Thanks Jun for your response.
Here is my scenario:

topicCountMap.put(topic, new Integer(2));
Map<String, List<KafkaStream<byte[], byte[]>>> consumerMap =
consumer.createMessageStreams(topicCountMap);
List<KafkaStream<byte[], byte[]>> streams = consumerMap.get(topic);

So from above scenario (only 1 partition) , there will be 2 threads C1 and
C2, and one of the thread would be holding one partition. In my scenario,
C1 would be hung after consuming the message from stream and in hung state
during processing of message. So in such scenario, looks like C2 will not
get hold of partition stream unless C1 didn't die..

So considering such scenarios, I am planning to use SimpleConsumer where I
will have more control on consuming  Partition content. But I have to
maintain offset state in zookeeper. Let me know if this approach is correct
for such hung scenarios.

Thanks and regards,
Srini


On Sun, Nov 9, 2014 at 9:12 PM, Jun Rao <jun...@gmail.com> wrote:

> If C1 dies, C2 will be owning that partition. However, C1 has to really
> die, which typically means that either you close the consumer connector or
> the jvm of C1 is gone.
>
> In your case, it seems that C1 didn't die, it just hung. Do you know why C1
> hung?
>
> Thanks,
>
> Jun
>
> On Fri, Nov 7, 2014 at 3:34 PM, Srinivas Reddy Kancharla <
> getre...@gmail.com
> > wrote:
>
> > Hi,
> >
> > I have a scenario where I have 1 partition and 1 consumer group having 2
> > consumer threads running say C1 and C2. Since there is only one partition
> > for a given topic, say C1 is holding that partition. Now due to some
> reason
> > if C1 dies, can C2 get hold of that partition?
> >
> > i.e. C1 was busy with KafkaStream instance, for any reason if C1 dies or
> in
> > hung state, Can we make C2 talking to KafkaStream (for Partition 0).
> > I am facing this issue where I have 10 messages in partition 0 and C1 was
> > consuming it. At message 4, C1 went into hung state. Now I would like to
> > make C2 to consumer other messages which are not consumed by C1.
> >
> > Thank and regards,
> > Srini
> >
>

Reply via email to