Services at a glance

Amazon EC2

Key concepts

• AMI (Amazon Machine Image): template with OS and software, region-scoped, copyable
• Instance type: hardware profile named family + generation + size, e.g. m6i.large
• Key pair: SSH key to log in (or use SSM Session Manager, no open ports)
• Security group: virtual firewall (5. AWS Networking)
• User data: script that runs at first boot
• Instance metadata (169.254.169.254): data about the instance, including role credentials (use IMDSv2)
• Tenancy: shared (default), dedicated instance, dedicated host

Instance families

Lifecycle

pending → running → stopping → stopped → terminated - Stop: no instance charge, EBS volume kept (and billed), public IP lost unless Elastic IP - Hibernate: saves RAM to EBS - Terminate: instance gone, root volume deleted by default

$ aws ec2 run-instances --image-id ami-0abcdef1234567890 --instance-type t3.micro \
    --key-name mykey --security-group-ids sg-0123 --subnet-id subnet-0123 \
    --iam-instance-profile Name=app-role
$ aws ec2 describe-instances --query 'Reservations[].Instances[].[InstanceId,State.Name]'
$ aws ec2 stop-instances --instance-ids i-0123456789abcdef0

Placement groups

• Cluster: low latency, one AZ
• Spread: separate hardware, up to 7 per AZ
• Partition: groups on separate racks (Hadoop, Kafka)

Purchase options are in 1. AWS Cloud Concepts.

Scaling and Load Balancing

Auto Scaling Groups (ASG)

• Define launch template, min / desired / max size, subnets across AZs
• Policies: target tracking (keep CPU at 50%), step, scheduled, predictive
• Replaces unhealthy instances using EC2 or ELB health checks

Elastic Load Balancing

• Target groups and health checks decide where traffic goes
• Cross-zone load balancing, sticky sessions, connection draining (deregistration delay)

Elastic Beanstalk

PaaS: upload a package (Java, .NET, Node, Python, PHP, Ruby, Go, Docker) and it provisions EC2, ASG, ELB and monitoring. Deployment policies: all at once, rolling, rolling with batch, immutable, blue/green, traffic splitting. You still own the underlying resources.

AWS Lambda

Run code without servers. Billed by requests and duration × memory (1 ms granularity).

• Triggers: API Gateway, S3, DynamoDB Streams, SQS, SNS, EventBridge schedule, and more
• Limits: 15 min max timeout, 128 MB to 10 GB memory (CPU scales with it), 10 GB container images, 250 MB zipped package
• Concurrency: default account limit 1000; reserved and provisioned concurrency
• Cold starts: mitigate with provisioned concurrency / SnapStart
• Layers share libraries, versions and aliases for safe deployment
• Permissions: execution role; invoke permissions via resource policy

import json

def handler(event, context):
    name = (event.get("queryStringParameters") or {}).get("name", "world")
    return {"statusCode": 200, "body": json.dumps({"message": f"Hello, {name}"})}

$ aws lambda invoke --function-name hello --payload '{}' out.json

More in 6. AWS Developer Services and Serverless.

Containers

ECS launch types: EC2 (you manage hosts) or Fargate. Task role gives containers AWS permissions.

Choosing Compute

• Full control, legacy apps → EC2
• Just deploy my web app → Elastic Beanstalk / App Runner
• Short event-driven code → Lambda
• Containers, no cluster ops → ECS on Fargate
• Kubernetes standard → EKS
• Cheap, simple VPS → Lightsail