fix(keda): correct EKS production autoscaling — Value semantics, reachable threshold, CPU safety net#10892
Open
skettkepalli wants to merge 3 commits into
Open
Conversation
…hold, CPU safety net Old trigger could never fire (threshold 0.75 vs observed max 0.08) and AverageValue math capped demand at 2 pods regardless of load. Pins query to the EKS cluster, codifies min 8, adds in-cluster CPU trigger. Co-Authored-By: skettkepalli <sridhar.ettkepalli@eclipse-foundation.org>
Each pod pins a 30-connection Hikari pool; 20 pods would need 600 of RDS max_connections=500. 15 x 30 = 450 leaves headroom for operations. Co-Authored-By: skettkepalli <sridhar.ettkepalli@eclipse-foundation.org>
Co-Authored-By: skettkepalli <sridhar.ettkepalli@eclipse-foundation.org>
netomi
reviewed
Jun 4, 2026
| esReplicaCount: 3 | ||
| keda: | ||
| enabled: true | ||
| clusterName: eks-staging |
Contributor
There was a problem hiding this comment.
for staging the name is eks-staging but for production it is eks-production-openvsx . Either use the suffix for both or none I guess?
Author
There was a problem hiding this comment.
We will need to recreate the staging cluster using IaC so will align it in IaC!
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Production autoscaling on EKS has never fired. Three issues in the ScaledObject:
AverageValuemetricType. Our query is a ratio that never exceeds 1.0, so the HPA could never ask for more thanceil(1/0.75)= 2 pods, even fully saturated.This also makes the manual minReplicas bump to 8 (done during cutover) permanent, so the next deploy doesn't silently drop it back to 6.
Changes:
metricType: Value— desired replicas now scale proportionally with the signaleks-production-openvsxvia a newkeda.clusterNamevalueValidation: server-side dry-run against prod, the exact rendered query checked against Grafana Cloud, and a replay of the proposed signal over the whole post-cutover week — zero scale events, no flapping, min-8 floor holds. We also rehearsed the full loop on eks-staging: under synthetic load the app scaled 2→6 in about 3 minutes with the expected math, and drained back 1 pod per 3 minutes after the 15-minute calm window. Staging was restored to its original state afterwards.
One known non-goal: the May 29 thread saturation was cold JVMs during a rollout, not a capacity problem — warmup/readiness gating is a separate follow-up.
Companion PR for aws-main: #10893
Signed-off-by: skettkepalli sridhar.ettkepalli@eclipse-foundation.org