Resource usage from workloads

[bassam]

We need to capture the concept of "using" a given resource like MySqlInstance or Bucket from a Workload or other kinds. We've discussed two approach for this usage, one that is weakly typed for example:

apiVersion: compute.conductor.io/v1alpha1
kind: Workload
...
  resources:
    - apiVersion: database.aws.conductor.io/v1alpha1
      kind: rdsinstance
      name: test-database
      namespace: default
...

where the the Workload has a list of references to resources that it "uses". This would help with the generation of secrets and other connection information, and also with future topology/region aware placement of the workload and its resources.

Another approach is to make the usage strongly typed, and similar to how volumes: in a PodSpec works. For example,

apiVersion: compute.conductor.io/v1alpha1
kind: Workload
...
  mysql:
    - name: test-database
       accessMode: readOnly
  bucket:
    - name: test-bucket
      accessMode: readWrite
...

The former approach enables extensibility by 3rd parties without needing to modify the workload spec. The latter providers a better user experience and enables us to set properties for "usage". One approach is to go with the strongly typed approach and work with APIMachinery SIG to enable strongly typed extensibility model.

This is related to #90.

[bassam]

This could also help with #149. We could automatically modify security groups, authorized networks, and such when we know that a workload is "using"/"consuming" a given resource.

[jbw976]

I don't think we will be able to address this in v0.1, but I think this is an important part of the experience nonetheless, moving to v0.2

[jbw976]

We already have support for declaring weakly typed usage of resources from workloads. This issue still captures whether we want to add strongly typed support, e.g.

apiVersion: compute.conductor.io/v1alpha1
kind: Workload
...
  mysql:
  ...
  bucket:

[ichekrygin]

Strongly typed resources looks like an interesting approach. However, it could be limiting when comes to trying to consume 3rd part (out of tree) resources inside the Workload.

I suspect we will need to support both.

[negz]

I don't believe this is relevant (at least in this specific context) with the direction we're going. We don't intend for Crossplane to manage workloads.