[
https://issues.apache.org/jira/browse/FLINK-32362?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17733363#comment-17733363
]
Rui Fan commented on FLINK-32362:
---------------------------------
Thanks for your quick feedback.
I'm not sure what should we do when some subtasks are not ready.
* Option1: just send event to all ready subtasks.
* Option2: Don't send any event before all subtasks are ready.
If we expect option1, your solution might work.
And your solution1 has a little bug, when one subtask isn't ready, we should
continue send event to next subtask. So the `try catch` should be moved to
inside of the `for loop`.
{code:java}
for (Integer subtaskId : subTaskIds) {
try {
context.sendEventToSourceOperator(
subtaskId, new WatermarkAlignmentEvent(maxAllowedWatermark));
} catch (Throwable ignore) {
LOG.warn("Announce the newest combined watermark to source failed, task
maybe during failover, wait next time to announce."
}
} {code}
> SourceAlignment announceCombinedWatermark period task maybe lost
> ----------------------------------------------------------------
>
> Key: FLINK-32362
> URL: https://issues.apache.org/jira/browse/FLINK-32362
> Project: Flink
> Issue Type: Bug
> Affects Versions: 1.16.0
> Reporter: Cai Liuyang
> Assignee: Cai Liuyang
> Priority: Major
>
> When we use sourcealignment，we also found there is another problem that
> announceCombinedWatermark may throw a exception (like "subtask 25 is not
> ready yet to receive events" , this subtask maybe under failover), which will
> lead the period task not running any more (ThreadPoolExecutor will not
> schedule the period task if it throw a exception)
> I think we should increase the robustness of announceCombinedWatermark
> function to avoid it throw any exception (if send fail, just wait next send)
> (code see
> [https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/source/coordinator/SourceCoordinator.java#L199]
> )
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
