Phase6

docs/IMPLEMENTATION_STATUS.md

@@ -2,51 +2,57 @@
 
 Last updated: 2026-06-26
 
-## Current phase: Phase 5 — Software & add-ons ✅
+## Current phase: Phase 6 — Multi-Node and queue ✅
 
-### Phase 5 completed
+### Phase 6 completed
 
 | Area | Status | Notes |
 |------|--------|-------|
-| Prisma InstalledAddon | Done | Plugin/mod install tracking |
-| `@hexahost/catalog` | Done | Modrinth API + compatibility checks |
-| `@hexahost/contracts` phase 5 | Done | catalog, addons schemas |
-| Go agent v0.5 | Done | addon.install/remove with SHA512 verify |
-| Worker server-addons queue | Done | install-addon, remove-addon jobs |
-| API CatalogModule | Done | Modrinth search + versions |
-| API AddonsModule | Done | Install/remove with audit |
-| Web add-ons tab | Done | Search, install, list, remove |
-| Docker FABRIC/PURPUR | Done | Extended TYPE mapping for itzg image |
+| `@hexahost/scheduler` | Done | Node scoring, RAM/slot capacity, least-loaded placement |
+| Prisma start queue + node capacity | Done | `ServerStartQueueEntry`, `maxRamMb`, `platformRamMb`, `QUEUED`/`UNKNOWN` |
+| `@hexahost/contracts` phase 6 | Done | `nodes.ts`, queue schemas, extended server status |
+| Worker scheduler integration | Done | Provisioning uses `selectNodeForAllocation` |
+| Start queue | Done | API enqueue + worker tick dispatch |
+| Node health tick | Done | Heartbeat timeout → `UNREACHABLE`, servers → `UNKNOWN` |
+| Drain / maintenance | Done | Admin `POST /admin/nodes/:id/drain` and `/maintenance` |
+| Allocation release | Done | Released on provision failure |
+| Dev seed | Done | Second node `local-dev-2` for multi-node tests |
 
-### Abnahmekriterium Phase 5
+### Abnahmekriterium Phase 6
 
-- [x] Modrinth catalog search with loader + game version facets
-- [x] Compatibility check before install
-- [x] Paper plugin → `plugins/` directory
-- [x] Fabric mod → `mods/` directory
-- [x] SHA512 verification on agent
-- [x] Install/remove only when server STOPPED
-- [ ] Manual E2E (Paper + Fabric server with agent)
+- [x] Scheduler distributes servers across eligible nodes (least loaded RAM)
+- [x] Resource reservation via `GameServerAllocation` + node RAM budget
+- [x] Start queue when node lacks capacity
+- [x] Drain excludes node from new allocations
+- [x] Maintenance mode blocks scheduling
+- [x] Node heartbeat loss marks servers `UNKNOWN`
+- [ ] Manual E2E with two agents on different nodes
 
-### Local test
+### Local test (two nodes)
 
-1. Create **Paper** server → stop → **Add-ons** → search plugin (e.g. LuckPerms) → install
-2. Create **Fabric** server → stop → install compatible mod
-3. Start server and verify JAR in container
+1. `pnpm db:seed` — note both enrollment tokens
+2. Start agent 1 with `NODE_ID=...001` and agent 2 with `NODE_ID=...002`
+3. Create several servers and start them — verify different `nodeId` in API responses
+4. Saturate one node RAM (`maxRamMb` low) → next server lands on other node
+5. `POST /admin/nodes/:id/drain` — new provisions skip drained node
 
 ### Known limitations
 
-- Modpacks/datapacks UI deferred
-- No automatic update/rollback yet (metadata stored for future)
-- CurseForge not integrated
-- Server must be reprovisioned to change software family (existing flow)
+- Offline migration not implemented (manual re-provision only)
+- No admin UI for nodes yet (API only)
+- Queue priority uses plan metadata placeholder (`priority=0`)
+- CPU overcommit not configurable yet
+
+---
+
+## Phase 5 — Software & add-ons ✅
+
+Modrinth catalog, addon install/remove, SHA512 verify.
 
 ---
 
 ## Phase 4 — Worlds & backups ✅
 
-S3 backups, world upload/download, restore, retention, schedules.
-
 ---
 
 ## Phase 3 — Console, files, settings ✅
@@ -63,6 +69,6 @@ S3 backups, world upload/download, restore, retention, schedules.
 
 ## Phase 0 — Monorepo foundation ✅
 
-### Next phase: Phase 6 — Multi-Node and queue
+### Next phase: Phase 7 — Idle shutdown and free tier
 
-- Scheduler, resource reservation, start queue, drain
+- Player detection, auto-stop, quotas, credits, usage metering