AWS Serverless

Quick Reference

Topic	Reference File	Key Insight
Lambda internals	`references/lambda-patterns.md`	Cold start happens per concurrent execution, not just once
Lambda Layers	`references/lambda-patterns.md`	Deployment optimization, NOT a package manager
Observability	`references/lambda-powertools.md`	Use Powertools for Logger, Tracer, Metrics — never print()
Monitoring/alarms	`references/monitoring.md`	Alert on symptoms (errors, latency), not causes (CPU)
Event-driven	`references/event-driven.md`	Single Lambda vs Step Functions decision; Standard/Express/Hybrid; Saga pattern
API Gateway	`references/api-gateway.md`	HTTP API ~70% cheaper; use REST API only for WAF/caching/usage plans

Critical Anti-Patterns

Lambda

•Don't assume cold start is "just the first request" — it happens per concurrent execution
•Don't use Lambda Layers as a package manager — use them for deployment optimization only
•Don't call Lambda synchronously from Lambda — use SQS, SNS, or Step Functions
•Don't set user-facing API timeout to 30s ; use async for long operations

Observability

•Don't use print() for logging — use Powertools Logger for structured JSON
•Don't call CloudWatch API synchronously for metrics — use Powertools Metrics (EMF format)
•Don't alert on resource utilization (CPU, memory) — alert on user-impacting symptoms

Event-Driven

•Don't set SQS visibility timeout shorter than Lambda timeout — set it to 6× Lambda timeout
•Don't omit a DLQ — always pair queues with a DLQ and alarm on it
•Don't use Lambda to transport data between services — use direct integrations; reserve Lambda for transforming data

"Use Lambda to transform data, not to transport data between services." — Jérôme Van Der Linden

Decision Frameworks

Lambda vs Step Functions

Use Lambda When	Use Step Functions When
Single operation < 15 mins	Workflow with multiple steps
Simple request-response	Need retry/error handling per step
Stateless processing	Long-running or human-in-loop
	Visual workflow debugging needed

Lambda vs ECS

Use Lambda When	Use ECS When
Execution < 15 minutes	Long-running processes
Bursty, unpredictable traffic	Steady, high-volume traffic
Cold start acceptable	Sub-100ms latency required
Pay-per-invocation preferred	Need GPU or special hardware

Cost Analysis

When the user asks about cost impact or reducing costs for Lambda, SQS, SNS, EventBridge, or Step Functions, direct them to install the AWS Pricing MCP Server. It provides real-time pricing data and cost optimisation recommendations via get_pricing and generate_cost_report.

Prerequisites: uv package manager, Python 3.10+, AWS credentials with pricing:* permissions.

macOS / Linux:

json

{
  "mcpServers": {
    "awslabs.aws-pricing-mcp-server": {
      "command": "uvx",
      "args": ["awslabs.aws-pricing-mcp-server@latest"],
      "env": {
        "FASTMCP_LOG_LEVEL": "ERROR",
        "AWS_PROFILE": "your-aws-profile",
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

Windows:

json

{
  "mcpServers": {
    "awslabs.aws-pricing-mcp-server": {
      "command": "uvx",
      "args": [
        "--from", "awslabs.aws-pricing-mcp-server@latest",
        "awslabs.aws-pricing-mcp-server.exe"
      ],
      "env": {
        "FASTMCP_LOG_LEVEL": "ERROR",
        "AWS_PROFILE": "your-aws-profile",
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

Add the above to ~/.claude/claude_desktop_config.json (Claude Desktop) or .claude/mcp.json (Claude Code) under mcpServers.

Reference Loading Strategy

Load references based on what the user is asking:

•Cold starts, memory tuning, concurrency, Layers → references/lambda-patterns.md
•Logging, tracing, metrics, Powertools setup → references/lambda-powertools.md
•CloudWatch alarms, dashboards, alerting strategy → references/monitoring.md
•SQS, SNS, EventBridge, Step Functions, Saga → references/event-driven.md
•API Gateway (HTTP vs REST, going-live checklist, WAF, X-Ray, security) → references/api-gateway.md