Prometheus metrics are broken

[Swift42]

System information

Geth version: 1.10.1
OS & Version: Linux

Expected behaviour

Prometheus should read the metrics without an error.

Actual behaviour

Prometheus scrape fails with "invalid metric type"
This happens because all of the new "les/clientcompact" and "les/clientdisk" metrics have a "." instead of "_" in the name.
E.g.
eth_db_les.clientcompact_input 0
should be
eth_db_les_clientcompact_input 0

[holiman]

I wonder if the problem isn't something else, namely that the namespace is "eth/db/les.client" and not "eth/db/les.client/".At least, that's one error, because eth/db/les.clientcompact is not a metric, but eth/db/les.client/compact should be

[holiman]

Could you check if #22459 solves it?

[Swift42]

I modified the source as suggested, but this doesn't solve the problem (at least the problem with prometheus).
Now instead of e.g.
eth_db_les.clientcompact_input 0
the metrics contain
eth_db_les.client_compact_input 0

There is still a "." which causes the problem.

And yes, I still think that the "." is the problem, because Prometheus reports (without the patch):
err="invalid metric type "clientcompact_input gauge""
and now (with the patch):
err="invalid metric type "client_compact_input gauge""

This means everything after the dot is not recognized correctly. And all other metrics don't contain a dot and these are working correctly.

Maybe this breaks non-prometheus metrics, but if I change the line
lesDb, err := stack.OpenDatabase("les.client", 0, 0, "eth/db/les.client/")
to
lesDb, err := stack.OpenDatabase("les.client", 0, 0, "eth/db/les/client/")
all is working again and Prometheus don't report an error anymore. Now the metrics contain correctly:

# TYPE eth_db_les_client_compact_input gauge
eth_db_les_client_compact_input 187338

[holiman]

Thanks for checking -- I've fixed both issues in the linked PR now!