AIOKafkaConsumer recovery from whole-cluster-dead

[YarinLowe]

Hey, I ran into the following bug (or unexpected behavior) and would be glad to get your help with that, or any idea to what I'm doing wrong. This bug is reproducing easily on my environment.

Thanks!

Describe the bug

1. Cluster of 3 brokers is up and running.
2. I run AIOKafkaConsumer with bootstrap_servers list which contains all 3 brokers.
3. I produce a message and it is consumed successfully by my AIOKafkaConsumer.
4. I kill all 3 brokers (as part of a test).
5. AIOKafkaConsumer repeatedly prints "NodeNotReadyError" about all 3 brokers (which is fine)
6. I start all 3 brokers
7. AIOKafkaConsumer stops printing "NodeNotReadyError" (which also looks fine)
8. I produce a message and AIOKafkaConsumer does not consume it, neither printing any error.

Expected behaviour Step 7 should successfully consume the message

Environment:

• aiokafka version: 0.7.0
• kafka-python version: 1.3.5
• Kafka Broker version: 2.13-2.7.0