Nsight Streamer 1.1.1 Helm Chart Broken

Developer Tools Nsight Cloud
1

I tried reporting this directly via email, but got an UNDELIVERABLE message back.

Your message to nsight-cloud-feedback@nvidia.com couldn’t be delivered.
nsight-cloud-feedback wasn’t found at nvidia.com.

Given the errors experienced, it appears that ClusterIP support is a yet incomplete afterthought. With service.type=ClusterIP, the chart cannot be deployed “as is” , due to nodePort being specified twice in the HTTP section of the service template:

  {{- if and (eq .Values.service.type "NodePort") (not (empty .Values.service.nodeHttpPort)) }}
      nodePort: {{ .Values.service.nodeHttpPort }}
  {{- end }}
      nodePort: {{ .Values.service.httpPort }}

The first occurrence is properly gated by service type, which is OK. The second unconditional one is likely a left-over, and MUST be removed to allow deploying the chart with ClusterIP.

With that fix, the chart can be deployed with ClusterIP, but it still won’t work as expected.

I’ve yet to find time to debug more carefully, but I suspect that it’s the TURN/ICE stuff that breaks, since the whole point withClusterIP is to avoid exposing internal services. Forwarding the TURN port made no difference.

I noticed that the connection info job is botched, since it’s looking for nodePort, when there’s none to be found, but I suspect that the job’s output is for information only, since fixing it didn’t help. (Or, maybe, usiing None, when nonodePort is present was the wrong fix.)

2

Additional “breakage” includes unused entries like env and defaultEnv in the values file…

That’s quite an oversight, I’d argue..

3

Running the chart with default values, except for selecting tool=ncu still doesn’t work.

I’m thoroughly surprised. I expected more.

4

@mpopov can you please help with the issue.

@mhari can you help with the bad email address.

5

We are aware of the issues concerning clusterId and environment configurations, and a new release addressing these will be available very soon.

Separately, to help us ensure the best possible experience, could you let us know if you’re encountering any specific problems with the default installation settings of the Helm chart?
You mentioned that the service is not working—could you please provide more details about the issue? For example, are you unable to connect to the HTTP port, or is the video stream not functioning?

We appreciate your assistance and valuable input.

6

Actually, the service started working even with ClusterIP after my fixes. Unsure why.

On my first attempts, I was using Firefox and I did get a placeholder and messages about
connecting, but the connections always failed, and had to be perpetually re-loaded.
Checking the pod logs, selkies was complaining about 2 things:

  • trying to read from a closed socket
  • wrong peer ID

Interestingly, all other browsers I tried (Edge, Safari, Chrome, and Brave) had the same issue initially,
but managed to recover after only a few such errors. Firefox did not.

Not until today, that is. WEIRD!