add a few simple prometheus metrics

[iameli]

Please let me know if any of these already exist or if there are better ways to handle some of these situations.

• livepeer_broadcaster_count — 1 if running with -broadcaster, 0 otherwise
• livepeer_orchestrator_count — 1 if running with -orchestrator, 0 otherwise
• livepeer_transcoder_count — 1 if running with -transcoder, 0 otherwise

These will let us sum the currently running counts of Bs, Os, and Ts without resorting to building the numbers from the local Kubernetes environment.

• livepeer_gpu_count — If running as a -transcoder with GPUs, how many GPUs are we running with?
• livepeer_gpu_nvidia_count — How many -nvidia GPUs are we using?
• livepeer_gpu_vaapi_count — How many VAAPI GPUs are we using?

[iameli]

@ya7ya Do these make sense to you? Anything you'd want to add here?

[darkdarkdragon]

@iameli
go-livepeer set node_type tag on all the exported metrics.
So, for getting number of running broadcasters you can do
sum(livepeer_versions{node_type="bctr"})
for orchestrators
sum(livepeer_versions{node_type="orch"})

You can't get number of running transcoders directly because standalone transcoders does not export Prometheus metrics (see discussion here #747)
But every orchestrator has metrics livepeer_transcoders_number - number of transcoders connected to that orchestrator - so you can use that to count transcoders.

livepeer_gpu_count

There is issue about that: #1341