Keep blue-green deployment strategy but with single minimum machine to reduce costs. During deployment, Fly.io will:
- Start new machine (green)
- Run health checks
- Switch traffic
- Stop old machine (blue)
Note: This may have brief startup time during deployment, but significantly reduces operating costs.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Changes:
- Added [deploy] strategy = "bluegreen" to fly.toml
- Increased min_machines_running from 1 to 2
- Disabled auto_stop_machines to keep both environments ready
- Updated GitHub Actions workflow with --strategy bluegreen flag
How Blue-Green Works:
1. Deploy creates new "green" environment alongside current "blue"
2. Health checks verify green environment is healthy
3. Traffic switches instantly from blue to green
4. Old blue environment kept briefly for instant rollback
5. Zero downtime during deployments
Cost Impact:
- Runs minimum 2 machines instead of 1
- Ensures true zero-downtime deployments
- Instant rollback capability
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Read worker count from WORKERS env var (default: 4)
- Set WORKERS=4 in fly.toml for production
- Allows tuning worker count without code changes
- Can adjust via fly secrets set WORKERS=X
- Upgrade Fly.io VMs from shared to performance CPUs for dedicated compute
- Remove --reload flag from docker-compose for better local performance
- Improves response times and consistency under load
- Cost increase: ~$10-15/month for dedicated CPU performance
Changed Fly.io configuration to maintain one active machine:
- min_machines_running: 0 → 1
- auto_stop_machines: 'stop' → 'suspend'
This eliminates cold start delays by keeping the application warm and ready
to serve requests immediately. The machine will suspend (not stop) when idle,
allowing much faster wake-up times.
With PRELOAD_INTERLINEAR enabled, the 14MB interlinear data stays loaded
in memory, making all requests fast without repeated decompression.
Trade-off: Slightly higher costs for always-on machine, but much better
user experience with instant page loads.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Enable preloading of interlinear data at application startup to eliminate
first-request delays. Configurable via PRELOAD_INTERLINEAR environment variable.
- Add preload_data() function to interlinear_loader.py with logging
- Add startup event handler in server.py to trigger preload
- Enable PRELOAD_INTERLINEAR=true in fly.toml and docker-compose.yml
- Update FLY_DEPLOYMENT.md with cache warming documentation
Performance impact:
- Startup time: ~7-10 seconds (vs ~5 seconds without preload)
- First request: <100ms (vs 2-3 seconds without preload)
- Memory usage: ~400-500MB total (139MB for interlinear data)
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add Dockerfile with multi-stage Python build
- Add GitHub Actions workflow for automatic deployment on main branch
- Add fly.toml with app configuration and auto-scaling settings
- Add .dockerignore to exclude development files from build context
kennethreitz