Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
Kafka >> mail # user >> Getting LeaderNotAvailableException in console producer after increasing partitions from 4 to 16.


Copy link to this message
-
Re: Getting LeaderNotAvailableException in console producer after increasing partitions from 4 to 16.
Cool! You can follow the process of creating a JIRA here:

http://kafka.apache.org/contributing.html

And submit patch here:

https://cwiki.apache.org/confluence/display/KAFKA/Git+Workflow

It will be great if you can also add an entry for this issue in FAQ since I
think this is a common question:

https://cwiki.apache.org/confluence/display/KAFKA/FAQ

Guozhang
On Tue, Aug 27, 2013 at 2:38 PM, Rajasekar Elango <[EMAIL PROTECTED]>wrote:

> Thanks Guozhang, Changing max retry to 5 worked. Since I am changing
> console producer code, I can also submit patch adding both
> message.send.max.retries
> and retry.backoff.ms to console producer. Can you let me know process for
> submitting patch?
>
> Thanks,
> Raja.
>
>
> On Tue, Aug 27, 2013 at 4:03 PM, Guozhang Wang <[EMAIL PROTECTED]> wrote:
>
> > Hello Rajasekar,
> >
> > The remove fetcher log entry is normal under addition of partitions,
> since
> > they indicate that some leader changes have happened so brokers are
> closing
> > the fetchers to the old leaders.
> >
> > I just realized that the console Producer does not have the
> > message.send.max.retries options yet. Could you file a JIRA for this and
> I
> > will followup to add this option? As for now you can hard modify the
> > default value from 3 to a larger number.
> >
> > Guozhang
> >
> >
> > On Tue, Aug 27, 2013 at 12:37 PM, Rajasekar Elango
> > <[EMAIL PROTECTED]>wrote:
> >
> > > Thanks Neha & Guozhang,
> > >
> > > When I ran StateChangeLogMerger, I am seeing this message repeated 16
> > times
> > > for each partition:
> > >
> > > [2013-08-27 12:30:02,535] INFO [ReplicaFetcherManager on broker 1]
> > Removing
> > > fetcher for partition [test-60,13] (kafka.server.ReplicaFetcherManager)
> > > [2013-08-27 12:30:02,536] INFO [Log Manager on Broker 1] Created log
> for
> > > partition [test-60,13] in
> > > /home/relango/dev/mandm/kafka/main/target/dist/mandm-kafka/kafka-data.
> > > (kafka.log.LogManager)
> > >
> > > I am also seeing .log and .index files created for this topic in data
> > dir.
> > > Also list topic command shows leaders, replicas and isrs for all
> > > partitions. Do you still think increasing num of retries would help or
> is
> > > it some other issue..? Also console Producer doesn't seem to  have
> option
> > > to set num of retries. Is there a way to configure num of retries for
> > > console producer ?
> > >
> > > Thanks,
> > > Raja.
> > >
> > >
> > > On Tue, Aug 27, 2013 at 12:52 PM, Neha Narkhede <
> [EMAIL PROTECTED]
> > > >wrote:
> > >
> > > > As Guozhang said, your producer might give up sooner than the leader
> > > > election completes for the new topic. To confirm if your producer
> gave
> > up
> > > > too soon, you can run the state change log merge tool for this topic
> > and
> > > > see when the leader election finished for all partitions
> > > >
> > > > ./bin/kafka-run-class.sh kafka.tools.StateChangeLogMerger --logs
> > > <location
> > > > to all state change logs> --topic <topic>
> > > >
> > > > Note that this tool requires you to give the state change logs for
> all
> > > > brokers in the cluster.
> > > >
> > > >
> > > > Thanks,
> > > > Neha
> > > >
> > > >
> > > > On Tue, Aug 27, 2013 at 9:45 AM, Guozhang Wang <[EMAIL PROTECTED]>
> > > wrote:
> > > >
> > > > > Hello Rajasekar,
> > > > >
> > > > > In 0.8 producers keep a cache of the partition -> leader_broker_id
> > map
> > > > > which is used to determine to which brokers should the messages be
> > > sent.
> > > > > After new partitions are added, the cache on the producer has not
> > > > populated
> > > > > yet hence it will throw this exception. The producer will then try
> to
> > > > > refresh its cache by asking the brokers "who are the leaders of
> these
> > > new
> > > > > partitions that I do not know of before". The brokers at the
> > beginning
> > > > also
> > > > > do not know this information, and will only get this information
> from
> > > > > controller which will only propagation the leader information after