[
https://issues.apache.org/jira/browse/HIVE-14876?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15563417#comment-15563417
]
Thejas M Nair commented on HIVE-14876:
--------------------------------------
Two suggestions -
1. I think we should have a different "max" fetch size and a "default" fetch
size.
The "max" would be a guard rail to prevent accidently adding extra zeroes to
the param, resulting in an OOM in HS2. This config in this case is called
"hive.server2.thrift.resultset.max.fetch.size" , it would be good to change the
description of that and use a new config param for purpose of this jira
("hive.server2.thrift.resultset.default.fetch.size").
2. Default to 1k for server -
We have 3 default values getting currently used -
10k from ODBC driver (overrides server settings)
1k from JDBC driver (overrides server settings)
100 in server (CliDriver uses Operation.DEFAULT_FETCH_MAX_ROWS)
1k seems like a reasonable value to me. If we assume 1 KB average record size,
it comes to around 1 MB per call. That should keep the network/rpc overheads
low enough. Unless we have data that shows 10 KB provides significantly better
performance, I think we can use 1k on the server side as well.
> make the number of rows to fetch from various HS2 clients/servers configurable
> ------------------------------------------------------------------------------
>
> Key: HIVE-14876
> URL: https://issues.apache.org/jira/browse/HIVE-14876
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-14876.patch
>
>
> Right now, it's hardcoded to a variety of values
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)