Add systemd uptime metric collection

[langesven]

The no longer available systemd_exporter provided uptime metrics for systemd services, as such I figured I'd propose this as an extension to node_epxorter.
This is somewhat relevant to #562 in that it exposes how long a service has been running for, not quite the timestamp of its start.
Exporting the duration makes it easier to define alerts on this metric in my opinion and would retain the functionality as provided by the previous project.

[discordianfish]

@langesven Thanks for the PR!
Though I think we should expose the start timestamp since it's much more efficient for prometheus to store. To alert on the duration, you can alert on time() - timestamp_metric.
See also this older, outdated PR with similar functionality: #709

[langesven]

Hi @discordianfish - fair enough, I changed it to a CounterValue that is either 0 or the unix timestamp of when the job was started. Or would a Gauge be better suited but still keep track of just the timestamp?

[arianvp]

Note that there is also this PR #709

[discordianfish]

@langesven After some back and forth we decided to use gauges for timestamps in general, so yeah please change this to a gauge.

[discordianfish]

Looking good in general, but we should also have some tests if possible.

[discordianfish]

Let's call this unit_start_time_seconds to be consistent with already existing process_start_time_seconds.

[SuperQ]

@discordianfish That won't work, because process_start_time_seconds metric family has no labels, so the collector will fail.

[SuperQ]

I think for consistency, we could name it unit_start_time_seconds.

[discordianfish]

@SuperQ That's what I suggested! Should have put it in ```` though :)

[SuperQ]

Sorry! I didn't read what you said closely enough. 🦆

[discordianfish]

@langesven Can you change the metric name? Beside this LGTM
/cc @SuperQ

[langesven]

Thanks for your feedback.
I changed the metric name to unit_start_time_seconds and made it a GaugeValue instead of a counter!

[SuperQ]

LGTM

[SuperQ]

Oh, would you mind adding a [FEATURE] entry to the changelog?

[langesven]

Added a feature changelog entry and rebased against master

[discordianfish]

@langesven Looks like this still (again?) needs rebasing

[langesven]

@discordianfish it was indeed again 😄 rebased