How to Handle Errors and Retries in a Stateless Environment
summary
Serverless platforms expose AWS Lambda up to 15 minute runs but also timeouts, out‑of‑memory and cold starts with containers reused minutes to hours. Cloud services like SNS retries three times and Kinesis can silently retry, which breaks non‑idempotent flows. Use Step Functions orchestration for managed retries and state, and automated instrumentation via Lambda Layers to get end‑to‑end traces and logs without touching code. Traces reveal surprises (OAuth call 700ms vs 50ms) and pinpoint performance and cost hotspots.
AI-written summary · about
chapters (11)
- 0:00Introduction to serverless
- 1:47What is serverless computing
- 4:37Benefits of serverless
- 8:06Challenges in serverless
- 12:58Handling errors with retries
- 18:15Orchestration with Step Functions
- 18:36Observability in serverless
- 22:23Distributed tracing explained
- 24:31Automating observability
- 28:22Best practices for serverless
- 34:28Q&A
related talks
43mThe Art of Embracing Failures with Serverless Architectures
25mHow to Fail at Serverless (without even trying)
48mServerless design patterns
36mObserving all the Serverless Things
35mEfficiency of Serverless Architectures with AWS Solutions
19m(N+1) Lessons I learned about serverless
39mBuilding Resilient Serverless Applications
44mServerless Architecture on Production: Good, Bad and Ugly Parts