[
https://issues.apache.org/jira/browse/KAFKA-2572?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
John Firth updated KAFKA-2572:
------------------------------
Description:
On two occasions, have seen zk session expiry, followed by a timeout during a
consumer rebalance following this expiry, followed by multiple successive zk
session expiries. Restarting the process using the zk client resolved the
problems.
Comparing these with a case in which a new stable zk session was created
following a session expiry, the timeout during rebalance is not seen in the
successful case.
This behavior was seen on 09/08 and 09/11 -- the attached 'full' logs show all
logs entries minus entries particular to our application. For 09/08, the time
span is 2015-09-08T12:52:06.069-04:00 to 2015-09-08T13:14:48.250-04:00; for
11/08, the time span is between 2015-09-11T01:38:17.000-04:00 to
2015-09-11T07:44:47.124-04:00. The digest logs are the result of retaining only
error and warning entries, and entries containing any of: "begin rebalancing",
"end rebalancing", "timed", and "zookeeper state". For the 09/11 digest logs,
entries from the kafka.network.Processor logger are also excised for clarity.
Unfortunately, debug logging was not enabled during these events.
was:
On two occasions, have seen zk session expiry, followed by a timeout during a
consumer rebalance following this expiry, followed by multiple successive zk
session expiries. Restarting the process using the zk client resolved the
problems.
Comparing these with a case in which a new stable zk session was created
following a session expiry, the timeout during rebalance is not seen in the
successful case.
This behavior was seen on 09/08 and 09/11 -- the attached 'full' logs show all
logs entries minus entries particular to our application. For 09/08, the time
span is 2015-09-08T12:52:06.069-04:00 to 2015-09-08T13:14:48.250-04:00; for
11/08, the time span is between 2015-09-11T01:38:17.000-04:00 to
2015-09-11T07:44:47.124-04:00. The digest logs are the result of retaining only
error and warning entries, and entries containing any of: "begin rebalancing",
"end rebalancing", "timed", and "zookeeper state". For the 09/11 digest logs,
entries from the kafka.network.Processor logger are also excised for clarity.
> zk connection instability, perhaps precipitated by zk client timeout during
> rebalance
> -------------------------------------------------------------------------------------
>
> Key: KAFKA-2572
> URL: https://issues.apache.org/jira/browse/KAFKA-2572
> Project: Kafka
> Issue Type: Bug
> Components: zkclient
> Affects Versions: 0.8.2.1
> Environment: zk version 3.4.6,
> CentOS 6, 2.6.32-504.1.3.el6.x86_64
> Reporter: John Firth
> Attachments: 090815-digest.log, 090815-full.log, 091115-digest.log,
> 091115-full.log.zip
>
>
> On two occasions, have seen zk session expiry, followed by a timeout during a
> consumer rebalance following this expiry, followed by multiple successive zk
> session expiries. Restarting the process using the zk client resolved the
> problems.
> Comparing these with a case in which a new stable zk session was created
> following a session expiry, the timeout during rebalance is not seen in the
> successful case.
> This behavior was seen on 09/08 and 09/11 -- the attached 'full' logs show
> all logs entries minus entries particular to our application. For 09/08, the
> time span is 2015-09-08T12:52:06.069-04:00 to 2015-09-08T13:14:48.250-04:00;
> for 11/08, the time span is between 2015-09-11T01:38:17.000-04:00 to
> 2015-09-11T07:44:47.124-04:00. The digest logs are the result of retaining
> only error and warning entries, and entries containing any of: "begin
> rebalancing", "end rebalancing", "timed", and "zookeeper state". For the
> 09/11 digest logs, entries from the kafka.network.Processor logger are also
> excised for clarity. Unfortunately, debug logging was not enabled during
> these events.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
