[
https://issues.apache.org/jira/browse/HIVE-1580?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Joydeep Sen Sarma updated HIVE-1580:
------------------------------------
Attachment: hive-1580.1.patch
cleanup multiple calls to getCounters (which turns out to be really expensive
call in JT) and don't print non-fatal stack traces to console.
> cleanup ExecDriver.progress
> ---------------------------
>
> Key: HIVE-1580
> URL: https://issues.apache.org/jira/browse/HIVE-1580
> Project: Hadoop Hive
> Issue Type: Bug
> Components: Query Processor
> Reporter: Joydeep Sen Sarma
> Assignee: Joydeep Sen Sarma
> Attachments: hive-1580.1.patch
>
>
> a few problems:
> - if a job is retired - then counters cannot be obtained and a stack trace is
> printed out (from history code). this confuses users
> - too many calls to getCounters. after a job has been detected to be finished
> - there are quite a few more calls to get the job status and the counters. we
> need to figure out a way to curtail this - in busy clusters the gap between
> the job getting finished and the hive client noticing is very perceptible and
> impacts user experience.
> calls to getCounters are very expensive in 0.20 as they grab a jobtracker
> global lock (something we have fixed internally at FB)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
