Hi, I have been reading about the Kafka design and I'm really impressed. I really like how you use OS page caching and message offsets as ids to achieve simplicity. I also like that you use disks very well and hence are able to support the pull model, which simplifies things so much. I have a couple questions.
If a Broker machine goes down, is restarted and hostname of the machine changes, in this case would the logs on the machine become available or would they be lost. I would really appreciate if you could provide some details on how this would work. I see that a consumer going down triggers a rebalance and the paper mentions that the rebalance happens pretty quickly. If we have a fickle consumer that for some reason starts up and crashes, again and again, would that block the entire Kafka system? Thanks in Advance, Pankaj
