[Suggestion] fix ordering in YAML output

[tabern]

Is your feature request related to a problem? Please describe.
Synthesized YAML shows up out of order from standard K8s format. It technically works, but for a user familiar with K8s YAML formatting can be confusing to read.

example:

spec:
  type: LoadBalancer
  ports:
    - port: 80
      targetPort: 8080
  selector:
    app: myappnginx37F28063
kind: Service
apiVersion: v1
metadata:
  name: myapp.nginx.service.5f289726

Describe the solution you'd like
Follow K8s YAML formatting conventions

1. apiVersion
2. kind
3. metadata
4. spec
5. etc...

example:

apiVersion: v1
kind: Service
metadata:
  name: myapp.nginx.service.5f289726
spec:
  type: LoadBalancer
  ports:
    - port: 80
      targetPort: 8080
  selector:
    app: myappnginx37F28063

There may be a styling guide somewhere but I can't find it - looking at K8s documentation

[eladb]

Pointer to where we serialize to yaml: https://github.com/awslabs/cdk8s/blob/fcb81a8b32dd37ccfdbb5de018ece93b92ce84bf/packages/cdk8s/lib/chart.ts#L77

[campionfellin]

I'll take this!

I think we can start with kubernetes/examples as our style guide for now and go from there.

For the top-level, I agree with the proposed solution of:

1. apiVersion
2. kind
3. metadata
4. spec (or other)

Looking at their examples, most generally follow this (though some, like cassandra-statefulset.yaml have 2 different orders in the same document.

I think we can put 1-3 in that order, and after that, order from shortest to longest, and in deeper levels of definitions, follow the shortest to longest as well. As such, when you scroll down in a file, you are reading inward on levels of deepness, and are not popping back out to a higher level without seeing its context.

Will poke around in the code mentioned above to find a solution.

[eladb]

I would take a bit of a more simple approach: apiVersion, kind, metadata, spec and then alpha-sort, and also do that recursively. It is very important that the order will be deterministic since people will likely use "diff" to identify changes to their definitions, so we can't afford any non-determinism.