Optimize cpufreq collector

[SuperQ]

Move metric descriptiions to package vars to avoid allocating them every time NewCPUFreqCollector() is called.

Signed-off-by: Ben Kochie [email protected]

[SuperQ]

CC @britcey, I don't think this will help your situation, unless you're using the collect[] URL param.

[dswarbrick]

Isn't NewCPUFreqCollector() only called once at startup, i.e. by NewNodeCollector() calling the various collectors' factory functions?

I don't really see how this is any significant optimization.

[discordianfish]

@SuperQ hrmm @dswarbrick is right, why this change?

[SuperQ]

No, this can be called on every scrape if you call it with the collect[] URL param.

[discordianfish]

Ah I see.. maybe then we should go through all collectors to look for similiar saving.

[dswarbrick]

Ah I see.. maybe then we should go through all collectors to look for similiar saving.

Oof. That's going to result in a LOT of package-level variables pollution.

[discordianfish]

@dswarbrick Yeah I'm worrying about that as well. I wanted to refactor the collector package for a while and move related functionality to subpackages but since nobody is paying me currently for anything prometheus related, I won't get to that anytime soon..

[SuperQ]

Eh, I don't see the package vars for the collector Descs as a big deal. I think we should migrate collectors to this new pattern. It's not a huge performance change for most users, but it would be good to keep things consistent.

We actually discussed this pattern as part of our refactoring the postgres_exporter to use a similar collector package. We decided to go with package level Desc vars to avoid any reallocaiton.