retries

Retry policies.

Terminology: an attempt is one execution of a task. The first execution is attempt 1; every later attempt is a retry. retries=3 therefore allows up to four attempts in total.

class RetryCopy Anchor Link

Raise from inside a task to request another attempt.

delay overrides the retry policy's computed backoff (seconds). The request still counts toward the task's maximum attempts, so explicit retries cannot loop forever.

__init__(delay: float | None = None, reason: str | None = None) -> NoneCopy Anchor Link

class RetryPolicyCopy Anchor Link

Exponential backoff with optional jitter.

The delay before retry n (n = 1 for the first retry) is min(max_delay, initial_delay * backoff ** (n - 1)). With jitter enabled the delay is drawn uniformly from [delay / 2, delay] ("equal jitter"), which spreads retries of simultaneously failing tasks while keeping a guaranteed minimum wait.

retry_on lists exception types that trigger an automatic retry; dont_retry_on takes precedence and marks exceptions as permanent failures. An explicit :class:~blitzq.Retry raised by the task is always honoured while attempts remain.

__init__(initial_delay: float = 1.0, max_delay: float = 300.0, backoff: float = 2.0, jitter: bool = True, retry_on: tuple[type[BaseException], ...] = (Exception,), dont_retry_on: tuple[type[BaseException], ...] = ()) -> NoneCopy Anchor Link

compute_delay(retry_number: int, rng: random.Random | None = None) -> floatCopy Anchor Link

Delay in seconds before retry number retry_number (1-based).

is_retryable(exc: BaseException) -> boolCopy Anchor Link

class TaskTimeoutCopy Anchor Link

A task exceeded its execution time limit.

For async tasks the coroutine is cancelled. For thread and process executors the worker stops waiting, but the underlying call keeps running until it returns (Python cannot safely kill a thread).