Deprecating current Kubernetes implementation

[TwiN]

Hello 👋

I've been thinking about deprecating the current (alpha) implementation of Kubernetes (see Kubernetes (ALPHA) in the README).

Here are the reasons why:

1. I'm not sure anybody even uses it

As much as I love Kubernetes from the depth of my heart, the dependencies are just such a pain to manage, and if nobody's using it, then I feel it's a pain I can live without.

2. I feel like the current implementation promotes bad practices

When I created Gatus, one of the biggest motivators was that I wanted an application that only notified me when something actually worth checking was happening.

The current Kubernetes implementation promotes having a 1-template-fits-all, which I think goes against that ideology; each service configuration should be tailored to that service, and even though you can do some basic filtering (through kubernetes.namespaces[].name and kubernetes.namespaces[].excluded-services), I personally think it's not specific enough.

Mind you, this reason is very opinionated. Really, you can use Gatus for whatever you want -- hell, I use Gatus for load testing, which is definitely not its intended purpose -- but as the maintainer, I have a vision for Gatus, and it strongly includes preventing Gatus from becoming "one of these tools that sends too many alerts and ends up being muted".

3. The current implementation could be externalized

By having another application whose primary task is auto discovery "inject" a configuration (e.g. in a configmap).

4. Readiness and liveness probes aren't supported

This is something that's been on my agenda, in fact, there's even an issue open about it: #5.

All the current implementation does is look for services (k8s) in a list of namespaces, and filter some of them out, then create a service (gatus) template that will fit every services (k8s) in a given namespace (e.g. "find all services in namespace X and evaluate the same endpoint and the same conditions on each of these services")

To clarify, I do plan on adding support for using the readiness & liveness probe statuses as a condition at one point, and may re-use some of the code that would be removed as a result of this deprecation, but I'd rather design it from scratch when I get to that point.

---

My mind is pretty much made up, but I'm just throwing it out there to see if anybody has something to say on the topic that might further convince me (or convince me otherwise).

[TwiN]

⚠️ A deprecation message for Kubernetes has been added as of v2.8.1. Said deprecation message has a link to this discussion.

[TwiN]

Considering that nobody has spoken against the idea and a sufficient grace period has been given, the current Kubernetes implementation will be officially removed from v3.0.0.

See #166

[TwiN]

v3.0.0 has been released today, and the Kubernetes auto discovery has been removed with it.