Expand description
Operation-level retry loops for business-layer flows (balance polling, faucet waits, bid submission races, etc.).
§Not to be confused with taskfast_client::retry
The client’s retry is scoped to HTTP transport: it retries only
Server/Network/RateLimited errors and exists to smooth over flaky
endpoints. This one wraps arbitrary async operations and retries on
any error. The policies differ in kind, not degree:
- HTTP retry = “the request didn’t land, try again”
- Agent retry = “the goal wasn’t met yet (e.g., balance still zero), try again — regardless of why”
§RateLimited special-case
If the returned error is Error::RateLimited with a retry_after, the
loop sleeps that exact duration instead of the next exponential step.
This matches the semantic of the platform’s 429 handler: the server told
us when to come back, so we respect it.
Structs§
- Backoff
Options - Knobs for
with_backoff. Defaults tuned for the worker hot loop: 5 attempts over ~15 s of exponential wait (0.5 → 1 → 2 → 4 → 8 s) is usually enough for transient hiccups without wedging an event-handler on a genuine outage.
Functions§
- with_
backoff - Run
opuntil it succeeds ormax_attemptsis exhausted.