[
https://issues.apache.org/jira/browse/KAFKA-1623?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14121473#comment-14121473
]
Guozhang Wang commented on KAFKA-1623:
--------------------------------------
Thanks for pointing it out Shlomi. I think this can be fixed by reducing the
select time (e.g. 100ms) and setting a proper tcp backlog time long enough so
that the thread will be awaken quickly once the connection is opened. The
reason I prefer not to make them configurable is that in practice a proper
universal value for such configs may be good enough for all scenarios and
adding more such configs may swamped the number of switches displayed to users.
> kafka is sometimes slow to accept connections
> ---------------------------------------------
>
> Key: KAFKA-1623
> URL: https://issues.apache.org/jira/browse/KAFKA-1623
> Project: Kafka
> Issue Type: Bug
> Components: core
> Affects Versions: 0.9.0
> Reporter: Shlomi Hazan
> Labels: performance
>
> from SocketServer.scala:144
> the Acceptor can wait up to 500 millis before processing the accumulated FDs.
> Also, the backlog of the acceptor socket seems not to be defined, which may
> be problematic if all 500 millis are elapsed before the thread awakes.
> setting the backlog is doable using the proper ServerSocket Ctor, and maybe
> better be provisioned via configuration.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
