adaptive-timeout

Adaptive timeout calculation with EWMA-based tracking and fallback strategies.

Install

npm install adaptive-timeout

Usage

const AdaptiveTimeout = require('adaptive-timeout')

const timeouts = new AdaptiveTimeout()

// Record a sample when you get a response
timeouts.put('10.0.0.1:8080', 150)

// Get timeout for next request (attempt 1, 2, 3...)
timeouts.get('10.0.0.1:8080', 1) // adaptive based on recorded samples
timeouts.get('unknown-peer', 1) // uses fallback strategy

API

const timeouts = new AdaptiveTimeout([options])

Create a new instance.

Options:

• maxSize - max cache entries (default: 65536)
• maxAge - entry TTL in ms (default: 600000 / 10 minutes)
• fallback - timeout sequence for unknown keys (default: [500, 750, 1000, 1500, 2000])
• min - minimum timeout for known keys (default: 300)
• max - maximum timeout for known keys (default: 5000)
• jitter - random jitter range in ms (default: 256)

timeouts.put(key, value)

Record a sample. Updates the exponentially weighted moving average.

Returns { avg, variance }.

timeouts.get(key, [attempt])

Get a timeout value.

For known keys: calculates adaptive timeout with linear backoff based on attempt number.

For unknown keys: returns value from fallback sequence based on attempt number.

timeouts.getValue(key)

Get raw stats for a key. Returns { avg, variance } or null.

timeouts.has(key)

Check if a key exists.

timeouts.delete(key)

Remove a key.

timeouts.clear()

Remove all entries.

timeouts.size

Number of entries.

How it works

Uses TCP-style EWMA (exponentially weighted moving average) to track values:

• avg blends 87.5% old + 12.5% new sample
• variance blends 75% old + 25% new deviation

Timeout for known keys: (avg + 2 * variance) * attempt + jitter

Unknown keys use an aggressive fallback sequence to probe quickly.

License

Apache 2.0