recovery: allow a custom delay strategy (calculateDelay option)

[intech]

The built-in recovery computes reconnect delays in lib/recovery.js with a fixed strategy: an exponential base capped at maxDelay, plus symmetric jitter around the base (offset = Math.random() * jitterPart * 2 - jitterPart). Two consequences for consumers:

• the jitter shape is fixed — full jitter (uniform[0, cap]), decorrelated jitter, or fixed-step schedules are not expressible;
• the cap applies to the base before jitter, so the effective delay can exceed maxDelay by up to the jitter fraction (with jitter: 1, up to 2× maxDelay).

Proposal: accept an optional function in the recovery options, following the precedent of the functional setup option:

await amqplib.connect(url, {
  recovery: {
    // ms; attempt starts at 1; falls back to the built-in strategy when absent
    calculateDelay: (attempt) => number,
  },
});

When absent, behavior is unchanged. normaliseRecoveryOptions currently drops unknown keys, so the addition is backward-compatible.

Context: we maintain an AMQP adapter built on the v2 recovery (Connectum framework). Some strategies can be approximated with the numeric knobs (e.g. full jitter via jitter: 1 plus halved initialDelay/maxDelay), but that is fragile against internal changes and does not cover step schedules or other shapes. A supported hook would remove the need to own a separate reconnect loop just for delay control.

Happy to send a PR if the direction is acceptable.

[cressie176]

Hi @intech

Thank you for taking the time to carefully explain the feature request. I would be happy to review a PR.