When you are running a monolithic app on your laptop, debugging is easy. You just look at the console. But when your code is running in ephemeral containers in the cloud, debugging can be a nightmare.
Today, I upgraded my Finance Agent with professional Observability tools.
- Structured Logging (JSON)
I refactored my Python Lambda to log in JSON format.
Before: print(f"Error: {e}") (Hard to parse)
After: print(json.dumps({"level": "ERROR", "component": "Bedrock", "details": str(e)}))
This small change allows me to use CloudWatch Logs Insights to run SQL-like queries on my logs, such as filtering only errors related to the AI model.
- Visualizing Latency with AWS X-Ray
I enabled "Active Tracing" in the Lambda configuration. Now, AWS automatically generates a Service Map (see cover image). I can visually see that my Plaid API call takes 200ms, while my Bedrock AI generation takes 1.5s. This visual "report" is invaluable for optimization.
Top comments (0)