echauchot commented on code in PR #22985:
URL: https://github.com/apache/flink/pull/22985#discussion_r1362263357
##########
flink-runtime/src/test/java/org/apache/flink/runtime/scheduler/adaptive/ExecutingTest.java:
##########
@@ -252,28 +259,94 @@ public void testTransitionToFinishedOnSuspend() throws
Exception {
}
@Test
- public void
testNotifyNewResourcesAvailableWithCanScaleUpTransitionsToRestarting()
+ public void
testNotifyNewResourcesAvailableBeforeCooldownIsOverScheduledStateChange()
+ throws Exception {
+ try (MockExecutingContext ctx = new MockExecutingContext()) {
+ // do not wait too long in the test
+ final Duration scalingIntervalMin = Duration.ofSeconds(1L);
+ final ExecutingStateBuilder executingStateBuilder =
+ new
ExecutingStateBuilder().setScalingIntervalMin(scalingIntervalMin);
+ Executing exec = executingStateBuilder.build(ctx);
+ ctx.setCanScaleUp(true); // => rescale
+ ctx.setExpectRestarting( // scheduled rescale should restart the
job after cooldown
+ restartingArguments -> {
+ assertThat(restartingArguments.getBackoffTime(),
is(Duration.ZERO));
+ assertThat(ctx.actionWasScheduled, is(true));
+ });
+ exec.onNewResourcesAvailable();
+ }
+ }
+
+ @Test
+ public void
testNotifyNewResourcesAvailableAfterCooldownIsOverStateChange() throws
Exception {
+ try (MockExecutingContext ctx = new MockExecutingContext()) {
+ final ExecutingStateBuilder executingStateBuilder =
+ new ExecutingStateBuilder()
+ .setScalingIntervalMin(Duration.ofSeconds(20L))
+
.setLastRescale(Instant.now().minus(Duration.ofSeconds(30L)));
+ Executing exec = executingStateBuilder.build(ctx);
+ ctx.setCanScaleUp(true); // => rescale
+ ctx.setExpectRestarting(
+ restartingArguments -> { // immediate rescale
+ assertThat(restartingArguments.getBackoffTime(),
is(Duration.ZERO));
+ assertThat(ctx.actionWasScheduled, is(false));
+ });
+ exec.onNewResourcesAvailable();
+ }
+ }
+
+ @Test
+ public void
testNotifyNewResourcesAvailableWithCanScaleUpWithoutForceTransitionsToRestarting()
throws Exception {
try (MockExecutingContext ctx = new MockExecutingContext()) {
Executing exec = new ExecutingStateBuilder().build(ctx);
ctx.setExpectRestarting(
restartingArguments -> {
- // expect immediate restart on scale up
+ // immediate rescale
assertThat(restartingArguments.getBackoffTime(),
is(Duration.ZERO));
+ assertThat(ctx.actionWasScheduled, is(false));
});
- ctx.setCanScaleUp(() -> true);
+ ctx.setCanScaleUp(true); // => rescale
exec.onNewResourcesAvailable();
}
}
@Test
- public void
testNotifyNewResourcesAvailableWithNoResourcesAndNoStateChange() throws
Exception {
+ public void
testNotifyNewResourcesAvailableWithCantScaleUpWithoutForceAndCantScaleUpWithForce()
+ throws Exception {
try (MockExecutingContext ctx = new MockExecutingContext()) {
- Executing exec = new ExecutingStateBuilder().build(ctx);
- ctx.setCanScaleUp(() -> false);
+ Executing exec =
+ new ExecutingStateBuilder()
+ .setScalingIntervalMax(Duration.ofSeconds(1L))
+ .build(ctx);
+ ctx.setCanScaleUp(
+ false, false); // => schedule force rescale but resource
lost on timeout =>
+ // no rescale
exec.onNewResourcesAvailable();
ctx.assertNoStateTransition();
+ assertThat(ctx.actionWasScheduled, is(true));
Review Comment:
Even if minMet and parallelismChangeAfterTimeout names kind of leaked
implementation details, I believe it is still better than the new names that
are too far from the tested use cases
##########
flink-runtime/src/test/java/org/apache/flink/runtime/scheduler/adaptive/ExecutingTest.java:
##########
@@ -252,28 +259,94 @@ public void testTransitionToFinishedOnSuspend() throws
Exception {
}
@Test
- public void
testNotifyNewResourcesAvailableWithCanScaleUpTransitionsToRestarting()
+ public void
testNotifyNewResourcesAvailableBeforeCooldownIsOverScheduledStateChange()
+ throws Exception {
+ try (MockExecutingContext ctx = new MockExecutingContext()) {
+ // do not wait too long in the test
+ final Duration scalingIntervalMin = Duration.ofSeconds(1L);
+ final ExecutingStateBuilder executingStateBuilder =
+ new
ExecutingStateBuilder().setScalingIntervalMin(scalingIntervalMin);
+ Executing exec = executingStateBuilder.build(ctx);
+ ctx.setCanScaleUp(true); // => rescale
+ ctx.setExpectRestarting( // scheduled rescale should restart the
job after cooldown
+ restartingArguments -> {
+ assertThat(restartingArguments.getBackoffTime(),
is(Duration.ZERO));
+ assertThat(ctx.actionWasScheduled, is(true));
+ });
+ exec.onNewResourcesAvailable();
+ }
+ }
+
+ @Test
+ public void
testNotifyNewResourcesAvailableAfterCooldownIsOverStateChange() throws
Exception {
+ try (MockExecutingContext ctx = new MockExecutingContext()) {
+ final ExecutingStateBuilder executingStateBuilder =
+ new ExecutingStateBuilder()
+ .setScalingIntervalMin(Duration.ofSeconds(20L))
+
.setLastRescale(Instant.now().minus(Duration.ofSeconds(30L)));
+ Executing exec = executingStateBuilder.build(ctx);
+ ctx.setCanScaleUp(true); // => rescale
+ ctx.setExpectRestarting(
+ restartingArguments -> { // immediate rescale
+ assertThat(restartingArguments.getBackoffTime(),
is(Duration.ZERO));
+ assertThat(ctx.actionWasScheduled, is(false));
+ });
+ exec.onNewResourcesAvailable();
+ }
+ }
+
+ @Test
+ public void
testNotifyNewResourcesAvailableWithCanScaleUpWithoutForceTransitionsToRestarting()
throws Exception {
try (MockExecutingContext ctx = new MockExecutingContext()) {
Executing exec = new ExecutingStateBuilder().build(ctx);
ctx.setExpectRestarting(
restartingArguments -> {
- // expect immediate restart on scale up
+ // immediate rescale
assertThat(restartingArguments.getBackoffTime(),
is(Duration.ZERO));
+ assertThat(ctx.actionWasScheduled, is(false));
});
- ctx.setCanScaleUp(() -> true);
+ ctx.setCanScaleUp(true); // => rescale
exec.onNewResourcesAvailable();
}
}
@Test
- public void
testNotifyNewResourcesAvailableWithNoResourcesAndNoStateChange() throws
Exception {
+ public void
testNotifyNewResourcesAvailableWithCantScaleUpWithoutForceAndCantScaleUpWithForce()
+ throws Exception {
try (MockExecutingContext ctx = new MockExecutingContext()) {
- Executing exec = new ExecutingStateBuilder().build(ctx);
- ctx.setCanScaleUp(() -> false);
+ Executing exec =
+ new ExecutingStateBuilder()
+ .setScalingIntervalMax(Duration.ofSeconds(1L))
+ .build(ctx);
+ ctx.setCanScaleUp(
+ false, false); // => schedule force rescale but resource
lost on timeout =>
+ // no rescale
exec.onNewResourcesAvailable();
ctx.assertNoStateTransition();
+ assertThat(ctx.actionWasScheduled, is(true));
Review Comment:
Yes it is expected: as long as resources are not met we schedule a force
rescale after scalingIntervalMax. This test is to simulate that the resource
was lost in between, hence there is no rescale (because parallelism has not
changed). But still it is normal that we have scheduled a forceRescale.
We are exactly in the case I mentioned yesterday, by renaming `minMet` to
`canScaleUpWithoutForce` and `parallelismChangeAfterTimeout` to
`canScaleUpWithForce` we have lost all the expressiveness of the tests.
Even if minMet and parallelismChangeAfterTimeout names kind of leaked
implementation details, I believe it is still better than the new names that
are too far from the tested use cases
--
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: issues-unsubscr...@flink.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
