Warn validators with slow hardware

[ggwpez]

Upon starting a node we currently print some brief hardware metrics.
This could be extended to include a warning in case the node has bad performance and was started with --validator.

The logic for comparing results and having a baseline is currently in utils/frame/benchmarking-cli, some of this would probably move to sc-sysinfo. sc-sysinfo would profit from this since it can use Throughput types (and other) instead of just u64 in some places.

[dvdplm]

@ordian Maybe this can be helpful in view of the disputes slashing roll-out as well?

[bkchr]

We already got the host performance check in Polkadot for this: https://github.com/paritytech/polkadot/blob/master/cli/src/host_perf_check.rs

[ggwpez]

We already got the host performance check in Polkadot for this …

Do you mean that we should not add it to Polkadot then, but still to Substrate? Or not at all?
I keep seeing people in the Kusama 1KV channel actually using the benchmark machine command, so they seem to like it.

[bkchr]

We can add it to Substrate, if it is done in a generic way. Aka chain implementors like Polkadot provide their own expected numbers that are treated as "good".

My comment was more an answer to @dvdplm.

But if we have your stuff, we can deprecate the host perf check.

[Szegoo]

@ggwpez I am gonna pick this up :)

[Szegoo]

@ggwpez You mentioned in https://github.com/paritytech/substrate/pull/12368 that we could move Metric to sc-sysinfo also, I will open a small PR for that if you didn't change your mind.

[ggwpez]

Please try to do it in one. I dont want to increase the review overhead for such a rather simple issue @Szegoo

[Szegoo]

@ggwpez Ok, I will close this issue and move Metric in the same PR.

[Szegoo]

@ggwpez Should we deprecate the host perf check in polkadot then and have it in substrate implemented in a generic way?

[ggwpez]

I have to look into the PVF check command again and chat with who created it. For now it is still fine to have both IMO.