Professional API Clients: httpx & Async Patterns

requests is fine for scripts. For production services consuming APIs at scale, you need httpx — it supports async, HTTP/2, connection pooling, and timeout control that requests simply can't match.

Why httpx Over requests?

# requests (sync, blocking, one connection at a time)
r = requests.get("https://api.example.com")

# httpx (async, non-blocking, connection pooling)
async with httpx.AsyncClient() as client:
    r = await client.get("https://api.example.com")

The async version can make thousands of concurrent requests without blocking threads. In a FastAPI app, blocking HTTP calls kill your concurrency. httpx keeps everything async-native.

Building a Production API Client

Async HTTP — The Real Superpower

In an async FastAPI endpoint, you cannot use requests.get() — it blocks the entire event loop. Here's the async pattern:

# ❌ WRONG — blocks all other requests
@app.get("/aggregate")
def aggregate():
    a = requests.get("https://api.service-a.com")   # BLOCKS
    b = requests.get("https://api.service-b.com")   # BLOCKS
    return combine(a, b)

# ✅ RIGHT — non-blocking, concurrent
@app.get("/aggregate")
async def aggregate():
    async with httpx.AsyncClient() as client:
        a, b = await asyncio.gather(
            client.get("https://api.service-a.com"),
            client.get("https://api.service-b.com"),
        )
    return combine(a.json(), b.json())

Circuit Breaker Pattern

When downstream APIs go down, retrying endlessly makes things worse. A circuit breaker stops requests after a threshold of failures:

Handling API Pagination

Almost every API paginates. There are 3 common patterns:

GitHub, Stripe, and Twitter all use cursor-based pagination. It's the only approach that doesn't break when items are inserted between pages.

Quiz

Key Takeaways

✅ httpx for async HTTP — don't block your event loop with requests.
✅ Rate limiters (token bucket) prevent hitting API limits.
✅ Exponential backoff + jitter for resilient retries.
✅ Circuit breakers stop cascading failures when downstream APIs die.
✅ Cursor pagination is the only stable pagination for mutable datasets.

Keep coding! 🚀