electric-cherry

Web app that manages a fleet of T5 Plutonium servers running under Komodo. Successor to push_scripts.sh — same job (push shared GSC scripts, run beta tests, restart empty servers, track deployments) with a live UI and real audit diffs.

Stack

SvelteKit (adapter-node) + TypeScript
SQLite via better-sqlite3 (replaces script_metadata.json as the source of truth)
Server-sent events for live pipeline / server-status streams
RCON (UDP) direct-to-game-server for player counts
Komodo API for restarts and per-server game-port discovery
Runs as a single Komodo-managed container on the same host as the game servers, with /root/zombies bind-mounted in

Layout

src/lib/server/db/ — schema, migrations, better-sqlite3 handle
src/lib/server/fleet/ — file walking, push/beta engines
src/lib/server/integrations/ — Komodo, RCON
src/lib/server/pipeline/ — sweep loop, event bus, worker singleton
src/routes/api/ — REST + SSE endpoints
src/routes/ — UI pages (dashboard, deployments, betas, per-server)
Dockerfile, docker-compose.yml, entrypoint.sh — container build
komodo/ — Komodo stack definition
scripts/ — legacy JSON importer, fixture seeder
tests/ — vitest unit tests

How restarts stay safe

File changes are staged sequentially. Their activation uses rolling restarts: by default only one server across the fleet can restart or recover at a time. Promotions, rollbacks, drift reconciliation, and beta restarts share this limit. Set DEPLOYMENT_RESTART_CONCURRENCY to a positive integer to raise it.

A background sweep runs every 15s. It checks recovering servers first, then admits queued deployments in oldest-first order when there is capacity. Busy or unreachable servers stay queued without taking a slot, so other empty servers can proceed. A successful Komodo restart request retains its slot until the container is healthy and RCON answers. Health checks start no sooner than 15 seconds after the request returns, avoiding an immediate check of the old process. An unhealthy server keeps its slot rather than letting failures spread across the fleet. A timed-out restart also keeps its slot because Komodo may have accepted the request.

Slots persist in SQLite across app restarts, and overlapping timer/API triggers share the active run. Beta options to skip health or match gates still work; skipping a beta gate does not release the fleet's recovery slot early. New-server creation is separate from deployment restarts and still starts its container directly.

No token, no session, no auth freeze. If Komodo or RCON aren't configured, the sweep marks pending rows manual so nothing hangs forever.

Local dev

Windows or macOS — no /root/zombies on disk, so seed a fixture tree first.

npm install
npm run seed:fixture

# Point env at the fixture (PowerShell)
$env:SHARED_SCRIPTS = "$PWD/test-fixtures/zombies/shared/scripts"
$env:SERVERS_ROOT   = "$PWD/test-fixtures/zombies/t5"
$env:DB_PATH        = "$PWD/data/cherry.db"

npm run dev
# open http://localhost:3000

Without RCON_PASSWORD / KOMODO_* set, the pipeline marks pending rows manual after a sweep — the UI still works, you just don't get real restarts. Enough to develop the push/beta flow.

Map selection and configs

The map selected when deploying a T5 zombies server controls its startup rotation automatically. No map-choice config overlay is required. New servers use the promoted game baseline and matching overlays when available; otherwise Plutainer seeds its usual defaults and Electric Cherry applies the selected map through a generated configs/electric_cherry_map.cfg, loaded after the main config and before map rotation. This file is tracked and restored by drift reconciliation; operators do not need to create or promote it. The launch order uses Plutainer's extra arguments and map-rotation controls. Plutainer's default main config is preserved until a game baseline is promoted.

Beta previews, promotion, and config drift repair all apply the server's stored catalog map after composing the baseline and overlays. An old sv_maprotation overlay cannot override that selection, but its other settings still apply. Existing servers receive this behavior on their next config deployment or drift reconciliation; this change does not immediately rewrite or restart them. Servers with an unknown map retain their existing config behavior. Rollback restores the exact prior config bytes.

Round telemetry experiment

Send POST /api/telemetry with Content-Type: application/json to the Electric Cherry host (port 3000 in local development):

curl http://localhost:3000/api/telemetry \
  -H 'Content-Type: application/json' \
  -d '{"event":"round_change","round":17,"server":"t5_zombie_prototype_1","players":4}'

The server must already be registered in Electric Cherry. round and players are required nonnegative integers. Success returns HTTP 200 with { "ok": true, "telemetry": { "server": "...", "round": 17, "players": 4, "receivedAt": "..." } }. Invalid payloads return 400, unknown/destroyed servers 404, and non-JSON content types 415. This endpoint uses the app's existing LAN/Tailscale access model; there is no API key requirement.

The latest report is saved across app restarts and streamed immediately to the dashboard and server detail page, including after browser reconnection. The UI labels it as the last reported round with its age: round-change events alone cannot detect a stopped script or a match ending. Send the initial round on each new match, then each round change. The latest received request wins (including lower rounds for new matches); there is no event ordering or history yet. Telemetry player counts are displayed with the report and never replace RCON occupancy or affect automated restart gates.

Running tests

npm test

Covers push (add/update/skip), beta lifecycle (open → rollback, open → promote, conflict blocks), sweep (integrations-off manual branch), and RCON parsing (fake UDP socket).

Deploy to the Komodo host

Build the image once locally or on the host:
```
docker compose build
```
Import komodo/stack.toml into Komodo (or point Komodo at the docker-compose.yml at repo root).

Set the env vars in Komodo:

KOMODO_ADDRESS, KOMODO_API_KEY, KOMODO_API_SECRET
RCON_PASSWORD              # single shared password across every T5 server
RCON_HOST                  # optional; defaults to host.docker.internal

First boot triggers the legacy import: if /root/zombies/script_metadata.json exists and /data/cherry.db is empty, entrypoint.sh runs scripts/import-legacy.ts before starting the server. The source JSON gets renamed to .migrated.
Load the app on the LAN or over tailscale (no web-app auth by design).

What still lives in the shell script

Nothing that's needed at runtime. Move push_scripts.sh to archive/ once you've run a side-by-side parity check between the shell's dry-run and the app's POST /api/push { dryRun: true }.

Non-goals

Editing GSC files in-browser. Do that in your editor of choice; the web app only cares about what's on disk in SHARED_SCRIPTS.
Multi-user auth. LAN/tailscale scoping is the security model.
Two-way sync with script_metadata.json. The importer runs once. If the shell script is ever used again after that, it'll write to the JSON file and the web app won't see it.

References into the old script

Every port carries a // Ports push_scripts.sh:NN comment pointing back to the original for the parts that had non-obvious behavior. Grep those if a behavior seems wrong — the shell script is the reference implementation until the parity check passes.