[
https://issues.apache.org/jira/browse/KAFKA-4685?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15834209#comment-15834209
]
Sinóros-Szabó Péter commented on KAFKA-4685:
--------------------------------------------
Yes. I was able to check ZK only 1 hour after the issue, at about 8:00. At that
time and still now:
{code}
WatchedEvent state:SyncConnected type:None path:null
[zk: localhost:2181(CONNECTED) 0] ls /brokers/ids
[]
[zk: localhost:2181(CONNECTED) 1] ls /controller
[]
[zk: localhost:2181(CONNECTED) 2]
{code}
I have checked the zookeeper logs and those show some error that I do not
understand (around 6:51), so it may happen that this is a ZK issue and Kafka
node managed to register themself successfully, but ZK forgot some if the data.
> All partitions offline, no conroller znode in ZK
> ------------------------------------------------
>
> Key: KAFKA-4685
> URL: https://issues.apache.org/jira/browse/KAFKA-4685
> Project: Kafka
> Issue Type: Bug
> Reporter: Sinóros-Szabó Péter
> Attachments: kafka-0-logs.zip, kafka-1-logs.zip, kafka-2-logs.zip,
> zookeeper-logs.zip
>
>
> Setup: 3 Kafka 0.11.1.1 nodes on kubernetes (in AWS), and another 3 nodes of
> Zookeeper 3.5.2-alpha also in kubernetes (in AWS).
> At 2017-01-23 06:51 ZK sessions expired. It seems from the logs that kafka-2
> was elected as the new controller, but I am not sure how to read that logs.
> I've checked the ZK data and both the /controller is empty and also the
> /brokers/ids is empty. Kafka reports that all partitions are offline,
> although it seems to be working because messages are coming and going.
> We are using an alpha version, I know that it may be a problem, but I suppose
> that Kafka should see that there is not any node registered as controller.
> I have attached the Kafka and ZK logs
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
