Hi,

I'm experimenting with the following scenario:

- 3 brokers are running (0,1 and 2) -- Kafka version 0.8.2.0
- Continuously: restart broker number 0 by triggering controlled shutdown.
Sleep rand(10) seconds. repeat.
- Continuously: create 'simple-test-topic' (RF=2), write and read messages,
then delete the topic. repeat.

After a while, broker 0 doesn't come back up any more due to "corrupt
index" error (but that's not my question for the moment). Looking at the
state of the topics:

Topic:simple-test-topic    PartitionCount:8    ReplicationFactor:2
 Configs:
    Topic: simple-test-topic    Partition: 0    Leader: -1    Replicas: 1,2
   Isr: 1
    Topic: simple-test-topic    Partition: 1    Leader: -1    Replicas: 2,0
   Isr: 2
    Topic: simple-test-topic    Partition: 2    Leader: -1    Replicas: 0,1
   Isr: 1
    Topic: simple-test-topic    Partition: 3    Leader: -1    Replicas: 1,0
   Isr: 1
    Topic: simple-test-topic    Partition: 4    Leader: -1    Replicas: 2,1
   Isr: 1
    Topic: simple-test-topic    Partition: 5    Leader: -1    Replicas: 0,2
   Isr: 2
    Topic: simple-test-topic    Partition: 6    Leader: -1    Replicas: 1,2
   Isr: 1
    Topic: simple-test-topic    Partition: 7    Leader: -1    Replicas: 2,0
   Isr: 2
Topic:test    PartitionCount:8    ReplicationFactor:3    Configs:
    Topic: test    Partition: 0    Leader: 1    Replicas: 1,2,0    Isr: 2,1
    Topic: test    Partition: 1    Leader: 2    Replicas: 2,0,1    Isr: 2,1
    Topic: test    Partition: 2    Leader: 2    Replicas: 0,1,2    Isr: 2,1
    Topic: test    Partition: 3    Leader: 1    Replicas: 1,0,2    Isr: 2,1
    Topic: test    Partition: 4    Leader: 2    Replicas: 2,1,0    Isr: 2,1
    Topic: test    Partition: 5    Leader: 2    Replicas: 0,2,1    Isr: 2,1
    Topic: test    Partition: 6    Leader: 1    Replicas: 1,2,0    Isr: 2,1
    Topic: test    Partition: 7    Leader: 2    Replicas: 2,0,1    Isr: 2,1

.. at which point:

- All 'simple-test-topic' partitions are leaderless.
- It's not possible to delete "simple-test-topic" any more.
- Calling 'kafka-preferred-replica-election.sh' successfully starts
election but doesn't have any effect.

The other topic, named "test" (RF 3), is just sitting there and not
actively participating in the test.

Now I'm wondering:

- Which of the steps above could have caused "simple-test-topic" partitions
to become leaderless?
- How to recover in such situation in cases where broker 0 can or cannot be
recovered?

Thanks,
Omid

Reply via email to