domains / / Azure/Azure-Sentinel

Error-handled API Calls

For every external API/data-store call, enforce a consistent error-handling contract: 1) Validate inputs up front (fail fast for predictable bad requests)

raw .md Error Handling Python

For every external API/data-store call, enforce a consistent error-handling contract:

1) Validate inputs up front (fail fast for predictable bad requests)

Example: if the computed window is invalid, skip/exit instead of calling the API.

2) Wrap the call in try/catch and log with traceback

Use logging.exception(...) (or equivalent) inside the except block so you keep the stack trace.
If you rethrow, prefer bare raise to preserve the original stack.

3) Retry only transient failures with status-aware logic

Retry on transient responses such as HTTP 429 and 5xx (e.g., 500/502/503/504).
Do not retry on non-transient/auth/client errors (e.g., 400/401); instead log a clear message and exit (or degrade gracefully).

4) Make retry termination explicit

Use a max-attempts counter/backoff and stop once exceeded.

Self-contained example (pattern to follow):

import logging
import time
import requests

TRANSIENT_STATUS = {429, 500, 502, 503, 504}

def call_with_retries(url, headers, params, max_retries=3, backoff_seconds=1):
    if params.get("from") and params.get("to") and params["from"] >= params["to"]:
        logging.info("Skipping call due to invalid time window")
        return None

    attempt = 0
    while True:
        try:
            resp = requests.get(url, headers=headers, params=params)
            if resp.status_code in TRANSIENT_STATUS:
                attempt += 1
                if attempt > max_retries:
                    logging.error(f"Exceeded retries for transient status {resp.status_code}")
                    return None
                time.sleep(backoff_seconds * attempt)
                continue

            if resp.status_code in {400, 401}:
                logging.error(f"Non-retryable API error {resp.status_code}: {resp.text}")
                return None

            resp.raise_for_status()
            return resp.json()

        except Exception:
            logging.exception("API call failed")
            attempt += 1
            if attempt > max_retries:
                raise
            time.sleep(backoff_seconds * attempt)

Apply this pattern consistently so failures are recoverable when appropriate, non-retryable errors don’t waste attempts, and logs/stack traces remain diagnosable.