Deployment model for trustee helm chart

[Xynnn007]

Based on last (6.15) trustee meeting, the topic is rised without blocking #1150

Different Models

Dimension	Model 1: All-in-one KBS+Builtin AS	Model 2: All services in one Pod	Model 3: One service per Pod (current)
Pod/Process topology	1 Pod, 1 main process	1 Pod, multiple containers	Multiple Deployments/Pods
Deployment complexity	Lowest	Medium	Highest
Independent scaling	No	No (pod-level scaling only)	Yes (KBS / AS / RVPS independently)
Independent rollout/rollback	No	Limited	Yes
Failure isolation	Weak	Medium	Strong
Network attack surface	Smallest (mostly in-process calls)	Small (can use localhost between containers)	Largest (service-to-service traffic across cluster network)
External access for KBS	Yes	Yes	Yes
External access for AS	No (not separately exposed)	Optional	Optional
Risk of exposing internal interfaces	Low	Low-Medium	Medium-High if misconfigured (AS/RVPS as LB)
Resource scheduling flexibility	Low	Low	High (e.g., AS on dedicated nodes)

From my side, there are needs that require directly connects to AS's evaluate API. Thus the deployment needs to expose AS API also. But we now do not have control for "which gRPC API" can be exposed, thus the set_policy would also be exposed if AS API is exposed.

Anyway, let me bring this topic out, and we can iterate without blocking current helm PR.

[Xynnn007]

cc @mythi @fitzthum

[mythi]

Thanks for the summary! It is clear that one model does not work for all usages so we just need to find a way to support those. I think there are additional dimensions to consider like: what is the model we target for the majority of user base and what is the model that is consistent with trustee-operator.

[Xynnn007]

If we cannot get agreement very quickly, my suggestion is to first merge the current full microservice model, and then support all models "simultaneously" later. It can be controlled by a option in values.yaml.