Allow scaling job workers down to 0

[npepinpe]

Is your feature request related to a problem? Please describe.

As an operator, I would like to be able to reduce my cloud footprint (including cost) to a minimum when unused. Right now, I need to have my workers connected to Zeebe all the time so they can receive jobs in order to know when there are some jobs.

Describe the solution you'd like

I'd like a way to know how many jobs I have per job type which I can scrape, and have an auto-scaler react to.

Describe alternatives you've considered

A custom metrics exporter which looks at job events to determine how many jobs there are per type, and exposes them via Prometheus. However, it's currently difficult to do so: see #18467.

Additional context

• Provide observation context to exporters #18467

[Zelldon]

Triage:

• moving to backlog as I don't see an urgency here
• Please speak up if you think otherwise

[tmetzke]

ZPA triage:

• Looks like a ZDP issue, removing it from our board for now. If you think there is anything for us to take over here, please re-add.

[npepinpe]

I would expect this is to be primarily ZPA issue, actually. It's about exposing a metric which displays the count of jobs per job type, which is application specific, behind a feature flag (since we may want to disable it in SaaS due to the "infinite" cardinality of such a metric). Alternatively, it could be a simple actuator which exposes the latest state, as this can also be consumed by a custom application metrics adapter for Kubernetes.

Just my two cents 🤷

[tmetzke]

I somehow considered it a more infrastructural issue provided by the platform. But the way you described it makes sense, @npepinpe. I'll re-add it to our board and we'll discuss it in the next team triage 🙂