Hi Michael, For the consumer side question. Yes, turning off auto offset commit is what you want. But you should also commit offset manually after you have written the processed data to somewhere else. Also, the offset is only associated with a particular consumer group. So if you restart your consumer with a different consumer group, by default it will consume from the log end of the partitions. And this might also the reason why you see the first message ³eaten² by producer - if you start producer before starting consumer and the consumer uses a new group id, it will consume from log end, which might miss the messages that are already produced.
Can you maybe clarify what did you do with the above? Thanks, Jiangjie (Becket) Qin On 7/2/15, 9:47 PM, "Michael J. Kitchin" <mcoyote...@gmail.com> wrote: >Hi there, > >These are questions re: the official camel-kafka integration. Since the >issues touch on both camel and kafka I'm sending to both users lists in >search of answers. > >- - - - > >I have a simple, inonly, point-to-point, synchronous camel route (a) >consuming from kafka topic (consumer), (b) running the resulting exchanges >(messages) through a processor chain and (c) forwarding the outcome on to >somewhere else. > >If the runtime dies while exchanges are in this pipeline, I want things to >pick up where they left off when restarted. I'm not picky about seeing the >same data more than once (as long as it's recent), I just can't lose >anything. > >In brief, everything's working great except this failure/recovery part -- >in-flight exchanges are getting lost and there is no, apparent re-delivery >on restart. My reading of the JMX data suggests the kafka logs are intact. > >I think this has to do with consumer auto-commit, which is the default >behavior. My reading of the kafka and camel-kafka docs suggests disabling >auto-commit will give me what I want, but when I try it I'm not seeing >re-delivery kick off when I restart. > >So, first question: >(1) Is auto-commit off the key to getting what I want and/or what else >might I need to do? > >- - - - - > >Meanwhile, on the producer side I'm seeing the first (and only the first) >message apparently get eaten. It's possible it's being buffered, but it >never seems to timeout. There are no error messages on startup and the >camel context, routes, etc. appear to have started successfully. The >second >message and everything that follows is golden. > >The payloads are ~70-character byte arrays, if it makes a difference. > >Second question, then: >(2) Is there a batching setting or something else I might be overlooking >behind this behavior? > >- - - - - > >Thanks, in advance for your time and consideration. We've been impressed >with kafka so far and are looking forward to employing it in production. > > >Please let me know if I may provide any additional information. Thanks. > > >-Regards, > >MjK > > >- - - - - > > > >*Michael J. Kitchin* > >Senior Software Engineer > >Operational Systems, Inc. > >4450 Arapahoe Avenue, Suite 100 > >Boulder, CO 80303 > > >Phone: 719-271-6476 > >Email: michael.kitc...@opsysinc.com > >Web: www.opsysinc.com
