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 2023/06/02 16:22:00 UTC

[jira] [Commented] (IMPALA-12142) Default fetch_size of 10240 is suboptimal

    [ https://issues.apache.org/jira/browse/IMPALA-12142?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17728815#comment-17728815 ] 

ASF subversion and git services commented on IMPALA-12142:
----------------------------------------------------------

Commit adfa6c83ec8ca371798b17021d49a8eacd954faf in impala's branch refs/heads/master from Csaba Ringhofer
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=adfa6c83e ]

IMPALA-12142: Decrease default fetch_size to 8192 in impala-shell

The previous fetch_size of 10240 turned out to be suboptimal for HS2
server side, likely because it leads to overallocation in result
'std::vector's. Changed to the closest power of 2 size (8192).

With this change RowMaterializationTimer decreased from 3.4s to 2.7s
for "SELECT * FROM tpch_parquet.lineitem".

Change-Id: I34973cb705db53c496b9944c74995b45cf720d46
Reviewed-on: http://gerrit.cloudera.org:8080/19965
Reviewed-by: Kurt Deschler <kd...@cloudera.com>
Reviewed-by: Daniel Becker <da...@cloudera.com>
Tested-by: Impala Public Jenkins <im...@cloudera.com>


> Default fetch_size of 10240 is suboptimal
> -----------------------------------------
>
>                 Key: IMPALA-12142
>                 URL: https://issues.apache.org/jira/browse/IMPALA-12142
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Clients
>            Reporter: Csaba Ringhofer
>            Priority: Major
>              Labels: performance
>             Fix For: Impala 4.3.0
>
>
> While working on IMPALA-12138 it turned out that the default settings of batch_size=1024 and  fetch_size=10240 are not ideal for the coordinator (RowMaterializationTimer). My guess for the cause is that HS2 results vectors are rounded up to power of 2 sizes leading to extra allocations and copying.
> query: select * from tpch_parquet.lineitem
> RowMaterializationTimer (before and after IMPALA-12138)
> fetch_size=10240: 3.6s ->3.2s
> fetch_size=8192: 2.8s->2.6s



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

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org