[
https://issues.apache.org/jira/browse/HIVE-7140?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14020866#comment-14020866
]
Ashutosh Chauhan commented on HIVE-7140:
----------------------------------------
[~brocknoland] Can you post some details about the setup where this occurs?
Like which api calls were these which were taking long time to complete. How
long they took? Was it remote metastore? How many partitions it had? Was
directSql successful or query on RDBMS was done using DataNucleus? How much
time was spent on RDBMS Vs how much in Metastore code?
May be we should also investigate in parallel why this is happening. There
looks like an opportunity to improve our metastore codebase here.
> Bump default hive.metastore.client.socket.timeout to 5 minutes
> --------------------------------------------------------------
>
> Key: HIVE-7140
> URL: https://issues.apache.org/jira/browse/HIVE-7140
> Project: Hive
> Issue Type: Improvement
> Affects Versions: 0.10.0, 0.12.0, 0.13.0
> Reporter: Brock Noland
> Assignee: Brock Noland
> Attachments: HIVE-7140.patch
>
>
> The issue is that OOTB clients often face timeouts when using HMS since many
> operations in the HMS completes are long running (e.g. many operations on a
> table with many partitions). A few supporting pieces of information:
> * The default value of hive.metastore.client.socket.timeout is 20 seconds.
> * Since the timeout is client only, the server happy continues doing the
> requested work
> * Clients retry after a small delay to perform the requested work again,
> often while the server is still trying to complete the original request
> * A few tests have actually increased this value in order to pass reliably.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
