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 "Tim Armstrong (JIRA)" <ji...@apache.org> on 2018/08/21 18:14:00 UTC

[jira] [Commented] (IMPALA-7462) HASH JOIN takes lot of memory

    [ https://issues.apache.org/jira/browse/IMPALA-7462?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16587822#comment-16587822 ] 

Tim Armstrong commented on IMPALA-7462:
---------------------------------------

The straight_join hit forces the join order to match the textual order of tables in the FROM clause. It's expected that join order will affect memory consumption. A good join order often results in lower memory consumption but it's not the only thing we optimise for.

What's the ask here? If we think that it's generating a bad plan we need more information like an explain plan and ideally table stats to understand what happened.

> HASH JOIN takes lot of memory
> -----------------------------
>
>                 Key: IMPALA-7462
>                 URL: https://issues.apache.org/jira/browse/IMPALA-7462
>             Project: IMPALA
>          Issue Type: Bug
>          Components: Frontend
>    Affects Versions: Impala 2.12.0, Impala 2.13.0
>            Reporter: Pranay Singh
>            Priority: Major
>
> While running a complex SQL query the peak memory used up by HASH JOIN node is 270 GB
> Operator                 #Hosts   Avg Time   Max Time    #Rows  Est. #Rows   Peak Mem  Est. Peak Mem  
> ------------------------------------------------------------------------------------------------------------
> 12:HASH JOIN                  4    0.000ns    0.000ns        0      27.89M  270.38 GB       34.00 MB  ====>
> When the same query is run with STRAIGHT_JOIN hint the peak memory consumed is 3.25 GB
> Operator                 #Hosts   Avg Time   Max Time    #Rows  Est. #Rows   Peak Mem  Est. Peak Mem  
> ------------------------------------------------------------------------------------------------------------
> 11:HASH JOIN                  1   10s171ms   10s171ms    1.55M     262.03K    3.25 GB        1.60 GB ===> 



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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