Retry Policies & Backoff

Why Smart Retries Matter

Naively retrying failed requests at fixed intervals (e.g. every 5 seconds) causes the Thundering Herd Problem: if an API recovers after an outage, hundreds of queued retries hit it simultaneously, driving it back into a crash loop.

Bunnylynk uses mathematical backoff models with randomized jitter to space retries smoothly over time.

Backoff Algorithms

1. Full Jitter Exponential Backoff (Default)

The delay before attempt $n$ is calculated as:

$$ \text = \text(0, \min(M, B \times 2^n)) $$

Where:

• $B$ is the initial base delay (default 5 seconds).
• $M$ is the maximum backoff cap (default 3600 seconds / 1 hour).
• $n$ is the zero-indexed retry count.
• $\text(0, x)$ prevents synchronized requests across multiple failing connections.

2. Linear Backoff

For applications requiring predictable intervals:

$$ \text = \min(M, B \times n) $$

Example Retry Schedule

For a connection with base_delay = 10s, max_delay = 3600s, and max_retries = 8:

Manual Replay

At any time, you can trigger a manual replay for an event using the API:

curl -X POST https://api.bunnylynk.com/v1/events/evt_01JMWP3KV9QZ/replay \
  -H "Authorization: Bearer $BUNNYLYNK_API_KEY"

Next: Learn how Circuit Breakers protect your services.