Limit X — Server Stack Deployment

Server deployment 🖥️ host x86-64

The host (Server2, 192.168.0.83) runs two containers — the docs site and the dash (UI + relay). No database or backend services in the telemetry pipeline itself — anything else on the box (e.g. a SQL server on :1433) is separate infrastructure, not part of this stack.

Before you start

1. Two machines on the same L2 network — the host and the robot (Jetson AGX Thor). No internet needed at runtime.
2. Docker Engine installed and sudo-free for the deploy user on both machines.
3. Images: production-limit-docs-docker and production-limit-dash-docker (both x86-64) on the host; production-limit-robot-docker (aarch64) on the Jetson. All three are untagged — no :1.0 tag exists.
4. Ports free on the host: 3010 (docs, published — on this fleet :3000 is taken by Gitea), 80 (dash UI, via --net host), 9000/udp (telemetry in from the robot), 8000 (relay API, localhost only).

Air-gapped image transfer

# 1. on the machine that BUILT the image
docker save production-limit-dash-docker -o limit-dash.tar
# 2. copy limit-dash.tar to the target (USB stick, sftp, whatever)
# 3. on the TARGET machine
docker load -i limit-dash.tar
docker images | grep production-limit   # confirm names + arch before you run

Docs site (this site, port 3010)

docker run -d \
  --name limit-docs \
  --restart unless-stopped \
  -p 3010:3000 \
  production-limit-docs-docker

• Static site — the nginx container (Alpine) is the whole thing. No database, no backend; nginx listens on :3000 inside the container and is published to 3010 on this fleet because :3000 is used by Gitea on the host. Any free port works — -p 3000:3000 is fine elsewhere.
• Pages: index, robot, robot_model, drive, robot_deploy, and server_deploy — this one — plus the bundled offline ZED 2i SDK reference under /offline_docs/.

Limit Dash (host port 80)

docker run -d \
  --name limit-dash \
  --restart unless-stopped \
  --net host \
  production-limit-dash-docker

• One container, three parts: nginx serves the UI on host port :80 and proxies /api/ to the C++ relay on :8000 (SSE, buffering off, 60 s read timeout); the relay ingests UDP JSON on :9000/udp. --net host is required — the relay binds the machine's own :9000/:8000 directly.
• Relay endpoints: GET /api/snapshot, GET /api/stream (SSE), GET /api/health.
• Live tiles fed by the robot stream: T04 lidar, T05L/T05R front cameras (/limitx/cam2), T06L/T06R rear cameras (/limitx/cam1), T08 IMU WIP, T09 GNSS, T12 source health, T28 system. 4 status LEDs (ENGINE, DOG LINK, MUSIC, DASH) and a SIM mode for no-robot testing. The remaining tiles (joints, battery, terrain, …) read NO SOURCE until their producers exist.

Verify

1. curl http://192.168.0.83/ — dashboard HTML.
2. curl http://192.168.0.83/api/health — relay reports alive.
3. curl -I http://192.168.0.83:3010/ — the docs site (this one) answers 200.
4. Open http://192.168.0.83 — tiles are grey/stale until the robot starts; flip SIM and they light up with synthetic data, proving the pipeline.

Day-to-day