Lambda

Run code without managing servers. You upload a function; AWS runs it in response to events, scales it horizontally, and bills per millisecond of execution.

• Event-driven; one concurrent execution per in-flight event.
• Stateless — persist anything durable in S3/DynamoDB/RDS.
• Pay for requests + GB-seconds (memory × duration). Idle = free.

Limits (defaults)

Resource	Limit
Timeout	15 min max
Memory	128 MB – 10,240 MB (CPU scales with it; ~6 vCPU at max)
/tmp ephemeral storage	512 MB – 10,240 MB
Deployment package	50 MB zipped / 250 MB unzipped; 10 GB as container image
Payload	6 MB sync, 256 KB async
Env vars	4 KB total
Account concurrency	1,000 (soft, raise via quota)

Invocation Models

• Synchronous — API Gateway, ALB, SDK Invoke. Caller waits; errors return to caller.
• Asynchronous — S3, SNS, EventBridge. Lambda queues + retries (2x), then DLQ / on-failure destination.
• Event source mapping (poll-based) — SQS, Kinesis, DynamoDB Streams. Lambda polls and batches.

Concurrency

• Reserved concurrency — caps (and guarantees) a function’s slice of the account pool.
• Provisioned concurrency — pre-warmed environments to eliminate cold starts (costs even when idle).
• Cold start = new execution environment init (download, runtime, init code). Mitigate with provisioned concurrency, smaller packages, SnapStart (Java/Python/.NET), and lazy-loading SDK clients.

Building Blocks

• Execution role (IAM) — what the function may do. Resource policy — who may invoke it.
• Layers — shared dependencies/runtime mounted at /opt (up to 5).
• Versions (immutable) + aliases (mutable pointers) → enable Canary Deployments.
• Environment variables — config; encrypt secrets with KMS or pull from Secrets Manager / Parameter Store.

Snippets

# Tail logs live
aws logs tail /aws/lambda/my-fn --follow --since 5m
 
# Invoke synchronously and capture the response
aws lambda invoke --function-name my-fn \
  --payload '{"key":"value"}' --cli-binary-format raw-in-base64-out \
  /dev/stdout
 
# Set reserved concurrency
aws lambda put-function-concurrency \
  --function-name my-fn --reserved-concurrent-executions 50

In front of a database?

Lambda scales faster than a relational DB can accept connections. Use RDS Proxy (or DynamoDB) to avoid connection-pool exhaustion.