[jira] [Created] (FLINK-27481) Flink checkpoints are very slow after upgrading from Flink 1.13.1 to Flink 1.14.3

["Monika Hristova (Jira)" <ji...@apache.org>]

Monika Hristova created FLINK-27481:
---------------------------------------

             Summary: Flink checkpoints are very slow after upgrading from Flink 1.13.1 to Flink 1.14.3
                 Key: FLINK-27481
                 URL: https://issues.apache.org/jira/browse/FLINK-27481
             Project: Flink
          Issue Type: Bug
          Components: Runtime / Checkpointing
    Affects Versions: 1.14.4
            Reporter: Monika Hristova


I am experiencing extremely slow checkpoints after upgrading flink version from 1.13.1 to 1.14.3. The checkpointing is slow because of the sink/source. Example [!https://i.stack.imgur.com/oChk9.png!|https://i.stack.imgur.com/oChk9.png]

This source is a kafka topic that have some messages inside and still, it might take up to minutes to be checkpointed.

I am using incremental checkpoints, unaligned checkpoints are not enabled. Configurations are the default ones, and this source was checkpointed for milliseconds with flink 1.13.1 and now it times out(the timeout is 3 minutes) for every second chechpoint.

[!https://i.stack.imgur.com/kTbSx.png!|https://i.stack.imgur.com/kTbSx.png]

[FlinkKafkaConsumer|https://nightlies.apache.org/flink/flink-docs-release-1.14/api/java/org/apache/flink/streaming/connectors/kafka/FlinkKafkaConsumer.html]/[Producer|https://nightlies.apache.org/flink/flink-docs-release-1.14/api/java/org/apache/flink/streaming/connectors/kafka/FlinkKafkaProducer.html] are used which are deprecated.

 

After removing the topic with the few configurational messages the checkpoint time improved drastically but it still takes around a minute for the checkpoint due to alignment duration in the kafka sinks. 



--
This message was sent by Atlassian Jira
(v8.20.7#820007)