trader.daemon.backoff module

Exponential backoff after a failed cycle.

Pure by design: this computes delays and never sleeps. The loop owns sleeping, which is what lets the growth sequence be asserted directly instead of being inferred from elapsed time.

class trader.daemon.backoff.Backoff(*, base_seconds, max_seconds)

Bases: object

Doubling delay with a ceiling, reset by a success.

Parameters:

• base_seconds (int) – the delay after the first failure.

• max_seconds (int) – the ceiling. Must not be below base_seconds — a lower ceiling makes every delay equal to the ceiling, so the delay never grows and the backoff is decorative.

property consecutive_failures: int

How many failures since the last success. For the log line.

next_delay()

Record a failure and return how long to wait before retrying.

The exponent is clamped before the shift rather than after. A daemon that has been failing all day would otherwise compute base * 2**2000 — a perfectly valid Python integer that costs real time and memory to build, only to be thrown away by min.

Return type:

int

reset()

Forget the failure streak. Called after a successful cycle.

Return type:

None