[ 
https://issues.apache.org/jira/browse/IMPALA-15265?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18104936#comment-18104936
 ] 

Joe McDonnell commented on IMPALA-15265:
----------------------------------------

This started with this commit: 
[https://github.com/apache/impala/commit/848ef4735f49e9707456ffd0f0f2079f2e83414c]

Basically, InternalServerTest's QueryTimeout test is running a SQL with a 
"where id=sleep(2000000)" but also a corresponding FETCH_ROWS_TIMEOUT_MS=1. The 
idea is that it hits that timeout and gets cancelled. The switch to Parquet 
means that we go through the logic for dictionary filters. This does a check 
whether evaluating the predicate on NULL is true. This is doing the real sleep 
during planning without the FETCH_ROWS_TIMEOUT_MS kicking in. We can avoid this 
by forcing it to use text tables or disabling dictionary filtering. In general, 
we may want to have a way to evaluate an expression where it won't respect any 
sleeps.

> internal-server-test slowed down considerably and now takes 30+ minutes
> -----------------------------------------------------------------------
>
>                 Key: IMPALA-15265
>                 URL: https://issues.apache.org/jira/browse/IMPALA-15265
>             Project: IMPALA
>          Issue Type: Bug
>          Components: Backend
>    Affects Versions: Impala 5.0.0
>            Reporter: Joe McDonnell
>            Priority: Critical
>
> internal-server-test is taking a long long time now (35+ minutes):
> {noformat}
> 04:42:28         Start  90: internal-server-test
> 05:18:15  90/166 Test  #90: internal-server-test ...................   Passed 
>  2146.37 sec{noformat}
> It used to run much faster (2.5 minutes):
> {noformat}
> 19:55:00         Start  89: internal-server-test
> 19:57:26  89/164 Test  #89: internal-server-test ...................   Passed 
>  146.09 sec{noformat}
> We should figure out what happened and when. This is relatively recent.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to