Hi Achintya,
Just to clarify, you did not take down either of the zookeepers in this
test, right? Having only two zookeepers in the ensemble would mean that if
either one of them failed, zookeeper wouldn't be able to reach quorum.
I'm not entirely sure why this would happen. One possibility is that the
consumer is failing to find the new coordinator, which might happen if all
the replicas for one of the __consumer_offsets partitions were located in
the "failed" datacenter. Perhaps you can enable DEBUG logging and post some
logs so we can see what it's actually doing during poll().
By the way, I noticed that your consumer configuration settings seem a
little mixed up. The new consumer doesn't actually communicate with
Zookeeper, so there's no need for those settings. And you don't need to
include the "offsets.storage" option since Kafka is the only choice. Also,
I don't think "consumer.timeout.ms" is an option.
-Jason
On Wed, Aug 31, 2016 at 6:43 PM, Ghosh, Achintya (Contractor) <
achintya_gh...@comcast.com> wrote:
> Hi Jason,
>
> Thanks for your response.
>
> I know that is a known issue and I resolved it calling wakeup method by
> another thread. But here my problem is different, let me explain , it's
> very basic
>
> I created one cluster with 6 nodes( 3 from one datacenter and 3 from
> another(remote) datacenter and kept replication factor 6 with 2 zookeeper
> servers one from each datacenter ). Now I brought down all 3 nodes of my
> local datacenter and produced few messages and I see producer is working
> fine even my local data center nodes are down. It successfully writes the
> messages to other data center nodes. But when I'm trying to consume the
> messages the consumer.poll method gets stuck as my local datacenter is down
> though other datacenter's nodes are up.
>
> My question is as the data has been written successfully to other
> datacenter why consumer part is not working?
>
> Here is my Producer settings:
>
> props.put("bootstrap.servers", "psaq1-wc.sys.comcast.net:61616,
> psaq2-wc.sys.comcast.net:61616,psaq3-wc.sys.comcast.net:61616,psaq1-ab.
> sys.comcast.net:61617,psaq2-ab.sys.comcast.net:61617,psaq3
> -ab.sys.comcast.net:61617");
> props.put("acks", "1");
> props.put("max.block.ms", 1000);
> props.put("key.serializer", "org.apache.kafka.common.serialization.
> StringSerializer");
> props.put("value.serializer", "com.comcast.ps.kafka.object.
> CustomMessageSer");
>
> and here is Consumer settings:
>
> props.put("group.id", "app-consumer");
> props.put("enable.auto.commit", "false");
> props.put("auto.offset.reset", "earliest");
> props.put("auto.commit.interval.ms", "500");
> props.put("session.timeout.ms", "120000");
> props.put("consumer.timeout.ms", "10000");
> props.put("zookeeper.session.timeout.ms", "120000");
> props.put("zookeeper.connection.timeout.ms", "60000");
> props.put("offsets.storage","kafka");
> props.put("request.timeout.ms", "150000");
> props.put("bootstrap.servers", "psaq1-wc.sys.comcast.net:
> 61616,psaq2-wc.sys.comcast.net:61616,psaq3-wc.sys.comcast.net:61616,
> psaq1-ab.sys.comcast.net:61617,psaq2-ab.sys.comcast.net:61617,psaq3
> -ab.sys.comcast.net:61617");
> props.put("key.deserializer", "org.apache.kafka.common.
> serialization.StringDeserializer");
> props.put("value.deserializer",
> "com.comcast.ps.kafka.object.CustomMessageDeSer");
>
> Is it because of consumer is not able to get the broker metadata if it is
> trying to connect other datacenter's zookeeper server? I tried with to
> increate the zookeeper session timeout and connection time out but no luck.
>
> Please help on this.
> Thanks
> Achintya
>
>
> -----Original Message-----
> From: Jason Gustafson [mailto:ja...@confluent.io]
> Sent: Wednesday, August 31, 2016 4:05 PM
> To: us...@kafka.apache.org
> Cc: dev@kafka.apache.org
> Subject: Re: Kafka consumers unable to process message
>
> Hi Achintya,
>
> We have a JIRA for this problem: https://issues.
> apache.org/jira/browse/KAFKA-3834. Do you expect the client to raise an
> exception in this case or do you just want to keep it from blocking
> indefinitely? If the latter, you could escape the poll from another thread
> using wakeup().
>
> Thanks,
> Jason
>
> On Wed, Aug 31, 2016 at 12:11 PM, Ghosh, Achintya (Contractor) <
> achintya_gh...@comcast.com> wrote:
>
> > Hi there,
> >
> > Kafka consumer gets stuck at consumer.poll() method if my current
> > datacenter is down and replicated messages are in remote datacenter.
> >
> > How to solve that issue?
> >
> > Thanks
> > Achintya
> >
>
