[
https://issues.apache.org/jira/browse/BOOKKEEPER-602?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13719292#comment-13719292
]
Sijie Guo commented on BOOKKEEPER-602:
--------------------------------------
The reason why we put getAddEntryTimeout to 1 is to put the best parameter for
most of cases. so the user doesn't need to tune it too much. based on this
consideration, we'd prefer putting the tuned value as default value and having
TestBKConfiguration to handle low-throughput case (this TestBKConfiguration
also exists in the journal improvements we made in BOOKKEEPER-657). if you feel
strongly about it, I could change. let me know.
1 second is based on performance evaluation, 1) we don't want a slow add
request to cause too much pending requests accumulated in client (which cause
bad gc behavior) 2) for latency consideration.
the failure test is also related to the configuration setting. I forgot to
bring the changes for hedwig when generating the patch. will add soon.
> we should have request timeouts rather than channel timeout in
> PerChannelBookieClient
> -------------------------------------------------------------------------------------
>
> Key: BOOKKEEPER-602
> URL: https://issues.apache.org/jira/browse/BOOKKEEPER-602
> Project: Bookkeeper
> Issue Type: Bug
> Affects Versions: 4.2.0, 4.2.1
> Reporter: Sijie Guo
> Assignee: Sijie Guo
> Fix For: 4.3.0
>
> Attachments: BOOKKEEPER-602.diff
>
>
> currently we only have readTimeout in netty channel, it timeouts only when
> there is no activities in that channel, but it can't track timeouts of
> individual requests. if a channel continues having read entry activities, it
> might shadow a slow add entry response, which is bad impacting add latency.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
