Hello @Gav Sturm,

Check and, if necessary, increase the timeout settings for the KEDA HTTP interceptor. Ensure that the readiness and liveness probes for your target pods are correctly configured and passing. Verify that the Nginx Ingress rules are properly routing traffic to the KEDA HTTP interceptor. Review the logs from the KEDA HTTP interceptor for any errors or warnings. Ensure that network policies allow traffic from the KEDA HTTP interceptor to your target pods. Confirm that the target pods have sufficient CPU and memory resources allocated.

If you have any further queries, do let us know.

Hello Gav Sturm,

You're encountering two known limitations in the current KEDA HTTP Add-on v0.10.0 that can be reproduced reliably in AKS environments.

Why 502 Bad Gateway After 22 Seconds happen?

Ans- When a request hits the Interceptor and the backend is scaled to 0, the Interceptor triggers a scale-up via the HTTP Scaled Object. However, the Interceptor doesn't wait long enough for the pod to become ready. Around the 20–25 second mark, it returns a 502 Bad Gateway because it didn’t receive a response from the backend in time. This timeout is controlled by interceptor.responseHeaderTimeout, which defaults to 30 secs in the Helm chart. This may not be long enough to cold-start workloads.

How to fix it?

Ans- You can try out below suggestions-

Set a longer timeout for the interceptor

something like-

interceptor:

  responseHeaderTimeout: 120s

and ensure the interceptor-proxy service is correctly configured. If targetPort is set to a named port like "proxy" instead of a number like 8080, the connection will silently fail.

If so then you can fix it using-

kubectl patch svc keda-add-ons-http-interceptor-proxy -n keda --type='json' \

  -p='[{"op": "replace", "path": "/spec/ports/0/targetPort", "value": 8080}]'

For your next question on why HPA ignores min: 0 — Always Sets minReplicas: 1

Ans- This is a known limitation in KEDA HTTP Add-on v0.10.0: the HPA created by the HTTP add-on is hardcoded to use minReplicas: 1, even if your HTTPScaledObject explicitly sets min: 0.

This breaks true scale-to-zero functionality once the deployment has been scaled up at least once the HPA keeps one replica running.

There is no configuration-based workaround in v0.10.0. You can track planned improvements here- https://github.com/kedacore/http-add-on/discussions or reach out to KEDA community for further clarity on this.
I haven't seen your configuration as of now. However, would like to suggest one additional point here as I was trying the same from my end and noticed that we need to avoid setting the name field inside scaleTargetRef. Only service and port should be set.

Why?
because if name is present, the generated workload reference becomes //backend, which breaks scaling entirely.

You edit it to something like this-

scaleTargetRef:

  service: backend

  port: 8080

Hope this clears your query. Thanks