Compute
AWS Lambda
AWS Lambda runs event-driven code without server management and scales per invocation. It works well for brief validation, orchestration adapters, and Bedrock agent actions rather than heavyweight model hosting.
Key points
- Execution duration, memory, payload, and ephemeral-storage quotas bound workloads
- Reserved concurrency can protect downstream model quotas
- A function execution may be retried, requiring idempotent side effects
When to use it
- Validate and authorize an agent tool request
- Transform a document event before starting ingestion
Exam tips
- Use Lambda for bounded stateless work; containers or SageMaker are better for long GPU inference
- Keep SDK clients outside the handler where safe and align timeouts with every downstream call