How to avoid common mistakes and misconceptions when working...
summary
Java Lambdas suffer two concrete cold starts: init duration ~1.8s (128 MB example) and JVM warmup that slows early requests; loading a different code path can spike latency by seconds. Provisioned concurrency removes many cold starts and a constructor pre-warm that executes common code paths cuts first-call latency from ~700ms to pre-warm in constructor -> 33ms. Cost math uses AWS Pricing Calculator and CloudWatch metrics; at scale (example: StackExchange 495 req/s -> 99 provisioned instances) Fargate often becomes cheaper. L1 in-process caches fail; use L2 (Redis) or Lambda extensions.
AI-written summary · about
chapters (16)
- 0:00Intro
- 3:47Lambda instance lifecycle
- 9:45Cold start type 1 (container recycle)
- 16:05Cold start type 2 (JVM JIT and lazy load)
- 17:32Reducing cold starts
- 23:47Cold start for provisioned concurrency
- 24:45Provisioned concurrency best practices
- 29:12Lambda pricing
- 31:00How to calculate AWS costs
- 32:28Assumptions
- 33:12Scenarios considered
- 34:06"Hello world", 1 request per second
- 35:13Fargate, 1 request per second
- 38:49Simulating Stack Exchange
- 46:37Caching in Lambda - challenges
- 48:09Caching in Lambda - approaches
related talks
19mTen things that improve the performance of your Java based AWS Lambda
53mHigh performance Serverless Java on AWS
51mHow To Reduce Cold Starts for Java Serverless Applications in AWS
46mThawing Java on AWS Lambda Reducing Cold Start Times from 11 Seconds to 1
34mThawing Java on AWS Lambda
46mReduce Cold Starts for Java Serverless Applications in AWS
51mAdapting Java for the Serverless world!
49mSpeeding Up AWS Lambda: JVM and Cold Start in Practice