zhuzhurk commented on code in PR #22506:
URL: https://github.com/apache/flink/pull/22506#discussion_r1189766188
##########
flink-runtime/src/main/java/org/apache/flink/runtime/jobmaster/JobMaster.java:
##########
@@ -473,26 +500,50 @@ public CompletableFuture<Acknowledge> cancel(Time
timeout) {
@Override
public CompletableFuture<Acknowledge> updateTaskExecutionState(
final TaskExecutionState taskExecutionState) {
- FlinkException taskExecutionException;
+ checkNotNull(taskExecutionState, "taskExecutionState");
+ // Use the main/caller thread for all updates to make sure they are
processed in order.
+ // (MainThreadExecutor i.e., the akka thread pool does not guarantee
that)
+ // Only detach for a FAILED state update that is terminal and may
perform io heavy labeling.
+ if
(ExecutionState.FAILED.equals(taskExecutionState.getExecutionState())) {
+ return labelFailure(taskExecutionState)
+ .thenApplyAsync(
+ taskStateWithLabels -> {
+ try {
+ return
doUpdateTaskExecutionState(taskStateWithLabels);
+ } catch (FlinkException e) {
+ throw new CompletionException(e);
+ }
+ },
+ getMainThreadExecutor());
+ }
try {
- checkNotNull(taskExecutionState, "taskExecutionState");
+ return CompletableFuture.completedFuture(
+ doUpdateTaskExecutionState(taskExecutionState));
+ } catch (FlinkException e) {
+ return FutureUtils.completedExceptionally(e);
+ }
+ }
+ private Acknowledge doUpdateTaskExecutionState(final TaskExecutionState
taskExecutionState)
+ throws FlinkException {
+ @Nullable FlinkException taskExecutionException;
+ try {
if (schedulerNG.updateTaskExecutionState(taskExecutionState)) {
Review Comment:
Thanks for updating the PR. @pgaref
I'm actually not against that restart strategies use failure labels. I'm
just hope that we can avoid some complication if possible. Totally +1 to the
concern about the risk to change some synchronous process into asynchronous
ones. But given that it is not avoidable, considering labeling failure from
with the JM, I prefer to do it in a common path. Absolutely later we need to
carefully review and test it.
Sorry that I did not have time to review it right now. I may take a look
tomorrow.
--
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]
