You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues-all@impala.apache.org by "ASF subversion and git services (Jira)" <ji...@apache.org> on 2021/09/10 16:28:00 UTC
[jira] [Commented] (IMPALA-10883) Dense runtime profile missing
some fragment instance profile
[ https://issues.apache.org/jira/browse/IMPALA-10883?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17413270#comment-17413270 ]
ASF subversion and git services commented on IMPALA-10883:
----------------------------------------------------------
Commit 108cf8d07c82a179ede6d0b3385ddc4a6709822c in impala's branch refs/heads/master from Riza Suminto
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=108cf8d ]
IMPALA-10883: Do not override existing counters with empty profile
Some profile information was missing when gen_experimental_profile flag
is enabled. This is because profile aggregation in the coordinator does
not anticipate a partial update from a backend. From a backend
perspective, if a fragment instance has sent its final report, that
instance will not participate in the subsequent profile report.
Therefore, the aggregated counters that belong to the finished instance
will be empty. This patch adds empty checks in the aggregation of
input_profile_names_, TAggTimeSeriesCounter, and TAggEventSequence to
prevent the existing value from being overridden by an empty profile.
Testing:
- Add BE test CountersTest.PartialUpdate
Change-Id: I9bb179bf739ffaa4e5ec8dc911480ac835ae387f
Reviewed-on: http://gerrit.cloudera.org:8080/17819
Reviewed-by: Impala Public Jenkins <im...@cloudera.com>
Tested-by: Impala Public Jenkins <im...@cloudera.com>
> Dense runtime profile missing some fragment instance profile
> ------------------------------------------------------------
>
> Key: IMPALA-10883
> URL: https://issues.apache.org/jira/browse/IMPALA-10883
> Project: IMPALA
> Issue Type: Bug
> Components: Distributed Exec
> Affects Versions: Impala 4.0.0
> Reporter: Riza Suminto
> Assignee: Riza Suminto
> Priority: Major
>
> I ran TPC-DS Q78 in the following setup:
> * Cluster of 20 nodes
> * MT_DOP=12
> * --gen_experimental_profile=true
> When I check the query profile of the completed query, I notice that couple fragment are missing profile from some instances. This missing profile is indicated in the missing instances id like this:
> {code:java}
> Fragment F00 [228 instances]:
> Instances: Instance 204c719f48777536:b83eb0ed00000001 (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000002 (host=ia0306.halxg.cloudera.com:27000), , Instance 204c719f48777536:b83eb0ed00000004 (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000005 (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000006 (host=ia0306.halxg.cloudera.com:27000), , , Instance 204c719f48777536:b83eb0ed00000009 (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed0000000a (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed0000000b (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed0000000c (host=ia0306.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed0000000d (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed0000000e (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed0000000f (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000010 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000011 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000012 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000013 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000014 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000015 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000016 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000017 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000018 (host=ia0318.halxg.cloudera.com:27000), Instance 204c719f48777536:b83eb0ed00000019 (host=ia0322.halxg.cloudera.com:27000), , , , Instance 204c719f48777536:b83eb0ed0000001d (host=ia0322.halxg.cloudera.com:27000), , , , Instance 204c719f48777536:b83eb0ed00000021 (host=ia0322.halxg.cloudera.com:27000), , , , Instance 204c719f48777536:b83eb0ed00000025 (host=ia0317.halxg.cloudera.com:27000), , , , ...{code}
> Note several missing instances from host ia0306, ia0322, ia0317, and so on.
> I tried to tune FLAGS_status_report_interval_ms faster to every 500ms and there are even more instance profile missing. On the other hand, setting FLAGS_status_report_interval_ms=0 (only send final report) will result in complete runtime profile.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org