Instruction file imported from rk8s-dev/rk8s (.github/instructions/rks.instructions.md). Copyright stays with the author.

RKS (Control Plane) – Path-specific Copilot Instructions

Scope

RKS provides the scheduler, state management, and reconciliation loops, aligning with Kubernetes control-plane patterns.
Priorities: consistency, recoverability, observability.

Coding & design

• Implement controllers as explicit state machines (Pending → Running → Succeeded/Failed) — idempotent and re-entrant.
• Storage/state backend should be pluggable with WAL/snapshot or event replay.
• Scheduler: start simple (binpack/spread), keep extensible for affinity, taints, and priorities.
• API layer (HTTP/gRPC): versioned, supports watch/streaming; avoid leaking internal structs.
• Trace critical decisions (node binding, preemption, rollback) via tracing.

Tests & benchmarks

• Property tests for controller logic (repeated/async events).
• Recovery tests after restart (snapshot or event replay).
• Scheduler load tests: large batch submission latency and failure injection (resource exhaustion, node loss).

Copilot guidance

• Suggest Rust code for controller skeletons, event types, and store traits.
• Emphasize re-entrancy, idempotency, and clear separation of desired vs actual state.