Summary
Replace MeshSync's flat, all-or-nothing discovery (one informer per kind over a hardcoded ~66-resource list, all namespaces, full fidelity) with a three-tier funnel that is user-configurable per connection for depth and scope:
- Tier 1 cheap/broad presence detection, Tier 2 deeper per-object inspection, Tier 3 infrastructure-specific deep discovery activated only when that infrastructure is detected present (optionally delegated to adapters over NATS).
- Depth is a per-kind projection (shallow = ObjectMeta only; full = today's behavior), trimmed at publish time - no new wire shape.
Key decisions
- Whitelist/blacklist becomes a degenerate tier config (full back-compat).
- Schema-first: a typed tier/depth contract surfaced via the MeshSync CRD (meshery-operator) plus a per-connection override in Connection metadata (meshery/schemas), with a Server API + UI controls.
- Fixes a real operability gap found along the way: editing the MeshSync CR's
watch-list currently requires a pod restart (no live reload). This adds a CR-watch hot-reload path - independently valuable and worth extracting as its own change.
Scope / phasing
MVP: depth toggle + tier config parsing + CR hot-reload (~1.5-2 wks across repos). Full: infra detection + conditional Tier-3 + adapter delegation.
Scale note: depth-trimming cuts publish/DB cost but not informer cache/watch memory; true cluster-scale relief needs metadata-only or namespace-scoped informers (separate).
Full design
docs/design/fd1-tiered-configurable-discovery.md (added in #576) - current-state analysis, per-repo file changes, schema/CRD/API changes, sequencing, test plan, and open questions.
Summary
Replace MeshSync's flat, all-or-nothing discovery (one informer per kind over a hardcoded ~66-resource list, all namespaces, full fidelity) with a three-tier funnel that is user-configurable per connection for depth and scope:
Key decisions
watch-listcurrently requires a pod restart (no live reload). This adds a CR-watch hot-reload path - independently valuable and worth extracting as its own change.Scope / phasing
MVP: depth toggle + tier config parsing + CR hot-reload (~1.5-2 wks across repos). Full: infra detection + conditional Tier-3 + adapter delegation.
Scale note: depth-trimming cuts publish/DB cost but not informer cache/watch memory; true cluster-scale relief needs metadata-only or namespace-scoped informers (separate).
Full design
docs/design/fd1-tiered-configurable-discovery.md (added in #576) - current-state analysis, per-repo file changes, schema/CRD/API changes, sequencing, test plan, and open questions.