How to Fail at Serverless (without even trying)
summary
Teams move fast and hit real failure modes: Lambda temp folder fills and causes OOMs because instance reuse makes state, image library leaks file descriptors so only max‑concurrency load tests reproduce it, session state lives in passport memory across instances, global S3 endpoint DDoS takes down builds, Athena 429s when S3 hot‑spots reads, replication doubles egress costs, and a lifecycle transition button would cost ~$100,000 to run — many fixes work but introduce new operational or cost surprises.
AI-written summary · about
chapters (15)
- 0:00Failure matters
- 1:26Lambda as storage
- 3:55Lambda file descriptor state
- 5:22Session management traps
- 6:47S3 endpoint consistency
- 8:52Athena hotspotting failures
- 11:00State machine retries
- 13:27S3 lifecycle oversights
- 15:52S3 transition costs
- 16:36Infrastructure as code pitfalls
- 18:48VPC multi-AZ misconfigs
- 20:36Lambda runtime issues
- 22:01CloudFormation limits
- 23:30API Gateway security risks
- 24:11Share your failures
related talks
51mConfusion In The Land Of The Serverless
43mThe Art of Embracing Failures with Serverless Architectures
39mBuilding Resilient Serverless Applications
28mCostly mistakes in serverless computing
44mServerless Architecture on Production: Good, Bad and Ugly Parts
49mYou Don't Want Serverless
25mWhere Did All My Servers Go?
01h 00mServerless - reality or BS - notes from the trenches