Rewind

An old idea of mine, revived with help of LLM forces. What if we could race around the world with offshore yatchs, but with accelerated time, like x10?

Would we be able to ride on back of storms, and dodge low wind areas? How would it feel to steer through the oceans with the tip of you fingers?

Could we replay historical wind conditions and compare ourselves with real skippers?

Codebase is 90% LLM-generated, but derivated from a hand-written core started 10 years ago. Lacks tests, a nice UI and proper deployment as my primary goal was to get a feel of the game. Paused since.

Features

• Real wind data - Historical GRIB wind forecasts from Vendée Globe 2020
• Multiplayer - WebSocket-based racing with server-authoritative positions
• Realistic physics - IMOCA 60 polar diagrams for boat speed
• 3D globe - Interactive Earth with wind visualization (particles + heatmap)
• Boat controls - Arrow keys to steer, space to tack, up arrow to lock TWA

Development

Create a .env at root from sample.env.

Prerequisites

Install cargo-watch for auto-reload:

cargo install cargo-watch

Start Services

Start Minio:

./server/bin/container up

Start the server:

cd server
./bin/dev-server

Start the client:

cd client
npm install
npm run dev

Ports:

• Client: http://localhost:3000
• Server: http://localhost:3001
• Minio: http://localhost:9000

Container Commands

./server/bin/container up       # Start minio
./server/bin/container down     # Stop containers
./server/bin/container logs     # Follow logs
./server/bin/container migrate  # Run migrations
./server/bin/container destroy  # Remove containers and volumes

Tech Stack

Client:

• React 18 + TypeScript
• Vite
• D3.js for globe projection and zoom/pan
• WebGL for wind texture rendering
• Tailwind CSS

Server:

• Rust with Tokio async runtime
• Axum web framework (HTTP + WebSocket)
• SQLite for wind report inventory
• S3 for wind raster storage

Infrastructure

Requirements

• AWS account with credentials configured
• Terraform
• Fly.io CLI (fly) authenticated

Provisioning

AWS (S3 + CloudFront):

cd infra
terraform init
terraform apply

This creates:

• ACM certificate for rewind.milox.dev
• S3 buckets for GRIB files and wind rasters
• CloudFront distribution for the client

After terraform apply, add DNS records in your DNS provider:

• rewind CNAME → CloudFront domain (from terraform output cloudfront_domain)
• ACM validation CNAMEs (from terraform output acm_validation_records)

Fly.io (Server):

cd server
fly launch --no-deploy

# Create volume for SQLite database
fly volumes create rewind_data --region cdg --size 1

# Set secrets
fly secrets set \
  RUST_LOG=info \
  REWIND_S3_GRIB_BUCKET=rewind-gribs \
  REWIND_S3_RASTER_BUCKET=rewind-wind-rasters \
  REWIND_S3_ENDPOINT=https://s3.eu-west-3.amazonaws.com \
  REWIND_S3_REGION=eu-west-3 \
  REWIND_S3_ACCESS_KEY=<from terraform> \
  REWIND_S3_SECRET_KEY=<from terraform>

The fly.toml is pre-configured to mount the volume at /data with REWIND_DB_PATH=/data/rewind.db.

Deployment

Client

Requires AWS CLI with profile rewind-frontend-uploader configured:

aws configure --profile rewind-frontend-uploader
# Use access key from: terraform state show aws_iam_access_key.frontend_uploader
# Use secret from: terraform output frontend_uploader_secret

Deploy:

./client/bin/deploy

Server

./server/bin/deploy

Scripts

Import GRIB files for courses defined in server:

fly ssh console
rewind import-courses-gribs

Rebuild wind report database from existing S3 PNG files:

fly ssh console
rewind rebuild-manifest