You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@hive.apache.org by "Laszlo Pinter (Jira)" <ji...@apache.org> on 2020/01/20 13:06:00 UTC
[jira] [Commented] (HIVE-21931) Slow compaction for tiny tables
[ https://issues.apache.org/jira/browse/HIVE-21931?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17019458#comment-17019458 ]
Laszlo Pinter commented on HIVE-21931:
--------------------------------------
[~csringhofer] Did you run compaction in blocking mode?
HIVE-22554 provides a way to configure the wait time out.
> Slow compaction for tiny tables
> -------------------------------
>
> Key: HIVE-21931
> URL: https://issues.apache.org/jira/browse/HIVE-21931
> Project: Hive
> Issue Type: Bug
> Affects Versions: 3.1.0
> Reporter: Csaba Ringhofer
> Priority: Major
> Labels: compaction
>
> I observed the issue in Impala development environment when (major) compacting insert_only transactional tables in Hive. The compaction could take ~10 minutes even when it only had to merge 2 rows from 2 inserts. The actual work was done much earlier, the new base file was correctly written to HDFS, and Hive seemed to wait without doing any work.
> The compactions are started manually, hive.compactor.initiator.on=false to avoid "surprise compaction" during tests.
> {code}
> hive.compactor.abortedtxn.threshold=1000
> hive.compactor.check.interval=300s
> hive.compactor.cleaner.run.interval=5000ms
> hive.compactor.compact.insert.only=true
> hive.compactor.crud.query.based=false
> hive.compactor.delta.num.threshold=10
> hive.compactor.delta.pct.threshold=0.1
> hive.compactor.history.reaper.interval=2m
> hive.compactor.history.retention.attempted=2
> hive.compactor.history.retention.failed=3
> hive.compactor.history.retention.succeeded=3
> hive.compactor.initiator.failed.compacts.threshold=2
> hive.compactor.initiator.on=false
> hive.compactor.max.num.delta=500
> hive.compactor.worker.threads=4
> hive.compactor.worker.timeout=86400s
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)