hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "mahesh kumar behera (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HIVE-20531) Repl load on cloud storage file system can skip redundant move or add partition tasks.
Date Mon, 01 Oct 2018 03:19:00 GMT

     [ https://issues.apache.org/jira/browse/HIVE-20531?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

mahesh kumar behera updated HIVE-20531:
---------------------------------------
    Attachment: HIVE-20531.07.patch

> Repl load on cloud storage file system can skip redundant move or add partition tasks.
> --------------------------------------------------------------------------------------
>
>                 Key: HIVE-20531
>                 URL: https://issues.apache.org/jira/browse/HIVE-20531
>             Project: Hive
>          Issue Type: Sub-task
>          Components: repl
>    Affects Versions: 4.0.0
>            Reporter: mahesh kumar behera
>            Assignee: mahesh kumar behera
>            Priority: Major
>              Labels: pull-request-available
>             Fix For: 4.0.0
>
>         Attachments: HIVE-20531.01.patch, HIVE-20531.02.patch, HIVE-20531.03.patch, HIVE-20531.04.patch,
HIVE-20531.05.patch, HIVE-20531.06.patch, HIVE-20531.07.patch
>
>
> In replication load, both add partition and insert operations are handled through import.
Import creates 3 major tasks. Copy, add partition and move. Copy does the copy of data from
source location to staging directory. Then add partition (which runs in parallel to copy)
creates the partition in meta store. Its a no op in case of insert and by the time this ddl
task is executed for insert partition would be already present. The third operation is move.
Which actually moves the file from staging directory to actual location. And then in case
of insert it adds the insert event to notification table. It does this for add partition operation
which is redundant as the event for add partition would have been written already by ddl task.
With the optimization to copy directly to actual table location in S3, move task can be avoided
for add partition operation replay and replay of insert need not create the add partition
(ddl) task.



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

Mime
View raw message