[
https://issues.apache.org/jira/browse/HIVE-9253?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14264610#comment-14264610
]
Dong Chen commented on HIVE-9253:
---------------------------------
This fix makes server know its client timeout, when corresponding TSocket
thread in server is blocking in read and have not read msg during timeout
period.
When I tested this patch in local env, I found it could go further.
The server continues the original requested work, since its timeout is only
detect when blocked in read. So the long running work is still going on in
server. And then arrive the blocked read in connected socket thread.
The control of server side connected socket timeout is in thrift TServer and
TServerSocket. If we want to leverage the socket timeout mechanism and stop
invalid requested work immediately, I think we might need customize a Thrift
Server and ServerSocket for more control (still not sure it is feasible).
Any thought?
> MetaStore server should detect client timeout
> ---------------------------------------------
>
> Key: HIVE-9253
> URL: https://issues.apache.org/jira/browse/HIVE-9253
> Project: Hive
> Issue Type: Sub-task
> Components: Metastore
> Reporter: Dong Chen
> Assignee: Dong Chen
> Attachments: HIVE-9253.patch
>
>
> In the description of HIVE-7195, one issue is that MetaStore client timeout
> is quite dumb. The client will timeout and the server has no idea the client
> is gone.
> The server should detect client timeout.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
