make it easy to run local builds of Java services as local docker builds (or Java mains) and have them register into kubernetes as Services

[jstrachan]

Imagine you're working on a service which has a number of containers invoking it. You don't want to have to run everything locally on your box; you just want to run this one service you're working on locally; while running everything else inside kubernetes.

So you want to reuse a kubernetes environment (namespace) running various pieces; but just work on 1 container locally using your local code before you commit.

So you wanna run a process locally (inside docker or outside) as if its part of kubernetes (like this issue https://github.com/fabric8io/fabric8/issues/3619) but you also want to register your service as a Service into kubernetes so that it can be discovered by other containers which are really running inside kubernetes.

e.g. imagine we want to experiment with or debug a local build of Fabric8MQ; but reuse the existing apps for the MQ producer & consumer inside kubernetes.

So we'd want to be able to:

• create a new kubernetes namespace for the developer
• run our local Fabric8MQ Java process and expose it on an IP address and port that's accessible via the kubernetes environment (this bit could be tricky - we may need some kind of ssh tunnel from your laptop into the kubernetes cluster - we can figure that bit out separately...)
• we define an environment variable / maven property for the IP address to listen on
• create a Service in the kubernetes namespace for the local process; using the portalIP of the IP address the service is listening on and port of the locally running service (there's no need to go the RC / pod / selector stuff - in this case just using our local service implementation as the only service implementation is totally fine)
• you can then create whatever kube applications/pods you need in this namespace to test out your service

Its then really easy to work on the code iteratively; debug your local build of the service - while reusing all the kubernetes infrastructure, apps, discovery etc.

i.e. its like you're laptop is part of kubernetes.

In many ways this is very much like the motivation behind https://github.com/fabric8io/fabric8/issues/3619 - but its focussing on working on a service implementation so that you can reuse other kubernetes resources to invoke / consume the service you're developing

[jstrachan]

Incidentally while developing various back end services (e.g. App Library, Fabric8 Forge, Api Library) I found myself doing something like the above; but as a horrid hack - I hacked the fabric8 console to redirect some services in the /api/v1beta2/proxy/services/$foo/* namespace to point at a locally running service.

This hack was invaluable in trying out services (which are not yet packaged/released as a docker image + app) from inside the fabric8 console. Though its a bit of a dirty hack really ;) What'd be much cleaner is if we had a way to run any Java process as if its really a Service implementation inside Kubernetes.

e.g. then I could reuse an entire hosted OpenShift namespace (its console & all its services) and just point one service implementation at my laptop's Java process its running.

Certainly for folks working on the console - this would be really really handy ;)

The hardest bit is figuring out the ssh tunnel / DNS / IP thing so that the kubernetes cluster can see my laptop's IP address.

@jimmidyson this sounds like a DNS problem - everything's a DNS problem right? :)

[jstrachan]

Maybe something like this could help being able to expose a local port to the internet so that it can be used from inside the Service definition inside kubernetes? https://ngrok.com/docs