LoveHeat commented on code in PR #22806:
URL: https://github.com/apache/flink/pull/22806#discussion_r1244957853
##########
flink-runtime/src/main/java/org/apache/flink/runtime/source/coordinator/SourceCoordinator.java:
##########
@@ -195,9 +195,19 @@ void announceCombinedWatermark() {
"Distributing maxAllowedWatermark={} to subTaskIds={}",
maxAllowedWatermark,
subTaskIds);
- for (Integer subtaskId : subTaskIds) {
- context.sendEventToSourceOperator(
- subtaskId, new
WatermarkAlignmentEvent(maxAllowedWatermark));
+ // Because of Java-ThreadPoolExecutor will not schedule the period task
+ // if it throws an exception, so we should handle the potential
exception like
+ // "subtask xx is not ready yet to receive events" to increase
robustness.
+ try {
+ for (Integer subtaskId : subTaskIds) {
+ context.sendEventToSourceOperator(
+ subtaskId, new
WatermarkAlignmentEvent(maxAllowedWatermark));
+ }
Review Comment:
After review code, partially-send-events is just ok, no negative impact(or i
haven't found).
There are two cases that gateway is not ready:
1. task during deploying
2. task during restore before re-deployed (see code:
[schedulerBase::notifyCoordinatorsOfSubtaskRestore](https://github.com/apache/flink/blob/master/flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SchedulerBase.java#L408)
)
for case1: it only limit the ready task consume speed to avoid it consume
too faster than unready task, i think this is acceptable.
for case2: partially sending events have no impact on case2, because all
task will be re-deployed.
@mas-chen
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
