[
https://issues.apache.org/jira/browse/HIVE-15572?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15818466#comment-15818466
]
Yongzhi Chen commented on HIVE-15572:
-------------------------------------
1) Add data structure to pass driverstate around.
2) Improve ZooKeeperHiveLockManager response time for canceling by checking
driver state before each acquiring lock call to zookeeper.
> Improve the response time for query canceling when it happens during
> acquiring locks
> ------------------------------------------------------------------------------------
>
> Key: HIVE-15572
> URL: https://issues.apache.org/jira/browse/HIVE-15572
> Project: Hive
> Issue Type: Improvement
> Reporter: Yongzhi Chen
> Assignee: Yongzhi Chen
>
> When query canceling command sent during Hive Acquire locks (from zookeeper),
> hive will finish acquiring all the locks and release them. As it is shown in
> the following log:
> It took 165 s to finish acquire the lock,then spend 81s to release them.
> We can improve the performance by not acquiring any more locks and releasing
> held locks when the query canceling command is received.
> {noformat}
> Background-Pool: Thread-224]: <PERFLOG method=TimeToSubmit
> from=org.apache.hadoop.hive.ql.Driver>
> 2017-01-03 10:50:35,413 INFO org.apache.hadoop.hive.ql.log.PerfLogger:
> [HiveServer2-Background-Pool: Thread-224]: <PERFLOG
> method=acquireReadWriteLocks from=org.apache.hadoop.hive.ql.Driver>
> 2017-01-03 10:51:00,671 INFO org.apache.hadoop.hive.ql.log.PerfLogger:
> [HiveServer2-Background-Pool: Thread-218]: </PERFLOG
> method=acquireReadWriteLocks start=1483469295080 end=1483469460671
> duration=165591 from=org.apache.hadoop.hive.ql.Driver>
> 2017-01-03 10:51:00,672 INFO org.apache.hadoop.hive.ql.log.PerfLogger:
> [HiveServer2-Background-Pool: Thread-218]: <PERFLOG method=Driver.execute
> from=org.apache.hadoop.hive.ql.Driver>
> 2017-01-03 10:51:00,672 ERROR org.apache.hadoop.hive.ql.Driver:
> [HiveServer2-Background-Pool: Thread-218]: FAILED: query select count(*) from
> manyparttbl has been cancelled
> 2017-01-03 10:51:00,673 INFO org.apache.hadoop.hive.ql.log.PerfLogger:
> [HiveServer2-Background-Pool: Thread-218]: <PERFLOG method=releaseLocks
> from=org.apache.hadoop.hive.ql.Driver>
> 2017-01-03 10:51:40,755 INFO org.apache.hadoop.hive.ql.log.PerfLogger:
> [HiveServer2-Background-Pool: Thread-215]: </PERFLOG method=releaseLocks
> start=1483469419487 end=1483469500755 duration=81268
> from=org.apache.hadoop.hive.ql.Driver>
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
