Adaptive Wait

When adaptive_wait = true, Penny adjusts the idle timeout based on recent traffic instead of using a fixed wait_period. Busier apps stay alive longer; idle apps shut down faster.

How It Works

Penny computes request rates over two time windows:

• Short window (5 min) — reacts quickly to traffic bursts
• Long window (30 min) — captures sustained traffic patterns

The higher of the two rates is mapped to a wait period between min_wait_period and max_wait_period using a smooth S-curve (smoothstep). The low_req_per_hour and high_req_per_hour thresholds control where the curve starts and saturates.

Configuration

Minimal Setup

Just enable it — defaults handle the rest:

["myapp.example.com"]
address = "127.0.0.1:3001"
command = "node server.js"
health_check = "/"
adaptive_wait = true

Custom Thresholds

For a high-traffic API with custom thresholds:

["api.example.com"]
address = "127.0.0.1:3002"
command = "python app.py"
health_check = "/health"
adaptive_wait = true
min_wait_period = "5m"
max_wait_period = "60m"
low_req_per_hour = 60
high_req_per_hour = 3000

Options

The Smoothstep Curve

The mapping uses a smoothstep function for smooth transitions between thresholds:

t = clamp((rate - low) / (high - low), 0, 1)
smoothstep = 3t² - 2t³
wait = min_wait + smoothstep × (max_wait - min_wait)

This means:

• Below low_req_per_hour: timeout stays at min_wait_period
• Above high_req_per_hour: timeout stays at max_wait_period
• Between the two: smooth transition with no sharp jumps

Notes

• When adaptive_wait is enabled, the wait_period field is ignored.
• The adaptive wait is recalculated each time Penny schedules a kill timer.