Fix Incorrect Cluster Size Update in ExperimentalControlLoop When Scaling Down is Disabled #728
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
fdc-ntflx
requested review from
calvin681,
sundargates,
Andyz26 and
hmitnflx
as code owners
fdc-ntflx
temporarily deployed
to
Integrate Pull Request
GitHub Actions
Inactive
Uploaded ArtifactsTo use these artifacts in your Gradle project, paste the following lines in your build.gradle. |
sundargates
approved these changes
Dec 5, 2024
| logger.info("scaleDownStage decrementing number of workers from {} to {}", numCurrentWorkers, desiredWorkers); | ||
| cancelOutstandingScalingRequest(); | ||
| final StageScalingPolicy scalingPolicy = stageSchedulingInfo.getScalingPolicy(); | ||
| if (scalingPolicy != null && scalingPolicy.isAllowAutoScaleManager() && !jobAutoscalerManager.isScaleDownEnabled()) { | ||
| logger.warn("Scaledown is disabled for all autoscaling strategy. For stage {} of job {}", stage, jobId); | ||
| return; | ||
| return false; | ||
| } | ||
| final Subscription subscription = masterClientApi.scaleJobStage(jobId, stage, desiredWorkers, reason) |
There was a problem hiding this comment.
why is this an observable? it feels like there's no need for a stream here?
There was a problem hiding this comment.
Good point. I didn't check the full logic, so I don't really know.
| @@ -38,7 +38,9 @@ protected Double processStep(Tuple3<String, Double, Integer> tup) { | |||
|
|
|||
| String reason = tup._1; | |||
| if (desiredNumWorkers < tup._3) { | |||
| scaler.scaleDownStage(tup._3, desiredNumWorkers, reason); | |||
| if (!scaler.scaleDownStage(tup._3, desiredNumWorkers, reason)) { | |||
| return tup._3 * 1.0; | |||
There was a problem hiding this comment.
can you add a comment on what this is doing?
fdc-ntflx
temporarily deployed
to
Integrate Pull Request
GitHub Actions
Inactive
fdc-ntflx
force-pushed
the
fix-worker-count-on-scale-disable
branch
from
1f3181e to
85a8242
Compare
fdc-ntflx
temporarily deployed
to
Integrate Pull Request
GitHub Actions
Inactive
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Context
Background
While monitoring the autoscaling behavior of our clusters, I observed an unexpected scale-down event in the shared MRE source ~3 AM PST, despite having scaling down disabled (new feature introduced for the next live event). This anomaly was not present in the PlayAPI source, which behaved correctly under the same conditions.
Issue
Upon investigation, I discovered that the
ExperimentalControlLoopclass, used by the Clutch RPS scaling policy, was incorrectly updating thecurrentSizeto thedesiredSizeeven when the actuator did not perform a scale-down due to scaling being disabled. This led to a discrepancy between thecurrentSizeand the actual cluster size, causing future scaling decisions to be based on incorrect data.Impact
This behavior resulted in a critical issue where, upon scaling up, the system could mistakenly reduce the cluster size. For example, with scaling down disabled:
Solution
The fix ensures that the currentSize is not updated when the actuator is unable to perform a scale-down due to scaling being disabled. This maintains the integrity of the cluster size data, preventing erroneous scaling operations.
Testing