Module runner

The retry runner behind Database::run.

The runner is built from three independent pieces:

• the attempt: run the closure on a transaction, then commit it. One attempt does exactly one pass and never retries.
• the hooks (RunnerHooks): purely observational callbacks fired around each attempt. They cannot change what the runner decides.
• the policy (RetryPolicy): the only thing that can change a decision. It sees the failure and the decision the runner proposes, and returns the decision to apply.

They are assembled by TransactionRunner, the builder returned by Database::runner:

let metrics = TransactionMetrics::new();
db.runner()
    .hooks(&MetricsHooks::new(&metrics))
    .run(|trx, _maybe_committed| async move {
        trx.set(b"key", b"value");
        Ok::<_, FdbBindingError>(())
    })
    .await?;
let report = metrics.get_metrics_data();

Retry semantics

fdb_transaction_on_error remains the single retry governor: backoff, max retry delay and TransactionOption::RetryLimit are applied by the C API, never by a Rust-side budget. A closure error asking for a retry without a native error underneath (RetryDecision::Retry) is routed through on_error with code 1020 (not_committed) so that it obeys the same budget.

Structs

• MetricsHooks
  Hooks collecting a MetricsReport into a TransactionMetrics.
• NativeRetryPolicy
  The default policy: it always applies what the runner proposes, leaving fdb_transaction_on_error the last word on retries.
• TransactionRunner
  Builder for a transactional run, returned by Database::runner.

Enums

• AttemptFailure
  The failure a RetryPolicy is asked about.

Traits

• RetryPolicy
  Decides what the runner does with a failed attempt.
• RunnerHooks
  Observation points of the retry runner.