Sorry for not being clear. Kafka ships with a tool, that allows to get a
dump of topic content in human readable form (ie, what is store on disk
in a topic).

You can call it via `bin/kafka-run-class.sh kafka.tools.DumpLogSegments`
and need to point it to the corresponding segment files. For example:

/tmp/kafka-logs/__consumer_offsets-1/00000000000000000000.log

with /tmp/kafka-logs being whatever you configured for `log.dir` for
your broker. Than it's topic-name and segment files are numbered and end
with .log



-Matthias


On 2/16/18 12:57 PM, Russell Teabeault wrote:
> Matthias,
> 
> I will try to get you the broker logs. As far as the partitions of
> __consumer_offsets topic, I am not quite clear on what you want me to
> provide. Are you wanting the output of
> 
> ./kafka-topics.sh --zookeeper zk-host --describe --topic __consumer_offsets
> 
> -russ
> 
> On Fri, Feb 16, 2018 at 1:45 PM, Matthias J. Sax <[email protected]>
> wrote:
> 
>> Ok. Let us know if bouncing works.
>>
>> Sounds like a bug thought, that the consumer group is not cleared. It
>> would be helpful to get the broker logs (of the node that hosts the
>> coordinator). Can you also provide a dump of the corresponding
>> partitions of __consumer_offsets topic?
>>
>> Thanks a lot.
>>
>> -Matthias
>>
>>
>> On 2/16/18 12:35 PM, Russell Teabeault wrote:
>>> Matthias,
>>>
>>> The session.timeout.ms is set to 10000. It has been in this "weird"
>> state
>>> now for going on 24 hours.
>>> We are going to try to bounce the group coordinator.
>>>
>>> Thanks,
>>> -russ
>>>
>>> On Fri, Feb 16, 2018 at 12:18 PM, Matthias J. Sax <[email protected]
>>>
>>> wrote:
>>>
>>>> Hi,
>>>>
>>>> this "weird" state can happen. Usually, a consumer sends a "leave group
>>>> request" to inform the broker it's shutting down. However, this is a
>>>> send-and-forget approach and if the message is lost, the broker might
>>>> think that a stopped consumer is still alive. For this reason, the
>>>> broker should eventually time out consumers and thus clean-up the
>>>> consumer group.
>>>>
>>>> How long did you wait?
>>>>
>>>> Did you modify the `session.timeout.ms` config?
>>>>
>>>> As a last resort, bouncing the broker that hosts the consumer group
>>>> coordinator should resolve the issue.
>>>>
>>>> -Matthias
>>>>
>>>> On 2/15/18 10:09 PM, Russell Teabeault wrote:
>>>>> It appears that kafka and/or kafka streams has gotten itself into a
>> weird
>>>>> state. My KS app is not running. I try to reset it using the reset
>>>> command
>>>>> line tool and I get:
>>>>>
>>>>> ERROR: java.lang.IllegalStateException: Consumer group
>>>>> 'dp-consumption-aggregation-processor-staging2' is still active. Make
>>>> sure
>>>>> to stop all running application instances before running the reset
>> tool.
>>>>> java.lang.IllegalStateException: Consumer group
>>>>> 'dp-consumption-aggregation-processor-staging2' is still active. Make
>>>> sure
>>>>> to stop all running application instances before running the reset
>> tool.
>>>>>
>>>>> But as I said the application is not running. So I ran
>>>>>
>>>>> ./kafka-consumer-groups.sh --bootstrap-server
>>>>> smf1-abi-09-sr1.prod.twitter.com:9092 --group
>>>>> dp-consumption-aggregation-processor-staging2 --describe
>>>>>
>>>>> and the first line is
>>>>>
>>>>> Warning: Consumer group 'dp-consumption-aggregation-
>> processor-staging2'
>>>> is
>>>>> rebalancing.
>>>>>
>>>>> How do I get it out of this state and how did I get it into this state?
>>>>>
>>>>> Thanks!
>>>>>
>>>>
>>>>
>>>
>>>
>>
>>
> 
> 

Attachment: signature.asc
Description: OpenPGP digital signature

Reply via email to