Add metrics to snet

[scrye]

Fixes #3107

Adds the following metrics:

# HELP lib_snet_closes_total Total number of Close calls.
# TYPE lib_snet_closes_total counter
lib_snet_closes_total 0
# HELP lib_snet_dials_total Total number of Dial calls.
# TYPE lib_snet_dials_total counter
lib_snet_dials_total 0
# HELP lib_snet_dispatcher_error_total Total number of dispatcher errors
# TYPE lib_snet_dispatcher_error_total counter
lib_snet_dispatcher_error_total 0
# HELP lib_snet_listens_total Total number of Listen calls.
# TYPE lib_snet_listens_total counter
lib_snet_listens_total 2
# HELP lib_snet_parse_error_total Total number of parse errors
# TYPE lib_snet_parse_error_total counter
lib_snet_parse_error_total 0
# HELP lib_snet_read_total_bytes Total number of bytes read
# TYPE lib_snet_read_total_bytes counter
lib_snet_read_total_bytes 1.163921e+06
# HELP lib_snet_read_total_pkts Total number of packetes read
# TYPE lib_snet_read_total_pkts counter
lib_snet_read_total_pkts 3105
# HELP lib_snet_scmp_error_total Total number of SCMP errors
# TYPE lib_snet_scmp_error_total counter
lib_snet_scmp_error_total 0
# HELP lib_snet_write_total_bytes Total number of bytes written
# TYPE lib_snet_write_total_bytes counter
lib_snet_write_total_bytes 579712
# HELP lib_snet_write_total_pkts Total number of packets written
# TYPE lib_snet_write_total_pkts counter
lib_snet_write_total_pkts 3037

---

This change is 

[lukedirtwalker]

Reviewed 6 of 6 files at r1.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @scrye)

---

go/lib/snet/dispatcher.go, line 103 at r1 (raw file):

	hdr, ok := pkt.L4Header.(*scmp.Hdr)
	if !ok {
		metrics.M.SCMPErrors().Inc()

Hm I wonder if that's what kormat meant with SCMP errors? I thought maybe more something along the lines of SCMP errors received.

[scrye]

Reviewable status: 5 of 6 files reviewed, 1 unresolved discussion (waiting on @lukedirtwalker)

---

go/lib/snet/dispatcher.go, line 103 at r1 (raw file):

Previously, lukedirtwalker (Lukas Vogel) wrote…

Hm I wonder if that's what kormat meant with SCMP errors? I thought maybe more something along the lines of SCMP errors received.

My approach was:

• if no handler, consider the "unhandled SCMP" as an error;
• handlers are free to then collect metrics for which SCMP parts they think are relevant;

I've changed the behavior in the PR a little bit right now:

• calling the SCMP handler with a non-SCMP message is no longer considered an error;
• if no handler found, assume the SCMP is an error (same as before), because it might very well be one;
• the default handler increases the errors counter appropriately now;

[lukedirtwalker]

Reviewed 1 of 1 files at r2.
Reviewable status: complete! all files reviewed, all discussions resolved