Resilience Patterns

Build systems that survive partial failures and degrade gracefully.

Core Patterns

Retry with Exponential Backoff

import asyncio
import random
from functools import wraps

def retry(max_attempts: int = 3, base_delay: float = 1.0, max_delay: float = 30.0):
    def decorator(func):
        @wraps(func)
        async def wrapper(*args, **kwargs):
            for attempt in range(max_attempts):
                try:
                    return await func(*args, **kwargs)
                except Exception as e:
                    if attempt == max_attempts - 1:
                        raise
                    delay = min(base_delay * (2 ** attempt) + random.uniform(0, 1), max_delay)
                    await asyncio.sleep(delay)
        return wrapper
    return decorator

@retry(max_attempts=3, base_delay=1.0)
async def fetch_data(url: str) -> dict:
    async with httpx.AsyncClient() as client:
        response = await client.get(url, timeout=10)
        response.raise_for_status()
        return response.json()

Circuit Breaker

import time
from enum import Enum

class CircuitState(Enum):
    CLOSED = "closed"       # Normal operation
    OPEN = "open"           # Failing, reject requests
    HALF_OPEN = "half_open" # Testing recovery

class CircuitBreaker:
    def __init__(self, failure_threshold: int = 5, recovery_timeout: float = 30.0):
        self.failure_threshold = failure_threshold
        self.recovery_timeout = recovery_timeout
        self.state = CircuitState.CLOSED
        self.failure_count = 0
        self.last_failure_time = 0.0

    async def call(self, func, *args, **kwargs):
        if self.state == CircuitState.OPEN:
            if time.time() - self.last_failure_time > self.recovery_timeout:
                self.state = CircuitState.HALF_OPEN
            else:
                raise CircuitOpenError(f"Circuit open, retry after {self.recovery_timeout}s")

        try:
            result = await func(*args, **kwargs)
            self._on_success()
            return result
        except Exception as e:
            self._on_failure()
            raise

    def _on_success(self):
        self.failure_count = 0
        self.state = CircuitState.CLOSED

    def _on_failure(self):
        self.failure_count += 1
        self.last_failure_time = time.time()
        if self.failure_count >= self.failure_threshold:
            self.state = CircuitState.OPEN

Timeout

async def with_timeout(coro, seconds: float):
    try:
        return await asyncio.wait_for(coro, timeout=seconds)
    except asyncio.TimeoutError:
        raise TimeoutError(f"Operation timed out after {seconds}s")

Bulkhead (Resource Isolation)

class Bulkhead:
    """Limit concurrent access to a resource."""
    def __init__(self, max_concurrent: int = 10):
        self.semaphore = asyncio.Semaphore(max_concurrent)

    async def execute(self, func, *args, **kwargs):
        async with self.semaphore:
            return await func(*args, **kwargs)

# Isolate different downstream services
payment_bulkhead = Bulkhead(max_concurrent=5)
inventory_bulkhead = Bulkhead(max_concurrent=20)

Fallback

async def get_user_profile(user_id: str) -> dict:
    try:
        return await primary_service.get_profile(user_id)
    except ServiceUnavailable:
        try:
            return await cache.get_profile(user_id)  # Stale cache fallback
        except CacheMiss:
            return {"user_id": user_id, "name": "Unknown", "_fallback": True}

Composition

Chain patterns for defense in depth:

Request → Timeout → Bulkhead → Circuit Breaker → Retry → Service Call

class ResilientClient:
    def __init__(self):
        self.circuit = CircuitBreaker(failure_threshold=5)
        self.bulkhead = Bulkhead(max_concurrent=10)

    @retry(max_attempts=3, base_delay=0.5)
    async def call(self, url: str) -> dict:
        return await with_timeout(
            self.bulkhead.execute(
                self.circuit.call, self._do_request, url
            ),
            seconds=15
        )

    async def _do_request(self, url: str) -> dict:
        async with httpx.AsyncClient() as client:
            response = await client.get(url, timeout=5)
            response.raise_for_status()
            return response.json()

Queue-Based Load Leveling

import asyncio
from collections import deque

class RateLimiter:
    def __init__(self, rate: int, per: float = 1.0):
        self.rate = rate
        self.per = per
        self.tokens = rate
        self.last_refill = time.time()
        self.lock = asyncio.Lock()

    async def acquire(self):
        async with self.lock:
            now = time.time()
            elapsed = now - self.last_refill
            self.tokens = min(self.rate, self.tokens + elapsed * (self.rate / self.per))
            self.last_refill = now
            if self.tokens >= 1:
                self.tokens -= 1
                return True
            return False

Health Check Patterns

from enum import Enum

class HealthStatus(Enum):
    HEALTHY = "healthy"
    DEGRADED = "degraded"
    UNHEALTHY = "unhealthy"

async def health_check() -> dict:
    checks = {
        "database": check_database(),
        "cache": check_cache(),
        "external_api": check_external_api(),
    }
    results = {}
    for name, check in checks.items():
        try:
            await asyncio.wait_for(check, timeout=5)
            results[name] = HealthStatus.HEALTHY
        except Exception:
            results[name] = HealthStatus.UNHEALTHY

    overall = (
        HealthStatus.HEALTHY if all(v == HealthStatus.HEALTHY for v in results.values())
        else HealthStatus.DEGRADED if any(v == HealthStatus.HEALTHY for v in results.values())
        else HealthStatus.UNHEALTHY
    )
    return {"status": overall.value, "checks": {k: v.value for k, v in results.items()}}

Failure Mode Analysis

Idempotency

async def process_payment(idempotency_key: str, amount: float):
    existing = await db.get_by_idempotency_key(idempotency_key)
    if existing:
        return existing  # Already processed

    result = await payment_gateway.charge(amount)
    await db.store(idempotency_key=idempotency_key, result=result)
    return result

Anti-Patterns

• Retry without backoff — Creates thundering herd on recovering services
• Retry on non-transient errors — 400 errors will never succeed on retry
• No timeout — Hanging requests consume resources indefinitely
• Cascading retries — Each layer retrying multiplies total attempts exponentially
• Circuit breaker too sensitive — Single failure shouldn't trip the circuit
• Ignoring partial failures — Assume any external call can fail