flexinfer-resilience

Production-ready resilience patterns for Python applications.

Features

• Circuit Breaker - Prevents cascading failures by failing fast when services are down
• Rate Limiter - Token bucket rate limiting with exponential backoff
• Retry - Automatic retries with configurable backoff strategies (constant, linear, exponential, decorrelated jitter)
• Bulkhead - Concurrency isolation using semaphores to prevent resource overload
• Timeout - Time-bounded operations with fallback support
• Async-first - Built for asyncio with full async/await support
• Type hints - Complete type annotations for IDE support
• Zero dependencies - Uses only Python standard library (optional Redis support)
• Distributed State - RedisCircuitBreaker for sharing circuit state across multiple instances.

Installation

pip install flexinfer-resilience

Import path stays resilience:

from resilience import CircuitBreaker

Or install from GitLab:

pip install git+https://gitlab.flexinfer.ai/libs/py-resilience.git

Quick Start

Circuit Breaker

from resilience import CircuitBreaker, CircuitBreakerConfig

# Create a circuit breaker with custom config
config = CircuitBreakerConfig(
    failure_threshold=5,      # Open after 5 failures
    recovery_timeout=30.0,    # Wait 30s before half-open
    success_threshold=2,      # Close after 2 successes in half-open
)
breaker = CircuitBreaker("my-api", config)

# Use as context manager
async def call_api():
    async with breaker:
        return await external_api.fetch()

# Or use execute method
result = await breaker.execute(external_api.fetch)

# Or use as decorator
@breaker.decorate
async def decorated_call():
    return await external_api.fetch()

Rate Limiter

from resilience import RateLimiter, RateLimitConfig

# Create a rate limiter
config = RateLimitConfig(
    requests_per_second=10,   # 10 requests per second
    burst_size=20,            # Allow bursts up to 20
)
limiter = RateLimiter(config)

# Acquire a token before making requests
async def rate_limited_call():
    await limiter.acquire()
    return await api.request()

# Or check without blocking
if await limiter.try_acquire():
    await api.request()
else:
    print("Rate limited, try again later")

Combined Usage with Retry

from resilience import (
    CircuitBreaker,
    RateLimiter,
    execute_with_retry,
    CircuitBreakerOpenError,
)

limiter = RateLimiter(RateLimitConfig(requests_per_second=5))

# Execute with automatic retry and rate limiting
result = await execute_with_retry(
    fetch_data,
    url="https://api.example.com",
    rate_limiter=limiter,
    max_retries=3,
    retry_exceptions=(TimeoutError, ConnectionError),
)

API Reference

CircuitBreaker

class CircuitBreaker:
    def __init__(self, name: str, config: CircuitBreakerConfig | None = None): ...

    @property
    def state(self) -> CircuitState: ...  # CLOSED, OPEN, or HALF_OPEN

    async def execute(self, func, *args, **kwargs) -> T: ...
    async def __aenter__(self) -> CircuitBreaker: ...
    async def __aexit__(self, ...): ...
    def decorate(self, func) -> func: ...
    def reset(self) -> None: ...

CircuitBreakerConfig

RateLimiter

class RateLimiter:
    def __init__(self, config: RateLimitConfig | None = None): ...

    async def acquire(self, tokens: int = 1) -> float: ...  # Returns wait time
    async def try_acquire(self, tokens: int = 1) -> bool: ...
    def backoff_delay(self, attempt: int) -> float: ...
    def reset(self) -> None: ...

RateLimitConfig

Registry Pattern

For managing multiple circuit breakers:

from resilience import CircuitBreakerRegistry

# Get or create circuit breakers by name (singleton)
registry = CircuitBreakerRegistry()
breaker = registry.get("api-service")

# Pre-configure breakers
registry.configure("api-service", CircuitBreakerConfig(failure_threshold=3))

# Get stats for all breakers
stats = registry.get_stats()

Circuit Breaker States

     ┌─────────────────────────────────────────┐
     │                                         │
     ▼                                         │
┌─────────┐  failure_threshold   ┌──────────┐  │
│ CLOSED  │ ──────────────────▶  │   OPEN   │  │
└─────────┘                      └──────────┘  │
     ▲                                │        │
     │                                │ recovery_timeout
     │ success_threshold              ▼        │
     │                          ┌───────────┐  │
     └───────────────────────── │ HALF_OPEN │──┘
                                └───────────┘ failure

• CLOSED: Normal operation, requests pass through
• OPEN: Service failing, requests rejected immediately
• HALF_OPEN: Testing recovery with limited requests

License

MIT License - see LICENSE