Page History

...

1. Producer idempotence could be done automatically and much more cheaply by optionally integrating support for this on the server.
2. The existing high-level consumer doesn't expose a lot of the more fine grained control of offsets (e.g. to reset your position). We will be working on that soon

Brokers

Why does controlled shutdown fail?

If a controlled shutdown attempt fails, you will see error messages like the following in your broker logs

WARN [Kafka Server 0], Retrying controlled shutdown after the previous attempt failed... (kafka.server.KafkaServer)

WARN [Kafka Server 0], Proceeding to do an unclean shutdown as all the controlled shutdown attempts failed

In addition to these error messages, if you also see SocketTimeoutExceptions, it indicates that the controller could not finish moving the leaders for all partitions on the broker within controller.socket.timeout.ms. The solution is to increase controller.socket.timeout.ms as well as increase controlled.shutdown.retry.backoff.ms and controlled.shutdown.max.retries to give enough time for the controlled shutdown to complete. If you don't see SocketTimeoutExceptions, it could indicate a problem in your cluster state or a bug as this happens when the controller is not able to move the leaders to another broker for several retries.

Why can't my consumers/producers connect to the brokers?

...