pnowojski commented on a change in pull request #10151: [FLINK-14231] Handle
the pending processing-time timers to make endInput semantics on the operator
chain strict
URL: https://github.com/apache/flink/pull/10151#discussion_r368994586
##########
File path:
flink-streaming-java/src/main/java/org/apache/flink/streaming/runtime/tasks/StreamTask.java
##########
@@ -1046,10 +1046,16 @@ TimerService getTimerService() {
return timerService;
}
- public ProcessingTimeService getProcessingTimeService(int
operatorIndex) {
- Preconditions.checkState(timerService != null, "The timer
service has not been initialized.");
- MailboxExecutor mailboxExecutor =
mailboxProcessor.getMailboxExecutor(operatorIndex);
- return new ProcessingTimeServiceImpl(timerService, callback ->
deferCallbackToMailbox(mailboxExecutor, callback));
Review comment:
Why was this removed? Yes, the method is ugly and the solution would be to
properly use `StreamOperatorFactory` concept, but the deletion of this method
is having a very nasty effect, where `ProcessingTimeServiceAware` can be
applied to a factory AND to an operator, which creates a confusion by allowing
for a bit pointless patterns, where operator has a factory and both the
operator and it's factory are `ProcessingTimeServiceAware`.
Unless I'm missing something, I think it would be better for the API to keep
this method, mark it `@Deprecated`, and never allow for
`ProcessingTimeServiceAware` to be applied on an operator.
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
