[jellyfin] Customizable readiness probe?

[onichandame]

Is your feature request related to a problem ?

Hi, I am running the jellyfin in a cluster with only 1 GPU available, so I have to scale it to zero when not needed and scale it back from zero when requests come in, so that my other services can take the GPU occasionally. I noticed that the scaling happens really quick after the request is initiated from the client side. But the pod stays in 0/1 Running state for like 10 seconds before it can handle the requests.

Having inspected the templates I suspect that the readiness probe's periodSeconds field is in play. So I manually edited the sts manifest to reduce this field to 5, and it takes only around 5s to become ready

Reading the logs I realize that the server finishes preparation in 4 seconds, so a 5-second delay of readiness probe makes sense in my case(probably also works in many people's case as my machine is using a very outdated CPU).

So what is the best solution in this case? I want to reduce the periodSeconds but I am not sure if it should stay hard coded in the templates or extracted to the values for more flexibility.

Describe the solution you'd like.

• reduced perioedSeconds in the template

Or

• customizable probes

Describe alternatives you've considered.

Fork the chart? IDK...

Additional context.

No response

[djjudas21]

Hey, thanks for your report. I'll have a proper look at the probes this afternoon, but I think it's also worth saying you can share a GPU between containers by specifying fractional requests, provided they sum to 1. You could set high requests for Jellyfin and it will get priority, but set a lower request for your other workload and it will also be able to use the GPU.

[onichandame]

@djjudas21 Thanks for the suggestion. My gpu is 2060 super and does not support any GPU virtualization technique(MIG, vGPU etc). So I didn't even think of sharing it across multiple pods at the same time. I'll try what you said