execution_policy 0.3.0

A Dart resilience and transient-fault-handling library (retry, backoff, timeout, circuit breaker, fallback) inspired by C# Polly.

execution_policy

A Dart resilience and transient-fault-handling library inspired by C# Polly, with built-in hooks for APM error logging.

---

Features

• Retry with fixed, linear, exponential, and jitter-ed backoff
• Timeout to fail long-running operations (per attempt)
• Fallback to recover gracefully — catches Exceptions only, so bugs stay visible
• Circuit Breaker to stop hammering failing services (single-probe half-open)
• Cancellation via CancellationToken — aborts further attempts and a sleeping backoff
• Fluent PolicyBuilder<T> API for composition
• PolicyDebugger for tracing start/success/failure of each policy
• Observability hooks: onError, onRetry, and breaker onStateChange

---

Installation

Add to your pubspec.yaml:

dependencies:
  execution_policy: ^0.3.0

Quick Start

Fluent Builder

Build one long-lived builder per endpoint and reuse it, so the circuit breaker (and any cancel token) keep their state across calls:

final _policy = PolicyBuilder<String>()
    .retry(
      RetryOptions.exponentialJitter.copyWith(maxAttempts: 4),
      retryIf: (error) => error is HttpException,
      onRetry: (error, attempt, delay) =>
          log('retry #$attempt in ${delay.inMilliseconds}ms'),
      onError: (error, stack, attempt) async {
        // Awaited defensively — a throwing hook can't derail the retry loop.
        // e.g. FirebaseCrashlytics.instance.recordError(error, stack);
      },
    )
    .timeout(const Duration(seconds: 2)) // per-attempt deadline
    .circuitBreaker(
      failureThreshold: 3,
      resetTimeout: const Duration(seconds: 10),
      onStateChange: (from, to) => log('circuit $from -> $to'),
    )
    .fallback((error) async => 'default value'); // receives the error

Future<String> fetchWithResilience() =>
    _policy.execute(() => httpClient.get('https://example.com').then((r) => r.body));

Composition order

Policies are applied by a fixed order regardless of the order you add them:

Fallback (outermost) → CircuitBreaker → Retry → Timeout (innermost).

Two consequences are intentional:

• the timeout is per attempt — each retry gets a fresh deadline (not a whole-execution budget); and
• the breaker sits outside retry — a single execute can record up to maxAttempts failures against it, so size failureThreshold accordingly.

Cancellation

retryIf/shouldContinue stop future attempts; a CancellationToken also interrupts a backoff that is already sleeping:

final token = CancellationToken();
final future = _policy.copy() // copy shares the breaker; adds a per-call token
    .reset()
    .retry(RetryOptions.exponentialJitter, cancelToken: token)
    .execute(fetch);

// later, e.g. in a widget's dispose():
token.cancel(); // aborts the in-flight backoff and stops retrying

Debug Execution

final result = await _policy.debugExecute(
  () async => unreliableOperation(),
  developer.log, // any void Function(String)
);

API Reference

• PolicyBuilder<T>

  • .retry(RetryOptions options, {retryIf, shouldContinue, cancelToken, onError, onRetry})
  • .timeout(Duration duration)
  • .circuitBreaker({failureThreshold, resetTimeout, onStateChange})
  • .fallback(FallbackFunction<T> fallbackFn, {shouldHandle})
  • .execute(FutureFunction<T> action) / .debugExecute(action, logger)
  • .reset() / .copy() (copy shares stateful policy instances)
  • Adding two policies of the same type throws StateError.

• RetryOptions — presets fixed, linear, exponential, exponentialJitter; .copyWith(...); .delayFor(int attempt) -> Duration (overflow-safe, capped by maxDelay).

• CancellationToken — cancel(), isCancelled, whenCancelled.

• CircuitBreakerPolicy<T> — state, failureCount; throws CircuitOpenException(retryAfter) when open.

• TimeoutPolicy<T>, FallbackPolicy<T>, PolicyDebugger<T>

Contributing

1. Fork the repo
2. Create a branch (git checkout -b feature/foo)
3. Commit your changes (git commit -am 'Add foo')
4. Push (git push origin feature/foo)
5. Open a PR

License

This project is licensed under the MIT License. See LICENSE for details.