Error handling and retries

Trying again, but not forever

Networks fail transiently. A client that gives up on the first 503 is fragile; a client that retries forever is worse, because it turns a small outage into a stampede.

The shape worth learning:

for attempt in range(max_attempts):
    response = requests.get(url)
    if response.ok:
        return response.json()
    if not should_retry(response.status_code):
        break            # a 404 will not fix itself
    time.sleep(backoff * (2 ** attempt))   # exponential backoff
raise RuntimeError("gave up")

Three ideas in there, all load-bearing:

1. Only retry what might succeed. Retrying a 4xx wastes everyone's time.
2. Back off exponentially. Waiting 1s, 2s, 4s gives the server room to recover; hammering every 100ms does not.
3. Have a limit. Unbounded retries are an outage amplifier.

The fake /flaky endpoint fails twice and then succeeds, so a retry loop is genuinely necessary.

Your turn: write fetch_with_retry(url, max_attempts=5) returning the parsed body, retrying only on retryable statuses, and returning None if it never succeeds.