[
https://issues.apache.org/jira/browse/HIVE-9736?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mithun Radhakrishnan updated HIVE-9736:
---------------------------------------
Attachment: HIVE-9736.4.patch
The reason this patch didn't apply to trunk/ is that it depends on HIVE-9681. :/
Here's the patch that incorporates [~cnauroth]'s suggestion to combine
{{FsActions}} into a single instance, to reduce RPCs. I'm afraid that still
doesn't obviate the overload we added to {{Hadoop*Shims*}} since we needed a
new overload anyway to pluralize the {{FileStatus}} argument.
The compromise in the patch is to reduce the RPC calls, but keep the overload.
> StorageBasedAuthProvider should batch namenode-calls where possible.
> --------------------------------------------------------------------
>
> Key: HIVE-9736
> URL: https://issues.apache.org/jira/browse/HIVE-9736
> Project: Hive
> Issue Type: Bug
> Components: Metastore, Security
> Reporter: Mithun Radhakrishnan
> Assignee: Mithun Radhakrishnan
> Attachments: HIVE-9736.1.patch, HIVE-9736.2.patch, HIVE-9736.3.patch,
> HIVE-9736.4.patch
>
>
> Consider a table partitioned by 2 keys (dt, region). Say a dt partition could
> have 10000 associated regions. Consider that the user does:
> {code:sql}
> ALTER TABLE my_table DROP PARTITION (dt='20150101');
> {code}
> As things stand now, {{StorageBasedAuthProvider}} will make individual
> {{DistributedFileSystem.listStatus()}} calls for each partition-directory,
> and authorize each one separately. It'd be faster to batch the calls, and
> examine multiple FileStatus objects at once.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
