AWS Compute Services — EC2, Lambda, Elastic Beanstalk, SAM

Key Takeaways

• Lambda scales automatically, but concurrency limits apply — understand reserved vs provisioned concurrency
• EC2 instance types matter for the exam: general-purpose (t/m), compute-optimized (c), memory-optimized (r/x)
• Elastic Beanstalk = PaaS — you upload code, AWS manages infra; .ebextensions customizes it
• SAM = shorthand for serverless — transforms to CloudFormation under the hood

Amazon EC2 (Elastic Compute Cloud)

EC2 is your foundational compute service on AWS. It gives you resizable compute capacity with full control over instance configuration, storage, networking, and security.

Instance Types

Different instance types optimize for different workloads:

Type	Family	Best For
General-purpose	t (burstable), m (standard)	Web apps, small databases, mixed workloads
Compute-optimized	c (high-performance)	Batch jobs, media transcoding, HPC
Memory-optimized	r (RAM-heavy), x (extreme)	In-memory databases, real-time analytics
Storage-optimized	i, d, h (I/O intensive)	NoSQL databases, data warehousing
Accelerated	p (GPU), f (FPGA)	ML training, graphics rendering

The exam heavily tests instance types—know which family fits which use case.

Instance Lifecycle & Management

Instances go through states: launch → running → stop (reversible) → start → terminate (final). Stopping an instance preserves the EBS root volume, but terminated instances are gone forever. Auto Scaling automatically adjusts instance count based on demand or schedules.

Interview Angle

When asked about EC2 security, always mention: Use an IAM instance profile (role) attached to the instance. Never put access keys on EC2—they’re visible in instance metadata and painful to rotate. This is a non-negotiable security best practice.

Elastic Block Store (EBS)

EBS provides persistent block storage attached to EC2 instances. Volume types include:

Volume Type	Use Case	Durability
gp2/gp3 (SSD)	General workloads, databases	Highly durable
io1/io2 (IOPS SSD)	High-speed databases, I/O intensive	Very high IOPS
st1 (HDD)	Big data, sequential I/O	Good throughput
sc1 (HDD)	Infrequent access, archival	Low cost

Critical for the exam: EBS volumes are AZ-specific. You cannot attach a volume from us-east-1a to an instance in us-east-1b. This is a common gotcha.

Networking & Security

Security Groups act as virtual firewalls, controlling inbound and outbound traffic at the instance level. Elastic IPs are static, publicly routable addresses that can be remapped between instances. Elastic Load Balancers (ALB, NLB, CLB) distribute traffic across multiple EC2 instances.

Common Gotchas

EBS volumes are locked to their AZ—can’t move between AZs directly. Use EBS snapshots to copy data to another AZ, then create a new volume.

AWS Lambda

Lambda is fully serverless—AWS manages servers, scaling, and patching. You pay only for compute time consumed.

Core Concepts

Lambda functions are event-driven and scale automatically. Triggers include S3, DynamoDB Streams, Kinesis, SNS, API Gateway, CloudWatch Events, and more. Execution happens in isolated execution environments.

Concurrency & Scaling

Lambda concurrency is the number of functions executing simultaneously. AWS allocates one execution environment per concurrent request.

Interview Angle

Interviewers ask: “How does Lambda scale?” Answer: One execution environment per concurrent request. Default account limit is 1,000 concurrent executions. Hit the limit → 429 Throttling error. Solutions: use reserved concurrency (guarantee capacity) or provisioned concurrency (keep warm, predictable).

Default concurrency limits depend on account and region, but reserved/provisioned concurrency gives you explicit control.

Cold Starts & Optimization

A cold start occurs when AWS must create a new execution environment to run your function. Initialization overhead (Java can take 1-2 seconds; Python faster) impacts latency.

Cold start factors: Function size, language runtime, VPC attachment (worst), and code complexity. VPC-attached Lambda is particularly slow because it must attach an ENI to the function.

Interview Angle

Cold start fixes: Use provisioned concurrency to keep functions warm, implement lightweight initialization, avoid VPC if possible, or use Snap Start (Java only—pre-compiled snapshot reduces cold start to ~50ms).

Timeout & Limitations

Max execution time is 15 minutes (900 seconds). Lambda is not suited for long-running processes—use Fargate or EC2 for batch jobs that exceed 15 minutes.

Common Gotchas

Lambda timeout max is 15 minutes—strict hard limit. Don’t use Lambda for video processing, heavy ML training, or long batch jobs. Use Fargate, Batch, or EC2 instead.

AWS Elastic Beanstalk

Elastic Beanstalk is a Platform as a Service (PaaS)—you upload code, AWS manages infrastructure (EC2, load balancing, auto-scaling, monitoring).

How It Works

Upload your application code, choose a platform (Java, Python, Node.js, Go, Docker, etc.), and Beanstalk handles the rest. It provisions EC2 instances, applies auto-scaling policies, configures ELBs, and manages monitoring.

Environments are all resources Beanstalk creates: EC2 instances, Auto Scaling Group, Elastic Load Balancer, and RDS (optional).

Common Gotchas

Elastic Beanstalk is NOT serverless—it runs on EC2 under the hood. You’re still paying for those instances even if traffic drops to zero. For true serverless, use Lambda or Fargate.

Configuration & Customization

.ebextensions YAML files (placed in .ebextensions/ at app root) customize Beanstalk environments. Common configs: install packages, set environment variables, tune OS parameters, or deploy custom code.

Beanstalk handles deployment orchestration, health monitoring, and rollback if deployment fails. It integrates with CodePipeline for CI/CD automation.

AWS Serverless Application Model (SAM)

SAM is a framework and shorthand syntax for defining serverless applications. It’s built on CloudFormation.

SAM Templates & Transform

A SAM template is a YAML file using simplified syntax for Lambda functions, API Gateway APIs, DynamoDB tables, and more. At deploy time, SAM transforms the template into standard CloudFormation.

Interview Angle

SAM deployment pipeline: SAM template → sam build (builds layers, packages function) → sam deploy (generates CloudFormation template + deploys stack). Know this progression for the exam—interviewers often ask “what happens when you run sam deploy?”

Step	Command	What Happens
Build	sam build	Compiles, installs deps, creates artifact in .aws-sam/
Package	sam package	Creates S3 bucket for artifacts (optional, implicit in deploy)
Deploy	sam deploy	Transforms to CloudFormation, deploys stack, creates resources

Local Testing

SAM CLI allows local testing and debugging without deploying to AWS. sam local start-api runs a local API Gateway mock. Requires Docker to run Lambda containers locally.

Common Gotchas

SAM local testing requires Docker—you must have Docker running and installed. Missing this causes “samcli: command not found” or runtime errors during local invocation.

Best Practices

SAM enforces least privilege IAM policies, secure environment variable handling, and Infrastructure as Code principles. Every resource gets a logical ID, making it easy to reference and update.

Key Concepts & Best Practices

Elasticity & Scalability: Design to leverage AWS’s elastic capacity. Use Auto Scaling, Lambda concurrency controls, and load balancers to handle variable traffic.

Security: Always use IAM roles (never raw access keys on EC2), enable Security Groups, encrypt EBS volumes, and implement least-privilege policies.

Cost Optimization: Use Spot Instances for fault-tolerant workloads, Reserved Instances for predictable baseline load, and Savings Plans for compute flexibility.

Monitoring & Logging: CloudWatch provides metrics, alarms, and logs. Set up dashboards and alarms for critical thresholds to catch issues early.

Infrastructure as Code: Use CloudFormation, SAM, or Terraform to codify and version your infrastructure. This enables reproducibility and disaster recovery.

Common Gotchas

Summary of exam traps:

• Lambda: 15-minute timeout is a hard limit—no workarounds
• EBS: Volumes are AZ-specific—can’t move between zones without snapshots
• Beanstalk: Uses EC2 under the hood—not serverless, always has baseline cost
• SAM: Requires Docker for local testing—easy to miss in lab scenarios
• EC2 security: Never use access keys—always use IAM instance roles