Optimizer Service

The Optimizer consumes enriched market snapshots from Redis streams and runs Optuna-based walk-forward optimization. Optimization runs only when triggered by a command (via the Gateway API or frontend UI). All data lives in Redis — no disk I/O.

How It Works

1. Data accumulation — Feed publishes to collector:{token}. Optimizer buffers via XREADGROUP.
2. Manual trigger — User clicks "Run Optimizer" or calls POST /api/optimizer/run. Gateway pushes command to optimizer:commands.
3. Optimization — WalkForwardOptimizer runs Optuna trials over snapshot-replay backtests using real orderbook data.
4. Output — Overrides written to state:overrides:{env} for the Signals service. Full run persisted to optimizer:run:{run_id}.

Walk-Forward Windows

Window granularity: day, hour, or 15min. Smaller units capture intra-day regime shifts. Optional auto_window_sizing runs a nested Optuna meta-study to find optimal window configuration.

Modes

• standalone — Single process: buffers, reads commands, runs optimizations
• coordinator — Buffers and dispatches jobs to remote optimizer workers
• optimizer — Worker that polls its job queue and runs optimizations (horizontally scalable)

Self-Healing

Keeps a local cache of the last uploaded payload and periodically re-uploads if the Redis key is missing (e.g., after a Redis restart).