AKS cold start?

Question

I have about 30 pods (F#/.NET Core + Azure Functions Runtime) running in a 3-node AKS cluster, each of which I am calling every 5 minutes externally (using Application Insights Availability/Web tests).
I am observing that if the pod is not called often (e.g. only 1x/every 5 minutes) then it takes 5000ms+ to call it (usually the call checks dependencies like if db is reachable, if storage blob container is reachable, or event hub). If I call the pod 10-20 times one after the other, I get very fast reponse times below 100ms.

First I thought the shared storage account (AzureWebJobsStorage env var) could be the issue, but then took 1 pod and reassigned it to a dedicated storage account, and after couple of minutes of inactivity again 5000ms+ response time.

Is there any kind of cold start, or automatic shut down of the pods in case they are inactive for a couple of minutes??

Answer 1

I have increased the number of calls to 1 pod and it seems that I do not experience such "cold-starts" with this pod now ... Somehow it seems there is some "idle timeout" and calling the pod 1x/every 5 minutes is too less to keep it alive - can this be??

Answer 2

Trying now with "functionTimeout": "-1" in host.json ....

Answer 3

I have found numerous posts about "5 seconds dns resolve kubernetes" - can this be the reason?

E.g.

Answer 4

Applied an old workaround I found googling (under spec/template/spec, same level as containers), seems to help ... Wondering why 1,5 year-old-problem is hitting on innocent AKS/K8s newbies like me in 2020 :(

  dnsConfig:  
    options:  
    - name: single-request-reopen