AWS Lambda Optimization

Cold start reduction, memory tuning, and performance patterns for Lambda

Language / Topic

Amazon Web Services

Rules

AWS Lambda Optimization · balanced balanced

- Minimize deployment package size: use tree-shaking, exclude dev dependencies, prefer ESM for Node.js

- Set memory based on benchmarks — use AWS Lambda Power Tuning to find the cost/performance sweet spot

- Initialize SDK clients and database connections outside the handler to reuse across invocations

- Use Provisioned Concurrency for latency-sensitive endpoints; SnapStart for Java Lambdas

- Avoid synchronous calls to other Lambda functions — use SQS, SNS, or Step Functions instead

- Minimize deployment package size: tree-shaking, exclude dev deps, use Lambda layers for shared code

- Use Lambda Power Tuning to benchmark memory settings — CPU scales linearly with memory allocation

- Initialize SDK clients, DB connections, and config outside the handler for reuse across warm invocations

- Use Provisioned Concurrency for sub-100ms latency requirements; SnapStart for Java cold start reduction

- Avoid synchronous Lambda-to-Lambda calls — use SQS for decoupling, Step Functions for orchestration

- Use ARM64 (Graviton2) runtime for ~20% better price-performance on most workloads

- Batch SQS messages: set batch size 10 and batch window 5s for throughput-oriented processing

- Use Connection pooling for RDS: RDS Proxy for Lambda, or keep-alive settings for direct connections

- Set reasonable timeouts: 10-30s for API handlers, up to 15 minutes for batch/async processing

- Enable X-Ray active tracing to identify slow downstream calls and optimize the critical path

- Use S3 Transfer Acceleration or multipart uploads for large file operations from Lambda