.blockmanagement.BlockManager$ReplicationMonitor@4e87e6e6]
blockmanagement.BlockManager(3480): BLOCK* BlockManager: ask 127.0.0.1:37975
to delete [blk_1073741834_1010]
{noformat}
> Disable region splits and merges per table
> ------------------------------------------
>
> Key: HBASE-17942
> URL: https://issues.apache.org/jira/browse/HBASE-17942
> Project: HBase
> Issue Type: New Feature
> Reporter: Vladimir Rodionov
> Assignee: Nihal Jain
> Priority: Major
> Attachments: HBASE-17942.master.001.patch
>
>
> Snapshot of a large tables usually fails when region split/merge happens
> during snapshot. HBASE-15128 has introduced enable/disable switch for split
> /merges, but only cluster - wide which is not always good in a large
> deployments. The new feature will improve snapshot stability ands performance
> for a single table.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
