Remove metrics global registry

[lilic]

I would like to suggest to remove the global registry as currently the controller-runtime global registry does not capture metrics registered onto prometheus.DefaultRegisterer. As a step forward and in line with the prosed KEP, the controller-runtime should be able to have the registry injected. (for a start with prometheus.DefaultRegisterer, with the long term goal of removing the global registry in the first place, I am open for suggestions on that.)

As per the proposal kubernetes/community#2909:

Kubernetes also makes extensive use of a global metrics registry to register metrics to be exposed. Aside from general shortcomings of global variables, Kubernetes is seeing actual effects of this, causing a number of components to use sync.Once or other mechanisms to ensure to not panic, when registering metrics. Instead a metrics registry should be passed to each component in order to explicitly register metrics instead of through init methods or other global, non-obvious executions.

Lastly, this may be my own opinion, but introducing globals as a library hides its dependencies, therefore I think having the registry be injected makes the API more clear.

[DirectXMan12]

Disclaimer: the below represents my current thoughts, but I'm generally open to being convinced otherwise :-)

The original goal was to make registering metrics "easy", while also not necessarily polluting the default Prometheus registry. IMO, this is a pattern that the default Prometheus docs get right, especially for smaller projects -- since the metrics registry is in a global, there's very little code or cognitive overhead in registering a new metric -- you just declare it at the top of your file, and start using it. Fundamentally, most programmers are lazy, and so the easier we can make it to add metrics, the more likely people are to do it. I don't want to give people excuses to put off writing metrics until later :-). Overhead leads to all sorts of excuses -- "oh, we can't put metrics here without refactoring to plumb through the registry", "oh we can't put metrics here because this is a free-standing function and not a method", etc. You might still get metrics, but you'll miss out on some that people didn't think were worth the effort.

With that motivation in mind, I think our current setup is a decent compromise -- it doesn't use the default registry -- it's actually separate, so you can choose to use it or not use it. However, you still get a similar feel: instead of just importing prometheus you have to import prometheus and metrics, but you get a similar feel -- declaring metrics.Registry.MustRegister at the top of your file, and then using the metrics as normal.

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[fejta-bot]

Stale issues rot after 30d of inactivity.
Mark the issue as fresh with /remove-lifecycle rotten.
Rotten issues close after an additional 30d of inactivity.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle rotten

[lilic]

/remove-lifecycle rotten

Personally, still think we should move away from it :)

[DirectXMan12]

yeah, I think this still bears further discussion. Likely what'll happen is that we'll eagerly go the direction that k/k goes -- if it looks like k/k is switching to opencensus/opentelemetry, we'll head in that direction eagerly (I expect we'll need to wait for opentelemetry's reference Go client in september or so, unless the opencensus --> opentelemetry switch is easy). At any rate, I expect when we switch we'll want to go in the direction of something like opentelemetry to make plumbing thing through easier.

[fejta-bot]

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

[DirectXMan12]

/lifecycle frozen

still want to do this