ttm-fleet

TTM Fleet — Concurrent Forecasting & Governance

Overview

The fleet endpoint runs one TTM forecast per truck in parallel. All workers are independent — a timeout on one truck does not fail the others.

---

Fleet Configuration

PER_TRUCK_TIMEOUT = 8        # seconds — one TTM native API round-trip
FLEET_WORKERS     = min(10, len(trucks))

Adjust PER_TRUCK_TIMEOUT if native API latency increases (e.g. cross-region calls). Increase FLEET_WORKERS only when fleet size exceeds 10 trucks.

---

Concurrent Forecasting Pattern

from concurrent.futures import ThreadPoolExecutor, TimeoutError

def forecast_fleet_optimization(trucks: list) -> list:
    FLEET_WORKERS     = min(10, len(trucks))
    PER_TRUCK_TIMEOUT = 8

    with ThreadPoolExecutor(max_workers=FLEET_WORKERS) as pool:
        future_map = {
            pool.submit(_forecast_one, t['truckId']): t['truckId']
            for t in trucks
        }

    results = []
    for future, truck_id in future_map.items():
        try:
            results.append(future.result(timeout=PER_TRUCK_TIMEOUT))
        except TimeoutError:
            # Timed-out truck falls back to current-temp-vs-threshold heuristic
            results.append(_fallback_heuristic(truck_id))
        except Exception as e:
            results.append({"truckId": truck_id, "error": str(e)})

    return results

Each worker calls _forecast_one(truck_id) which:

1. Fetches the truck's telemetry
2. Applies cargo normalisation
3. Calls ttm_model.forecast() via Path 1 (native API) or Path 2 (local)
4. Returns a structured result dict