[
https://issues.apache.org/jira/browse/FLINK-4702?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15529580#comment-15529580
]
ASF GitHub Bot commented on FLINK-4702:
---------------------------------------
Github user tzulitai commented on the issue:
https://github.com/apache/flink/pull/2559
@StephanEwen
On a second look, I think the `commitSpecificOffsetsToKafka` method was
designed to commit synchronously in the first place. `AbstractFetcher` holds a
Map of all current pending offsets for committing by checkpointID, and on every
`notifyCheckpointComplete` the offsets are removed from the Map before
`commitSpecificOffsetsToKafka` is called.
So, for async committing, I think we need to remove cleaning up the offsets
in `AbstractFetcher#notifyCheckpointComplete()` and instead clean them up in a
new separate callback handle method in `AbstractFetcher`.
> Kafka consumer must commit offsets asynchronously
> -------------------------------------------------
>
> Key: FLINK-4702
> URL: https://issues.apache.org/jira/browse/FLINK-4702
> Project: Flink
> Issue Type: Bug
> Components: Kafka Connector
> Affects Versions: 1.1.2
> Reporter: Stephan Ewen
> Assignee: Stephan Ewen
> Priority: Blocker
> Fix For: 1.2.0, 1.1.3
>
>
> The offset commit calls to Kafka may occasionally take very long.
> In that case, the {{notifyCheckpointComplete()}} method blocks for long and
> the KafkaConsumer cannot make progress and cannot perform checkpoints.
> Kafka 0.9+ have methods to commit asynchronously.
> We should use those and make sure no more than one commit is concurrently in
> progress, to that commit requests do not pile up.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
