[KAFKA-9917] Consumer not assigned any partitions in group after broker restart

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.1.1
• Component/s: consumer
• Environment: Image solsson/kafka:2.1.1@sha256:8bc8242c649c395ab79d76cc83b1052e63b4efea7f83547bf11eb3ef5ea6f8e1 deployed in Kubernetes v 1.12.7 (minikube v1.4.0) ; 1 broker
• Assignee: Unassigned
• Reporter: Igor Baltiyskiy
• Created: 24/Apr/20 21:30
• Updated: 29/Apr/20 10:27

Description

I've only reproduced it once for now, so I don't have a ready reproduction sequence.

I was trying to reproduce KAFKA-9912, as advised, and kept restarting Kafka (via kubectl scale --replicas 0 / 1). At some point my application entered the state where there is an active KafkaConsumer in a group ID, but it is not assigned any partitions, although the broker is functional.

 bin/kafka-consumer-groups.sh --bootstrap-server kafka:9092 --describe --group tfp-9
Consumer group 'tfp-9' has no active members.TOPIC              PARTITION  CURRENT-OFFSET  LOG-END-OFFSET  LAG             CONSUMER-ID     HOST            CLIENT-ID
increasing-topic-0 0          270574          500000          229426          -               -               - 

bin/kafka-console-consumer.sh --bootstrap-server kafka:9092 --topic increasing-topic-0  --partition 0 --offset 270574 --max-messages 5
270575
270576
270577
270578
270579
Processed a total of 5 messages

The app is active, I can see that KafkaConsumer is actively polling the broker. App logs are full of "Resuming partitions []" and "Pausing partitions []" messages.

I'm attaching broker and application logs (without "Resuming|Pausing partitions []" messages, also truncated from the beginning so that they start just before the broker is started).