Bottlenecks in Serverless Applications

on YouTube ↗53m

" summary

Shows concrete failure modes and fixes: managed APIs and services (Lambda, Kinesis, DynamoDB, CloudWatch, SNS/SQS, RDS) create capacity and latency bottlenecks that inflate cost and queue depth, exemplified by a $30,000 cost spike from aggressive CloudWatch polling. Presents tooling and numbers: benchmarking memory vs duration shows 128MB vs 512MB tradeoffs and CPU/network effects, explains shard/partition key limits in Kinesis, and describes the successful CloudWatch→Kinesis streaming fix that replaces polling and eliminates the queue backlog.

AI-written summary · about

" chapters (14)

1. 0:00Start
2. 0:35Topics to discuss
3. 2:20What is Serverless?
4. 4:02Why Serverless?
5. 5:16The limitations of Faas
6. 7:59The properties of Serverless applications
7. 9:40A real example - HSBC Solution Overview
8. 13:14Non-Serverless Example
9. 21:35Benchmark your Lambda
10. 29:50More Serverless bottlenecks
11. 30:57A real life example
12. 33:17Scaling
13. 35:28The correct approach
14. 36:28Conclusions

" related talks

• 51mServerless architectural patternsYan Cui · NDC Oslo 2018
• 39mBuilding Resilient Serverless ApplicationsYan Cui · Reliable Web Summit 2021
• 36mObserving all the Serverless ThingsPeter Elger, Guilherme Dalla Rosa · GOTO EDA Day 2022
• 43mThe Art of Embracing Failures with Serverless ArchitecturesAnahit Pogosova · GOTO EDA Day London 2024
• 25mWhere Did All My Servers Go?Garth Henson · JSConf Hawai 2019
• 40mSh... Fail Happens: Fail-aware Events Processing on AWSMarcin Sodkiewicz · GOTO EDA Day London 2024
• 01h 00mServerless Architecture - Tales from a world without serversRobin Weston · NDC London 2017
• 51mConfusion In The Land Of The ServerlessSam Newman · JavaDay Ukraine 2017