[
https://issues.apache.org/jira/browse/HIVE-9508?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Thiruvel Thirumoolan updated HIVE-9508:
---------------------------------------
Attachment: HIVE-9508.4.patch
Thanks [~vgumashta] for taking a look at this.
Uploading rebased patch for tests to run. I modified the LOG.info to debug
since I see the number of log entries to be high in our usage.
> MetaStore client socket connection should have a lifetime
> ---------------------------------------------------------
>
> Key: HIVE-9508
> URL: https://issues.apache.org/jira/browse/HIVE-9508
> Project: Hive
> Issue Type: Sub-task
> Components: CLI, Metastore
> Reporter: Thiruvel Thirumoolan
> Assignee: Thiruvel Thirumoolan
> Labels: metastore, rolling_upgrade
> Fix For: 1.2.0
>
> Attachments: HIVE-9508.1.patch, HIVE-9508.2.patch, HIVE-9508.3.patch,
> HIVE-9508.4.patch
>
>
> Currently HiveMetaStoreClient (or SessionHMSC) is connected to one Metastore
> server until the connection is closed or there is a problem. I would like to
> introduce the concept of a MetaStore client socket life time. The MS client
> will reconnect if the socket lifetime is reached. This will help during
> rolling upgrade of Metastore.
> When there are multiple Metastore servers behind a VIP (load balancer), it is
> easy to take one server out of rotation and wait for 10+ mins for all
> existing connections will die down (if the lifetime is 5mins say) and the
> server can be updated.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
