zhuzhurk commented on a change in pull request #10067: [FLINK-14375][runtime]
Avoid to notify scheduler about fake or outdated state update
URL: https://github.com/apache/flink/pull/10067#discussion_r342084999
##########
File path:
flink-runtime/src/main/java/org/apache/flink/runtime/scheduler/SchedulerBase.java
##########
@@ -417,12 +417,55 @@ public void cancel() {
@Override
public final boolean updateTaskExecutionState(final TaskExecutionState
taskExecutionState) {
final Optional<ExecutionVertexID> executionVertexId =
getExecutionVertexId(taskExecutionState.getID());
- if (executionVertexId.isPresent()) {
- executionGraph.updateState(taskExecutionState);
-
updateTaskExecutionStateInternal(executionVertexId.get(), taskExecutionState);
+
+ boolean updateSuccess =
executionGraph.updateState(taskExecutionState);
+
+ if (updateSuccess && executionVertexId.isPresent()) {
+ Optional<TaskExecutionState> validTaskExecutionState =
getValidTaskExecutionState(
+ executionVertexId.get(),
+ taskExecutionState);
+
+ if (validTaskExecutionState.isPresent()) {
+
updateTaskExecutionStateInternal(executionVertexId.get(),
validTaskExecutionState.get());
+ }
return true;
+ } else {
+ return false;
+ }
+ }
+
+ private Optional<TaskExecutionState> getValidTaskExecutionState(
+ final ExecutionVertexID executionVertexId,
+ final TaskExecutionState taskExecutionState) {
+
+ final ExecutionVertex executionVertex =
getExecutionVertex(executionVertexId);
+ final ExecutionAttemptID currentExecutionAttemptID =
executionVertex.getCurrentExecutionAttempt().getAttemptId();
+
+ // check whether this state update is outdated
+ if
(!currentExecutionAttemptID.equals(taskExecutionState.getID())) {
+ return Optional.empty();
}
- return false;
+
+ // only notifies FINISHED and FAILED states which are needed at
the moment.
+ // can be refined in FLINK-14233 after the legacy scheduler is
removed and
+ // the actions are factored out from ExecutionGraph.
+ switch (taskExecutionState.getExecutionState()) {
Review comment:
I think it would be simpler to only consider these two required state
changes.
Notifying other states might be no hard but I have not thought it
thoroughly. Moreover, it's hard to notify all states cleanly at the moment
before we factored all actions out from ExecutionGraph. So I'd prefer to do it
in a follow up ticket in Flink 1.11, i.e. FLINK-14233
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
