Meminfo metric naming inconsistency

[l4l]

Currently /proc/meminfo metrics have suffix _bytes (which is great), e.g MemTotal becomes MemTotal_bytes:

node_exporter/collector/meminfo_linux.go

Lines 52 to 63 in 654f19d

	if meminfo.ActiveBytes != nil {
	metrics["Active_bytes"] = float64(*meminfo.ActiveBytes)
	}
	if meminfo.ActiveAnonBytes != nil {
	metrics["Active_anon_bytes"] = float64(*meminfo.ActiveAnonBytes)
	}
	if meminfo.ActiveFileBytes != nil {
	metrics["Active_file_bytes"] = float64(*meminfo.ActiveFileBytes)
	}
	if meminfo.AnonHugePagesBytes != nil {
	metrics["AnonHugePages_bytes"] = float64(*meminfo.AnonHugePagesBytes)
	}

however their NUMA-counterpart does not (i.e MemTotal remains "as is"):

node_exporter/collector/meminfo_numa_linux.go

Lines 152 to 154 in 654f19d

	// Active(anon) -> Active_anon
	metric = re.ReplaceAllString(metric, "_${1}")
	memInfo = append(memInfo, meminfoMetric{metric, prometheus.GaugeValue, parts[1], fv})

I suppose this is a bug?

[discordianfish]

I guess we could append _bytes there as well. In general we expose metrics as-is if we collect them dynamically (e.g by looping over some procfs files etc), while we change names according to naming best practices if we expose them explicitly.