AWS Lambda Node.js: Cold Starts, Layers, and Production Patterns

Lambda Execution Model

Containers are created (cold start ~300ms), then reused (warm ~0ms). Initialize clients outside handlers.

Cold Start Optimization

// Initialize OUTSIDE handler — reused on warm invocations
const db = DynamoDBDocumentClient.from(new DynamoDBClient({ maxAttempts: 3 }))
const pgPool = new Pool({ connectionString: process.env.DATABASE_URL, max: 1 })

export const handler = async (event: APIGatewayEvent) => {
  // db and pgPool are already initialized — no cold start overhead
}

SQS Partial Batch Failures

export const handler = async (event: SQSEvent) => {
  const failures: SQSBatchItemFailure[] = []

  await Promise.allSettled(event.Records.map(async (record) => {
    try {
      await processMessage(JSON.parse(record.body))
    } catch (err) {
      failures.push({ itemIdentifier: record.messageId })
    }
  }))

  return { batchItemFailures: failures }
}

Cached SSM Secrets

const secretCache = new Map<string, string>()

async function getSecret(name: string) {
  if (secretCache.has(name)) return secretCache.get(name)!
  const { Parameter } = await ssm.send(new GetParameterCommand({ Name: name, WithDecryption: true }))
  secretCache.set(name, Parameter!.Value!)
  return Parameter!.Value!
}

Cost Optimization

-> Encode Lambda vars with the Base64 Converter.