import boto3
import datetime
import pandas as pd

def analyze_serverless_costs(region="us-east-1", days=30):
    cloudwatch = boto3.client("cloudwatch", region_name=region)
    lambda_client = boto3.client("lambda", region_name=region)
    
    end_time = datetime.datetime.utcnow()
    start_time = end_time - datetime.timedelta(days=days)
    
    # Fetch Lambda metrics
    functions = lambda_client.list_functions()["Functions"]
    cost_data = []
    
    for func in functions:
        name = func["FunctionName"]
        memory = func["MemorySize"]
        
        # Get duration & invocations
        duration = cloudwatch.get_metric_statistics(
            Namespace="AWS/Lambda",
            MetricName="Duration",
            Dimensions=[{"Name": "FunctionName", "Value": name}],
            StartTime=start_time, EndTime=end_time,
            Period=86400, Statistics=["Average"]
        )["Datapoints"]
        
        invocations = cloudwatch.get_metric_statistics(
            Namespace="AWS/Lambda",
            MetricName="Invocations",
            Dimensions=[{"Name": "FunctionName", "Value": name}],
            StartTime=start_time, EndTime=end_time,
            Period=86400, Statistics=["Sum"]
        )["Datapoints"]
        
        avg_duration_ms = duration[0]["Average"] if duration else 0
        total_invocations = invocations[0]["Sum"] if invocations else 0
        
        # AWS Lambda pricing (approximate, update as needed)
        gb_sec_rate = 0.0000166667  # per GB-second
        request_rate = 0.20  # per 1M requests
        gb_sec_cost = (memory / 1024) * (avg_duration_ms / 1000) * total_invocations * gb_sec_rate
        request_cost = (total_

invocations / 1_000_000) * request_rate total_cost = gb_sec_cost + request_cost

    cost_data.append({
        "Function": name,
        "Memory_MB": memory,
        "Avg_Duration_ms": avg_duration_ms,
        "Invocations": total_invocations,
        "Estimated_Cost_USD": round(total_cost, 4),
        "Cost_Efficiency": round(total_cost / max(total_invocations, 1), 6)
    })

df = pd.DataFrame(cost_data)
return df.sort_values("Estimated_Cost_USD", ascending=False)

if name == "main": report = analyze_serverless_costs() print(report.head(10))

### 2. Right-Sizing & Execution Pattern Optimization

Cost efficiency hinges on aligning memory allocation with actual CPU/memory utilization and reducing unnecessary compute cycles.

**Connection Pooling & Initialization Caching**
Serverless functions reuse execution environments. Move heavy initialization outside the handler to leverage the execution context cache:

```javascript
// ❌ Anti-pattern: Connection created per invocation
export const handler = async (event) => {
  const db = await connectToDatabase(); // Cold start + network latency
  return await db.query(event.query);
};

// ✅ Optimized: Connection cached across invocations
const db = connectToDatabase(); // Initialized once per execution environment

export const handler = async (event) => {
  return await db.query(event.query);
};

Payload Compression & Async Offloading Reduce data transfer and execution time by compressing payloads and offloading non-critical work:

import gzip
import json
import boto3

def handler(event, context):
    payload = event.get("body", "")
    if len(payload) > 10_000:  # >10KB threshold
        compressed = gzip.compress(payload.encode())
        # Store in S3, pass reference to downstream consumer
        s3 = boto3.client("s3")
        key = f"compressed/{context.aws_request_id}.gz"
        s3.put_object(Bucket="payload-archive", Key=key, Body=compressed)
        return {"statusCode": 202, "body": json.dumps({"reference": key})}
    
    # Process small payloads synchronously
    return {"statusCode": 200, "body": process(payload)}

3. IaC Cost-Optimized Defaults

Infrastructure as Code enforces cost-aware defaults across environments. Below is a Terraform configuration with production-ready cost controls:

resource "aws_lambda_function" "optimized" {
  function_name    = "cost-optimized-func"
  runtime          = "nodejs18.x"
  handler          = "index.handler"
  memory_size      = 512          # Right-sized; adjust based on telemetry
  timeout          = 15           # Prevent runaway executions
  publish          = true
  architectures    = ["arm64"]    # Graviton2: ~20% cheaper, faster

  environment {
    variables = {
      NODE_OPTIONS = "--enable-source-map"
      ENABLE_DEBUG = "false"
    }
  }

  tracing_config {
    mode = "PassThrough"          # X-Ray only when needed
  }

  dead_letter_config {
    target_arn = aws_sqs_queue.dlq.arn
  }

  tags = {
    Environment = "production"
    CostCenter  = "engineering"
    Optimized   = "true"
  }
}

resource "aws_cloudwatch_log_group" "lambda_logs" {
  name              = "/aws/lambda/${aws_lambda_function.optimized.function_name}"
  retention_in_days = 14          # Reduce log storage costs
}

Pitfall Guide

#	Pitfall	Why It Happens	Mitigation
1	Provisioned Concurrency Overuse	Teams provision concurrency to avoid cold starts, but pay for idle reserved capacity regardless of traffic.	Use provisioned concurrency only for latency-critical endpoints (<50ms SLA). Route everything else to on-demand scaling.
2	Data Transfer Blind Spot	Developers focus on compute costs but ignore egress fees, which scale with payload size and cross-region calls.	Compress payloads, cache responses, use regional endpoints, and route through CloudFront or API Gateway caching.
3	Memory-CPU Misalignment	Conservative memory settings prolong execution, increasing GB-sec costs and invocation counts.	Run load tests across memory tiers (128MB–1024MB). Pick the tier with lowest total cost, not lowest per-GB price.
4	Unbounded Retry Storms	Misconfigured retry policies or dead-letter queues trigger exponential invocations during downstream failures.	Implement exponential backoff with jitter, set maximum retry attempts, and monitor DLQ depth with alerts.
5	Verbose Logging in Production	Debug-level logging generates massive CloudWatch costs and increases function duration.	Use structured logging with severity levels. Route debug logs to S3 or disable in production via environment variables.
6	Missing Cost Allocation Tags	Without consistent tagging, costs cannot be attributed to teams, features, or environments, obscuring optimization targets.	Enforce tagging via IAM policies and CI/CD gates. Use AWS Cost Explorer tags to split spend by project, owner, and environment.

Production Bundle

✅ Serverless Cost Optimization Checklist

Pre-Deployment

• Right-size memory based on load test telemetry
• Set timeout ≤ actual max execution time + 20% buffer
• Enable ARM64 architecture where compatible
• Configure DLQ with retry limits & jitter
• Apply mandatory cost allocation tags

Runtime & Monitoring

• Enable CloudWatch metrics for Duration, Errors, Throttles
• Set budget alerts at 50%, 80%, 100% of forecast
• Monitor data transfer egress vs. compute cost ratio
• Track cold start frequency & duration
• Validate logging level matches environment

Optimization Cycle

• Run weekly cost analysis script
• Identify top 3 cost drivers & optimize execution patterns
• Archive or delete unused functions/layers
• Review provisioned concurrency utilization (<60% = overprovisioned)
• Document cost baselines per service/feature

📊 Decision Matrix: When to Use Serverless vs. Alternatives

Workload Profile	Serverless Fit	Cost Threshold	Recommended Alternative
Event-driven, spiky traffic, <15s runtime	✅ High	< $500/mo per service	N/A
Steady-state, >50% utilization, long-running	❌ Low	> $2k/mo	Containers (ECS/EKS) or VMs
Batch processing, predictable throughput	⚠️ Medium	$500–$2k/mo	Step Functions + Lambda or Fargate
GPU/ML inference, custom OS dependencies	❌ Low	N/A	SageMaker, EC2, or Kubernetes
High-throughput API, strict latency SLA	⚠️ Medium	Depends on concurrency	API Gateway + Lambda + Provisioned Concurrency

📄 Config Template: Serverless Framework (Cost-Optimized)

service: cost-optimized-api

provider:
  name: aws
  runtime: nodejs18.x
  architecture: arm64
  memorySize: 512
  timeout: 10
  stage: ${opt:stage, 'dev'}
  region: ${opt:region, 'us-east-1'}
  environment:
    LOG_LEVEL: ${env:LOG_LEVEL, 'info'}
    NODE_OPTIONS: '--enable-source-map'
  tags:
    Environment: ${self:provider.stage}
    CostCenter: 'platform-engineering'
    ManagedBy: 'serverless-framework'
  iamRoleStatements:
    - Effect: Allow
      Action:
        - logs:CreateLogGroup
        - logs:CreateLogStream
        - logs:PutLogEvents
      Resource: 'arn:aws:logs:*:*:*'
  deploymentBucket:
    name: ${self:service}-${self:provider.stage}-deploy
  apiGateway:
    restApiId: ${env:API_GATEWAY_ID}
    restApiRootResourceId: ${env:API_GATEWAY_ROOT_ID}

functions:
  main:
    handler: src/handler.main
    events:
      - http:
          path: /process
          method: post
          cors: true
          integration: lambda
          timeout: 10
    layers:
      - ${env:LAYER_ARN}
    reservedConcurrency: 50
    provisionedConcurrency: 0  # Enable only for critical paths

plugins:
  - serverless-prune-plugin
  - serverless-cost-reporter

custom:
  prune:
    automatic: true
    number: 3

🚀 Quick Start: 5-Step Optimization Pipeline

1. Instrument: Deploy the cost analysis script to a scheduled Lambda or CI/CD job. Tag all resources with CostCenter, Environment, and Owner.
2. Baseline: Run load tests across memory tiers (128MB, 256MB, 512MB, 1024MB). Record duration, invocations, and total cost. Select the most cost-efficient tier.
3. Harden: Set timeouts, enable DLQs, disable debug logging in production, and apply ARM64 architecture. Configure CloudWatch budget alerts.
4. Automate: Integrate cost checks into your CI/CD pipeline. Block deployments if estimated monthly cost exceeds threshold or if tags are missing.
5. Iterate: Review telemetry weekly. Archive unused functions, right-size provisioned concurrency, compress payloads, and offload async work. Document savings per optimization cycle.

Serverless cost optimization is not a FinOps afterthought; it's an architectural discipline. By embedding telemetry, enforcing IaC defaults, and continuously aligning execution patterns with pricing models, teams transform serverless from a cost center into a scalable, predictable, and financially efficient compute engine.