THIS IS A TEST INSTANCE. ALL YOUR CHANGES WILL BE LOST!!!!
...
- on startup the consumer issues a list_groups(my_group) to a random broker to find out the location of the controller for its group (each server is the controller for some groups)
- knowing where the controller is, it connects and issues a RegisterConsumer RPC, then it awaits a RegisterConsumerResponse
- on receiving the RegisterConsumer request the server sends an error code in the HeartbeatResponse to all alive group members and waits for a new RegisterConsumer request from each consumer, except the one that sent the initial RegisterConsumer.
- on receiving an error code in the HeartbeatResponse, all consumers stop fetching, commit offsets, re-discover the controller through list_groups(my_group) and send a RegisterConsumer request to the controller
- on receiving RegisterConsumer from all consumers of the group membership change, the controller sends the RegisterConsumerResponse to all the new group members, with the partitions and the respective offsets to restart consumption from. If there are no previous offsets for the consumer group, -1 is returned. The consumer starts fetching from earliest or latest, depending on consumer configuration
- on receiving the RegisterConsumerResponse the consumer is now able to start consuming its partitions and must now start sending heartbeat messages back to the controller with the current generation id.
...