MeshSync: non-Kubernetes / multi-platform discovery (design)

[leecalcote]

Summary

Let MeshSync discover non-Kubernetes platforms (Docker first, cloud later) into Server's already kind-agnostic Connection state machine, via an infrastructure-agnostic model - incrementally, not a big-bang rewrite.

Key decisions

• Phase 0 enabling refactor: a Discoverer interface (Discover/Watch/ScopeID/Shutdown) with today's Kubernetes machinery as the first adapter (a thin wrapper, zero behavior change). Independently low-risk; unblocks everything non-K8s.
• Schema-first: a new platform-agnostic resource envelope construct in meshery/schemas; MeshSync ships a hand-written mirror first (the codegen swap is deferred to avoid compounding scope). ScopeID generalizes clusterID (dual-written for one migration window).
• Server's Connection model + FSM are already kind-generic (the grafana/prometheus register-only machines are the precedent); a Docker machine slots into getMachine the same way.

Scope / phasing

Phase 0 refactor (days, MeshSync-only, runtime-verified for zero regression). Phase 1 Docker MVP end-to-end (weeks, 4 repos). Cloud = per-provider projects, each >= Docker-sized.

Honest caveats: cloud inventory APIs have no push/watch (poll-diff), and are rate-limited/metered; fingerprinting robustness and any Kanvas Kind-enum assumptions are open questions; standalone (non-embedded) deployment topology is deferred.

Full design

docs/design/fd6-non-kubernetes-discovery.md (added in #576).

[Om-Beast]

Thanks for sharing the design proposal. I like the incremental approach, especially introducing a Discoverer interface first to keep Phase 0 behavior-preserving while enabling future platform-specific implementations.

I'll read through the design document in more detail to better understand the proposed abstraction and migration path. Looking forward to following the implementation as it progresses.