Count usage of process start-up

[PierreF]

When a process started between two Update(), count the usage between process creation and the first Update() which see it.

This is especially useful for short-lived processes & not too fast gathering (like process lifetime of 1 minutes and metric gathering every 10 seconds), where I had about 15% error in my case:

I'm measuring CPU usage of a PostgreSQL server behind a pgbouncer configured to keep connection 60 seconds (which means that every 60 seconds, the connection is closed & re-opened, which on Postgres means a process is terminated and a new one is re-created).

The real CPU usage is measured using ps --cumulative on parent process (which is a good approximation on long term, and a correct minimal value. Using this does not account CPU usage of currently running child but do count exactly CPU usage of all terminated children).

Without this PR, I have the following delta (over 1h30):

• CPU seconds between two ps: 77 seconds
• CPU seconds between two namedprocess_namegroup_cpu_seconds_total (both user + system): 65.2 seconds
• Error: 11.80 seconds or 15%

Without this PR, I have the following delta (over 30 minutes):

• CPU seconds between two ps: 28 seconds
• CPU seconds between two namedprocess_namegroup_cpu_seconds_total (both user + system): 28.04 seconds
• Error: none :)

[PierreF]

Is there something missing for this PR to be merged ?

[ncabatoff]

Hi @PierreF,

The patch looks good, thanks very much! Sorry for the delay, I haven't worked on this project in quite a while, I needed to carve out some time to remember how it works so I could make sense of your change.

I made a minor suggestion to fix a typo in a comment, if you wouldn't mind fixing that I'll be happy to merge this.

[PierreF]

Sorry, missed the review. Fixed now!

[ncabatoff]

Thanks @PierreF !