rpuch commented on code in PR #6330:
URL: https://github.com/apache/ignite-3/pull/6330#discussion_r2239571638
##########
modules/low-watermark/src/main/java/org/apache/ignite/internal/lowwatermark/LowWatermarkImpl.java:
##########
@@ -212,12 +222,54 @@ public void getLowWatermarkSafe(Consumer<@Nullable
HybridTimestamp> consumer) {
}
private void scheduleUpdateLowWatermarkBusy() {
+ scheduleUpdateLowWatermarkTaskLock.lock();
+
+ try {
+ ScheduledUpdateLowWatermarkTask lastTask =
lastScheduledUpdateLowWatermarkTask.get();
+ ScheduledUpdateLowWatermarkTask newTask = new
ScheduledUpdateLowWatermarkTask(this, State.NEW);
+
+ State lastTaskState = lastTask == null ? State.COMPLETED :
lastTask.state();
+
+ switch (lastTaskState) {
+ case NEW:
+ if (lastTask.tryCancel()) {
+ boolean casResult =
lastScheduledUpdateLowWatermarkTask.compareAndSet(lastTask, newTask);
+
+ assert casResult : "It is forbidden to set a task in
parallel";
+
+ scheduleUpdateLowWatermarkTaskBusy(newTask);
+ }
+
+ break;
+ case IN_PROGRESS:
+ case CANCELLED:
Review Comment:
Why will it be rescheduled? And why don't we need to reschedule our new task
if the previous task cancelled (and, hence, it did not schedule any
continuation)?
##########
modules/low-watermark/src/main/java/org/apache/ignite/internal/lowwatermark/LowWatermarkImpl.java:
##########
@@ -124,6 +128,10 @@ public class LowWatermarkImpl extends
AbstractEventProducer<LowWatermarkEvent, L
private final Map<UUID, LowWatermarkLock> locks = new
ConcurrentHashMap<>();
+ private final AtomicReference<ScheduledUpdateLowWatermarkTask>
lastScheduledUpdateLowWatermarkTask = new AtomicReference<>();
Review Comment:
This seems to be only accessed under the lock. Why is `AtomicReference`
needed here?
##########
modules/low-watermark/src/main/java/org/apache/ignite/internal/lowwatermark/LowWatermarkImpl.java:
##########
@@ -339,16 +395,12 @@ CompletableFuture<Void> updateAndNotify(HybridTimestamp
newLowWatermark) {
.whenCompleteAsync((unused, throwable) -> {
if (throwable != null) {
if (!(hasCause(throwable,
NodeStoppingException.class))) {
- LOG.error("Failed to update low
watermark, will schedule again: {}", throwable, newLowWatermark);
+ LOG.error("Failed to update low
watermark: {}", throwable, newLowWatermark);
failureManager.process(new
FailureContext(CRITICAL_ERROR, throwable));
-
- inBusyLock(busyLock,
this::scheduleUpdateLowWatermarkBusy);
Review Comment:
Why is this removed?
##########
modules/low-watermark/src/test/java/org/apache/ignite/internal/lowwatermark/LowWatermarkImplTest.java:
##########
@@ -336,6 +342,36 @@ void
testUpdateAndNotifyNotInvokeFailureManagerWhenGetNodeStoppingException() {
verify(failureManager, never()).process(any());
}
+ @Test
+ void testParallelScheduleUpdates() throws Exception {
+ assertThat(lowWatermarkConfig.updateIntervalMillis().update(300L),
willCompleteSuccessfully());
+
+ assertThat(lowWatermark.startAsync(new ComponentContext()),
willCompleteSuccessfully());
+
+ runRace(
+ () -> lowWatermark.scheduleUpdates(),
+ () -> lowWatermark.scheduleUpdates(),
+ () -> lowWatermark.scheduleUpdates(),
+ () -> lowWatermark.scheduleUpdates()
+ );
+
+ Thread.sleep(1_000);
+
+ verify(lwmChangedListener, atLeast(2)).notify(any());
Review Comment:
Why is it at least 2? It seems that each subsequent task might cancel the
previous one, so why can't it be just 1 (the last one)?
--
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: notifications-unsubscr...@ignite.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
