Hello Vadim, Producers maintain a cache about who's the leader for each topic/partition, and when got NotLeader exceptions while producing they will try to refresh their cache from controller. Hence after a while all producers should be producing to the only-up brokers.
In this case even with one broker up you should be able to publish messages as long as your ack <=1 (meaning no need to wait for more than one replica to commit before returning the produce request). Guozhang On Wed, Aug 14, 2013 at 6:10 PM, Vadim Keylis <vkeylis2...@gmail.com> wrote: > We have cluster of 3 Kafka nodes. All topics created with replication > factor of 3. Lets say two crashed for what ever reason. I ran tool prefer > replica election leaders script. it look to me as leaders were moved to the > node that is up. However once trying to publish exception about replicas is > thrown and that is expected. however Should we be still be able to publish > to a single node? How many minimum replicas need it minimum during failure > scenario for Kafka to work. > > > Thanks so much. > > Sent from my iPhone -- -- Guozhang
