Is the zookeeper process you're referring to the server or the client (broker
in the case of Kafka)?
If you're referring to the server, then the ensemble recovers from the disk
state of the servers. If it is a follower, then there is nothing to recover,
the leader already has all the necessary data. If it is the leader, then it
will get the ephemeral/session data from disk.
If you're referring to the Kafka broker, then you should be seeing a message
like this in your log in the case the broker finds a znode upon registration:
case e: ZkNodeExistsException =>
throw new RuntimeException("A broker is already registered on the path " +
brokerIdPath
+ ". This probably " + "indicates that you either have configured a
brokerid that is already in use, or "
+ "else you have shutdown this broker and restarted it faster than
the zookeeper "
+ "timeout so it appears to be re-registering.")
Does it help?
-Flavio
> On 30 Aug 2016, at 14:35, J316 Services <j316servi...@icloud.com> wrote:
>
> Setup is 3 participants + 1 observer.
>
> We used 3.5.0 for dynamic configs.
>
> What happens when a system forced killed the zookeeper process and it did not
> have time to clean up ephemeral? Does it recover in that scenario?
>
>
> Thanks.
>
> Sent from my iPad
>
>> On Aug 30, 2016, at 8:53 AM, Flavio Junqueira <f...@apache.org> wrote:
>>
>> I think you're saying that the session isn't expiring that and the ephemeral
>> isn't getting deleted. Or maybe the session is expiring but the ephemeral
>> isn't being deleted? It'd be great if you could check that the broker
>> session is expiring eventually.
>>
>> Since you're on the 3.5 branch, you may want to use 3.5.2-alpha or simply
>> try 3.4.8 as Harsha suggested.
>>
>> -Flavio
>>
>>> On 29 Aug 2016, at 18:44, Harsha Chintalapani <ka...@harsha.io> wrote:
>>>
>>> how many brokers you've in this cluster. Do you try using a stable
>>> zookeeper release like 3.4.8?
>>> -Harhsa
>>>
>>>> On Mon, Aug 29, 2016 at 5:21 AM Nomar Morado <nomar.mor...@gmail.com>
>>>> wrote:
>>>>
>>>> we are using kafka 0.9.0.1 and zk 3.5.0-alpha
>>>>
>>>> On Mon, Aug 29, 2016 at 8:12 AM, Nomar Morado <nomar.mor...@gmail.com>
>>>> wrote:
>>>>
>>>>> we would get this occasionally after a weekend reboot/restart.
>>>>>
>>>>> we tried restarting a couple of times all to naught.
>>>>>
>>>>> we had to delete dk's directory to get his going again.
>>>>>
>>>>> any ideas what might cause this issue and suggestions on how to resolve
>>>>> this?
>>>>>
>>>>>
>>>>> thanks.
>>>>
>>>>
>>>>
>>>> --
>>>> Regards,
>>>> Nomar Morado
>>
