tenacity retry storm auditor

I have audited the ./services/payment-processor directory. I identified 3 instances of retry amplification where nested decorators and library defaults were creating up to 27x request spikes per user action.

Tenacity Retry Storm Audit

Amplification Analysis: Payment Flow

• Layer 1: @task(bind=True, max_retries=3) in tasks/process.py
• Layer 2: @retry(stop=stop_after_attempt(3)) in logic/billing.py
• Layer 3: urllib3.Retry(total=3) in clients/http.py
• Worst-case Multiplier: 27x attempts
• Result: One failure triggers 27 calls, likely causing the 429s you are seeing.

Recommended Fix (Patched in ./output/patches/STM-1-A.py)

from tenacity import retry, stop_after_attempt, stop_after_delay, wait_exponential_jitter

def wait_retry_after(fallback, cap: float = 60.0):
    def _wait(retry_state):
        exc = retry_state.outcome.exception() if retry_state.outcome else None
        resp = getattr(exc, "response", None)
        header = resp.headers.get("Retry-After") if resp is not None else None
        if header and header.strip().isdigit():
            return min(float(header), cap)
        return fallback(retry_state)
    return _wait

# Flattened: Single retry at the idempotent boundary
@retry(
    stop=stop_after_attempt(4) | stop_after_delay(30),
    wait=wait_retry_after(wait_exponential_jitter(initial=1, max=10)),
    reraise=True,
)
def unified_payment_call():
    # Inner layers now set to 0 retries / 1 attempt
    return stripe_client.execute(retries=0)

Next steps

1. Review the detailed amplification table in ./output/tenacity-retry-storm-auditor-report.md.
2. Apply the flattened retry patch to clients/gateway.py to reduce gateway load.
3. Update environment variables to set the new 30s global deadline.