Trait RetryPolicy

pub trait RetryPolicy<E> {
    // Required method
    fn decide(
        &self,
        failure: AttemptFailure<'_, E>,
        proposed: RetryDecision,
        attempt: usize,
    ) -> RetryDecision;
}

Decides what the runner does with a failed attempt.

The runner classifies the failure on its own and passes its verdict as proposed; the policy returns the decision to apply, which is proposed itself unless it wants to override it:

• RetryDecision::Fatal ends the run and returns the original error, the one the closure produced or the commit error, never something the policy made up.
• RetryDecision::Fdb hands that error to fdb_transaction_on_error, which judges retryability and applies the backoff.
• RetryDecision::Retry does the same with code 1020 (not_committed), which is retryable by definition.

A policy cannot corrupt the MaybeCommitted flag handed to the closure: it is computed from the original error before the policy is consulted.

Example: cap the number of attempts

use foundationdb::runner::{AttemptFailure, RetryPolicy};
use foundationdb::{FdbBindingError, RetryDecision};

struct MaxAttempts(usize);

impl RetryPolicy<FdbBindingError> for MaxAttempts {
    fn decide(
        &self,
        _failure: AttemptFailure<'_, FdbBindingError>,
        proposed: RetryDecision,
        attempt: usize,
    ) -> RetryDecision {
        if attempt + 1 >= self.0 {
            RetryDecision::Fatal
        } else {
            proposed
        }
    }
}

Required Methods

fn decide( &self, failure: AttemptFailure<'_, E>, proposed: RetryDecision, attempt: usize, ) -> RetryDecision

Returns the decision to apply for failure, proposed being what the runner would do on its own.

Implementors

impl<E> RetryPolicy<E> for NativeRetryPolicy