Retry budget
When a dependency goes down, every client retries — and the load multiplies exactly when the system is weakest. This calculator shows the worst case of your retry policy before an incident does: how much extra load you generate and how long a client takes to give up.
How to read it
With 3 attempts, 100 ms ×2 backoff and 100 clients: the failing dependency receives 300 requests instead of 100 (×3 amplification) and each client takes up to ~6.5 s to give up. If your caller has a 5 s timeout, your retry policy no longer fits inside it.
Assumptions and limits
- Assumes the dependency is fully down (every attempt fails): worst case, not average.
- Jitter is modeled as a ceiling (+%) for worst-case time; in practice it spreads load, it does not reduce it.
- Does not model circuit breakers or queues: if you have them, your real worst case is better than this.
- Amplification is per client; with multiple retrying layers, it multiplies per layer.
The article behind it
Timeout, Retry and Circuit Breaker: the order matters · Circuit Breaker: production implementation
Version 1.0 · July 2026 · Runs 100% in your browser: nothing is sent to any server.
Want to know when the next tool lands? Follow the newsletter.