Fitting a sizing engine inside API Gateway's 29 seconds

API Gateway gives you a hard 29-second integration timeout. For a solar PV and battery-storage sizing engine that is not a lot of room: you have to read the inputs, run the numerical model across thousands of hourly steps, and return a result — all before the gateway hangs up.

flowchart LR
  C[Client] -->|request| AGW["API Gateway · 29s limit"]
  AGW --> L[Lambda]
  L -->|NumPy vectorized| R[Result]
  R --> AGW

The naive version

The first version looped in Python, hour by hour, accumulating state. Correct, readable, and far too slow once the simulation horizon grew. A single request flirted with the timeout, and any retry storm took the whole thing down.

Vectorize the model

The fix was not a bigger Lambda. It was deleting the loop. NumPy lets you express the whole horizon as array operations:

# instead of: for h in range(8760): soc = step(soc, ...)
charge = np.minimum(surplus, max_charge_rate)
soc = np.clip(np.cumsum(charge - discharge), 0, capacity)

The per-hour Python overhead disappears into a handful of C-level array passes. The same calculation that brushed the 29-second ceiling now returns in a couple of seconds, with room to spare for larger installations.

The lesson

The constraint was real and fixed — you cannot negotiate with API Gateway. The degree of freedom was the algorithm. Determinism first, vectorize the hot path, and let the timeout become headroom instead of a wall.