[jira] [Updated] (HIVE-15250) Reuse partitions info generated in MoveTask to its subscribers (StatsTask)

["Rajesh Balamohan (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/HIVE-15250?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Rajesh Balamohan updated HIVE-15250:
------------------------------------
    Attachment: HIVE-15250.1.patch

Observed 10% improvement on top of HIVE-15240 in 5 node cluster running on EC2 in populating web_returns dataset in tpc-ds-200.

> Reuse partitions info generated in MoveTask to its subscribers (StatsTask)               
> -----------------------------------------------------------------------------------------
>
>                 Key: HIVE-15250
>                 URL: https://issues.apache.org/jira/browse/HIVE-15250
>             Project: Hive
>          Issue Type: Improvement
>          Components: Metastore
>            Reporter: Rajesh Balamohan
>            Priority: Minor
>         Attachments: HIVE-15250.1.patch
>
>
> When dynamic partitions are enabled, {{StatsTask}} loads partition information by querying metastore. In cases like {{insert overwrite table}}, this can be expensive operation depending on the number of partitions involved (for e.g, in tpcds populating web_returns table would incur 2184 DB calls just on this function).
> It would be good to pass on the partition information generated in MoveTask to its subscribers to reduce the number of DB calls.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)