[jira] [Created] (SAMZA-2785) Reuse of KafkaCheckpointManager causes memory leak

["Katie Liu (Jira)" <ji...@apache.org>]

Katie Liu created SAMZA-2785:
--------------------------------

             Summary: Reuse of KafkaCheckpointManager causes memory leak
                 Key: SAMZA-2785
                 URL: https://issues.apache.org/jira/browse/SAMZA-2785
             Project: Samza
          Issue Type: Bug
            Reporter: Katie Liu


Calling start() and stop() multiple times on the same KafkaCheckpointManager, while stopConsumerAfterFirstRead == true, causes the SystemConsumer left unclosed. The unclosed SystemConsumer can cause memory leaks in some implementations.

 

Evidence:

In production logs, SystemConsumer was started 1741 times, but only closed 14 times.
{code:java}
[katxxxx@xxxxxx xxxx]$ grep "Starting the checkpoint SystemConsumer from oldest offset" xxxxxx.log|wc -l
1741
[katxxxxx@xxxxxx xxxx]$ grep "Stopping system consumer" xxxxxxxxxx.log |wc -l
14 {code}
We also have a heap dump of KafkaSystemConsumer taking up 8Gbs of memory.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)