[jira] [Commented] (HIVE-24313) Optimise stats collection for file sizes on cloud storage

["Manish Maheshwari (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HIVE-24313?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17221306#comment-17221306 ] 

Manish Maheshwari commented on HIVE-24313:
------------------------------------------

Also it would be good to persist the stats collected into HMS to ensure that they can be used for subsequent queries.

> Optimise stats collection for file sizes on cloud storage
> ---------------------------------------------------------
>
>                 Key: HIVE-24313
>                 URL: https://issues.apache.org/jira/browse/HIVE-24313
>             Project: Hive
>          Issue Type: Improvement
>          Components: HiveServer2
>            Reporter: Rajesh Balamohan
>            Priority: Major
>
> When stats information is not present (e.g external table), RelOptHiveTable computes basic stats at runtime.
> Following is the codepath.
> [https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/optimizer/calcite/RelOptHiveTable.java#L598]
> {code:java}
> Statistics stats = StatsUtils.collectStatistics(hiveConf, partitionList,
>                 hiveTblMetadata, hiveNonPartitionCols, nonPartColNamesThatRqrStats, colStatsCached,
>                 nonPartColNamesThatRqrStats, true);
>  {code}
> [https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/stats/StatsUtils.java#L322]
> {code:java}
> for (Partition p : partList.getNotDeniedPartns()) {
>         BasicStats basicStats = basicStatsFactory.build(Partish.buildFor(table, p));
>         partStats.add(basicStats);
>       }
>  {code}
> [https://github.com/apache/hive/blob/master/ql/src/java/org/apache/hadoop/hive/ql/stats/BasicStats.java#L205]
>  
> {code:java}
> try {
>             ds = getFileSizeForPath(path);
>           } catch (IOException e) {
>             ds = 0L;
>           }
>  {code}
>  
> For a table & query with large number of partitions, this takes long time to compute statistics and increases compilation time.  It would be good to fix it with "ForkJoinPool" ( partList.getNotDeniedPartns().parallelStream().forEach((p) )
>  
>  



--
This message was sent by Atlassian Jira
(v8.3.4#803005)