Skip to content

Fix AdaptiveAllocationsScalerServiceTests.test_scaleDownToZero_whenNo… #137731

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jan-elastic merged 1 commit into from
Nov 7, 2025
Merged

Fix AdaptiveAllocationsScalerServiceTests.test_scaleDownToZero_whenNo… #137731

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions ...csearch/xpack/ml/inference/adaptiveallocations/AdaptiveAllocationsScalerServiceTests.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -275,7 +275,7 @@ public void test_scaleDownToZero_whenNoRequests() {
inferenceAuditor,
meterRegistry,
true,
ONE_SECOND,
2 * ONE_SECOND,
Copy link
Contributor Author

@jan-elastic jan-elastic Nov 7, 2025 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

These two settings:

  • timeIntervalSeconds of ONE_SECOND
  • scaleToZeroAfterNoRequestsSeconds of ATOMIC_SECOND

lead to a race condition. Adaptive allocations service checks after 1 second (the timeIntervalSeconds) whether there's been no request for 1 second (the scaleToZeroAfterNoRequestsSeconds).

On Unix-based systems this seems to always pass. Locally on my MacBook the timeWithoutRequestsSeconds is always something like 1.001 seconds, so the model is scaled to zero.

This test is probably failing on Windows builds on x86 hardware, because of a less precise system clock, see:
https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/high-resolution-timers#controlling-timer-accuracy. Not sure 100% sure, I'm no Windows expert by any means.

Increasing the timeIntervalSeconds to 2 seconds, and then checking whether there's been no request for 1 second, solves the issue. Unfortunately, this add 2 more seconds of unit test time, because the times are set in seconds instead of milliseconds.

ATOMIC_SECOND,
TWO_THOUSAND_MILLISECONDS
);
Expand All @@ -295,7 +295,7 @@ public void test_scaleDownToZero_whenNoRequests() {
return Void.TYPE;
}).when(client).execute(eq(GetDeploymentStatsAction.INSTANCE), eq(new GetDeploymentStatsAction.Request("test-deployment")), any());

safeSleep(1500);
safeSleep(2500);

verify(client, times(1)).threadPool();
verify(client, times(1)).execute(eq(GetDeploymentStatsAction.INSTANCE), any(), any());
Expand All @@ -317,7 +317,7 @@ public void test_scaleDownToZero_whenNoRequests() {
return Void.TYPE;
}).when(client).execute(eq(UpdateTrainedModelDeploymentAction.INSTANCE), any(), any());

safeSleep(1000);
safeSleep(2000);

verify(client, times(2)).threadPool();
verify(client, times(1)).execute(eq(GetDeploymentStatsAction.INSTANCE), any(), any());
Expand Down