kafka-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Caesar Ralf Franz Hoppen <caesar.r...@gmail.com>
Subject Re: (auto.commit.interval = heartbeat.interval) => infinite rebalance loop
Date Mon, 04 Apr 2016 11:15:53 GMT
Yes, it seems that it can be something related to that. Well, glad that
someone already mapped it. Waiting for the fix and see if the solves my
issue.

On 31 March 2016 at 20:49, Jason Gustafson <jason@confluent.io> wrote:

> This could be caused by a bug in our client's network layer which
> occasionally prevents multiple requests from being sent at the same time.
> Usually for both heartbeats and periodic commits, we expect the next one to
> be sent successfully, so it hasn't been a big problem. However, when the
> intervals match, we might be hitting the same problem over and over. Here
> is a link to a JIRA where we're trying to address the underlying problem:
> https://issues.apache.org/jira/browse/KAFKA-3488.
>
> -Jason
>
> On Thu, Mar 31, 2016 at 1:16 AM, Caesar Ralf Franz Hoppen <
> caesar.ralf@gmail.com> wrote:
>
> > Hej,
> >
> > So there seems to be a weird problem that happens if you happen to setup
> > your heartbeat.interval.ms equal to your autocommit.interval.ms.
> >
> > From what I see in the logs, it never successfully send a heartbeat when
> > there was a commit before, so the consumer expires after the
> > session.timeout.ms, there's a group rebalance, the consumer asks to join
> > the group, enters it, again never succeeds to send a heartbeat, and then
> > you have you loop again.
> >
> > The same setup, but heartbeats not being overlapped by autocommit works
> > just fine. Why?
> >
> > Can someone clarify this to me?
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message