Report

Reported in discussion #1382

KEDA_HTTP_CONNECT_TIMEOUT is used as the base duration for exponential backoff retries, not just as TCP connection timeout. This causes unnecessarily slow cold start responses.

With default KEDA_HTTP_CONNECT_TIMEOUT=500ms I'm seeing these numbers, its even worse with higher timeouts:

• Actual TCP connection time: <5ms
• Cold start response time: ~3.5s
• Expected response time: ~1.5s

Expected Behavior

Cold start requests should complete in ~1-1.5 seconds (pod startup time + minimal connection overhead).

Actual Behavior

Cold start requests take 3-3.5 seconds due to 500ms retry backoff intervals, even though TCP connections succeed in <5ms within the cluster.

Steps to Reproduce the Problem

1. Set high connect timeout: kubectl set env deployment/keda-add-ons-http-interceptor -n keda KEDA_HTTP_CONNECT_TIMEOUT=500ms
2. Deploy HTTPScaledObject with minReplicas: 0
3. Scale scaled deployment to 0: kubectl scale deployment <app> --replicas=0
4. Make request: time curl http://<ingress>/<path> -H "Host: <host>"

Logs from KEDA HTTP operator

no logs

HTTP Add-on Version

0.11.1

Kubernetes Version

1.33

Platform

Any

Anything else?

I will investigate this and create a PR.