Sync Retry

Minimal, transport-aware retry wrapper. Prefer retry(fn) with defaults in almost all call sites.

Decision Guide

• One-shot Redis/network call: retry(() => call()) — defaults are sufficient.
• Blocking stream loop (queue/topic/ephemeral reader): use attempts: Number.POSITIVE_INFINITY + signal so the loop retries indefinitely until aborted.
• Business/domain error: do NOT wrap with retry. Retry is only for transient transport errors.

Gotchas

• retryIf default (isRetryableTransportError) matches error codes (ECONNRESET, ETIMEDOUT, ECONNREFUSED, ENOTFOUND, EPIPE, EHOSTUNREACH, ECONNABORTED) and message substrings (connection, socket, broken pipe, network, clusterdown, tryagain, loading). Any error not matching is thrown immediately.
• signal.aborted is checked before each attempt — no wasted work after cancellation.
• Library internals (queue/topic/ephemeral stream({ wait: true })) already use retry with unbounded attempts. Do not double-wrap those calls.
• attempts: 1 means no retry — only one try. The default is 8.

API Reference

Read full API/types/defaults in references/api.md.