Explicitly define default requests and limits for engine container

[groszewn]

Describe the bug

The controller only defines a default CPU Request currently, with custom specification of resources on the engine available through the svcOrch specification. As we tighten up our underlying K8s installation with OPA policies (specifically a policy requiring both requests and limits for CPU and memory be set), SeldonDeployments without engine resources defined in the svcOrch spec fail to deploy.

To reproduce

1. Deploy any SeldonDeployment with svcOrch resources specified in a cluster with an OPA policy requiring that containers have resources defined.

Expected behaviour

The default installation can come up without being blocked by OPA.

Environment

K8s installation with OPA policies enabled (specifically requiring requests/limits).

Model Details

• Images of your model: [Output of: kubectl get seldondeployment -n <yourmodelnamespace> <seldondepname> -o yaml | grep image: where <yourmodelnamespace>]
• Logs of your model: [You can get the logs of your model by running kubectl logs -n <yourmodelnamespace> <seldonpodname> <container>]

[ukclivecox]

@groszewn do you have some suggestions of reasonable defaults?

[groszewn]

I've been specifying 1 CPU and 512Mi of memory, but I haven't really run a stress test in a while. Definitely open to guidance if you have any thoughts here.

[ukclivecox]

I think that is reasonable. The memory may be dependent on the concurrent model payloads running through the executor.