phoenix-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "James Taylor (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PHOENIX-4751) Support client-side hash aggregation with SORT_MERGE_JOIN
Date Thu, 24 May 2018 05:42:00 GMT

    [ https://issues.apache.org/jira/browse/PHOENIX-4751?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16488452#comment-16488452
] 

James Taylor commented on PHOENIX-4751:
---------------------------------------

Yes, you're right [~sangudi] - since the aggregation is done after the join, the SORT-MERGE-JOIN
will be completely done on the client (and hence would benefit from being able to do a hash
aggregation instead). The SpillableGroupByCache is used on the server-side hash aggregation.
It would work on the client side as well (if you can write to the file system). It basically
tries to do everything in memory and then past a memory threshold will spill to disk.

> Support client-side hash aggregation with SORT_MERGE_JOIN
> ---------------------------------------------------------
>
>                 Key: PHOENIX-4751
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-4751
>             Project: Phoenix
>          Issue Type: Improvement
>    Affects Versions: 4.14.0, 4.13.1
>            Reporter: Gerald Sangudi
>            Priority: Major
>
> A GROUP BY that follows a SORT_MERGE_JOIN should be able to use hash aggregation in
some cases, for improved performance.
> When a GROUP BY follows a SORT_MERGE_JOIN, the GROUP BY does not use hash aggregation.
It instead performs a CLIENT SORT followed by a CLIENT AGGREGATE. The performance can be
improved if (a) the GROUP BY output does not need to be sorted, and (b) the GROUP BY input is
large enough and has low cardinality.
> The hash aggregation can initially be a hint. Here is an example from Phoenix 4.13.1
that would benefit from hash aggregation if the GROUP BY input is large with low cardinality.
> CREATE TABLE unsalted (
>  keyA BIGINT NOT NULL,
>  keyB BIGINT NOT NULL,
>  val SMALLINT,
>  CONSTRAINT pk PRIMARY KEY (keyA, keyB)
>  );
> EXPLAIN
>  SELECT /*+ USE_SORT_MERGE_JOIN */ 
>  t1.val v1, t2.val v2, COUNT(\*) c 
>  FROM unsalted t1 JOIN unsalted t2 
>  ON (t1.keyA = t2.keyA) 
>  GROUP BY t1.val, t2.val;
>  +-------------------------------------------------------------+----------------++------------------+
> |PLAN|EST_BYTES_READ|EST_ROWS_READ| |
> +-------------------------------------------------------------+----------------++------------------+
> |SORT-MERGE-JOIN (INNER) TABLES|null|null| |
> |    CLIENT 1-CHUNK PARALLEL 1-WAY FULL SCAN OVER UNSALTED|null|null| |
> |AND|null|null| |
> |    CLIENT 1-CHUNK PARALLEL 1-WAY FULL SCAN OVER UNSALTED|null|null| |
> |CLIENT SORTED BY [TO_DECIMAL(T1.VAL), T2.VAL]|null|null| |
> |CLIENT AGGREGATE INTO DISTINCT ROWS BY [T1.VAL, T2.VAL]|null|null| |
> +-------------------------------------------------------------+----------------++------------------+



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

Mime
View raw message