hive-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mithun Radhakrishnan (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HIVE-17188) ObjectStore runs out of memory for large batches of addPartitions().
Date Thu, 27 Jul 2017 22:25:02 GMT

    [ https://issues.apache.org/jira/browse/HIVE-17188?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16104014#comment-16104014
] 

Mithun Radhakrishnan commented on HIVE-17188:
---------------------------------------------

@[~vihangk1]: Thank you for your attention. :]

bq. Can you please update the patch with HIVE specific JIRA number and description of this
JIRA as per our convention?
Sorry, it's been a while, so perhaps you could clarify for me. My memory of the convention
is that patches are named {{HIVE-<jira-number>.<version-of-patch>.patch}}. If
the patch is a port to another branch, then it's {{HIVE-<jira-number>.<branch>.patch}}.

>From perusing the JIRAs included in [the Hive 2.2 release|https://issues.apache.org/jira/projects/HIVE/versions/12335837],
this seems like the format of choice. Could you please clarify what I'm missing?

bq. You can add a line in the description where this patch was cherry-picked from I you like..
This is a port from Yahoo's internal production branch. The commit dates back to April of
2014. :]

bq. If there are hundreds of partitions being added, aren't they already in memory in the
{{List<Partition>}} parts object?
A fair question. :] I can try answer this, although [~cdrome] and [~thiruvel] are really the
experts on this one. 
The problem being addressed here isn't so much with the size of the hundreds of {{Partition}}
objects, but the cruft that builds with the {{PersistenceManager}}, in the JDO layer, as confirmed
through memory-profiling.

Our larger commit also plugged leaks from neglecting to call {{Query::close()}}, etc. It looks
like those have independently been solved already.


> ObjectStore runs out of memory for large batches of addPartitions().
> --------------------------------------------------------------------
>
>                 Key: HIVE-17188
>                 URL: https://issues.apache.org/jira/browse/HIVE-17188
>             Project: Hive
>          Issue Type: Bug
>          Components: Metastore
>    Affects Versions: 2.2.0
>            Reporter: Mithun Radhakrishnan
>            Assignee: Chris Drome
>         Attachments: HIVE-17188.1.patch
>
>
> For large batches (e.g. hundreds) of {{addPartitions()}}, the {{ObjectStore}} runs out
of memory. Flushing the {{PersistenceManager}} alleviates the problem.
> (Raising this on behalf of [~cdrome] and [~thiruvel].)



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message