Distributed Systems Model

Distributed Systems Model

Audience: Engine Developer Status: ✍️ Draft

The consistency and scaling philosophy that ties the engine together: single-writer per zone, dynamic placement, leases and fencing, and the exactly-once substrate. This page states the invariants the rest of the distributed design depends on.

Intended contents

• The core invariant: single-writer per zone — no in-zone locks, deterministic order, the unit of consistency
• Placement model: world servers claim zones via leases; director-coordinated; no static declaration
• Consistency model: what's strongly consistent (a zone's state) vs. eventually consistent (cross-scope views); where Postgres is authority
• Fencing and exactly-once: state_version/DurableEvent.Seq watermarks, lease fence tokens, idempotency
• Failure model: shard loss, director loss, dependency (PG/Redis/NATS) partition; fail-closed defaults (pack-set divergence, keyless boot)
• Scaling axes and observed capacity; how handoff enables zero-drop scale in/out
• Mermaid: a state diagram of zone ownership (unclaimed→claimed→draining→handed-off)
• Cross-links: Zone Runtime & Actor Model, Cross-Shard Handoff, Orchestration & Directors, Running at Scale

---

Scaffold page — full content pending the wiki build-out.