roadmap
- 01
- 02
- 03
- 04
- 05
- 06
- 07
- 08
Handle the AWS questions cloud and backend interviews raise: what the core services do, when to reach for each, and how they fit into a working architecture.
473 of 473 questions in your plan: 0 known, 0 in progress, 0 due, 0 skipped, 473 new. 0 cut, outside your plan.
For a backend candidate, AWS is vocabulary to speak fluently rather than a subject scheduled separately. It surfaces inside the system design conversation, when the interviewer asks where the queue lives, what sits in front of the service and how it reads its database password. Where the role is cloud, DevOps, SRE or platform, expect a round built around it: often an architecture you draw, then an incident staged on that drawing. Data engineering loops pull it toward S3, Kinesis and Redshift. Naming a service is only the opening move; the interviewer then pushes on the choice.
The pushing has specific targets. A role with administrator access still gets an access-denied error: was it the service control policy, the permissions boundary or a resource policy, given that an explicit deny anywhere overrides every allow? An instance in a private subnet cannot download a package: what is missing from its route table, and why does a stateless network ACL drop return traffic that a stateful security group would have let through? An SQS consumer charges the same order twice: what does at-least-once delivery have to do with a visibility timeout shorter than the work? A Lambda function scales out under a spike, and the relational database behind it runs out of connections.
Early-career candidates are mostly asked to tell neighbours apart: a role from a user, a public subnet from a private one, a standard queue from a FIFO queue. Once the level is middle or above, a question tends to arrive as a symptom or a bill, such as a NAT gateway charge larger than the instances it serves, or cross-AZ traffic nobody budgeted for. The answer that holds up names the metric or log it would read first, the change it would make and what that change gives up.
IAM opens the plan because nothing on AWS happens outside an authorized API call, and every later module assumes you can say who made a call and why it was allowed. The module runs from the policy document to how policies are evaluated against one another, then to roles and temporary credentials, how workloads and people actually act, and widens to workforce and customer identity and to many accounts under organization guardrails. Protecting data follows, from KMS and envelope encryption to secrets storage, and audit and detection close the module. Networking is the other half of the foundation. The VPC decides what can reach what, and the rest of the module is the front doors traffic comes in through, from DNS to the load balancers.
Compute comes once identity and network are in place, since every instance, function and task runs under a role, and most of them inside a subnet. EC2 and Lambda come first as the two ends of the range, and choosing a compute model sits between them and the container services, because the question is more often why ECS rather than Lambda than how ECS works. EKS appears here as one entry on that menu, ahead of the managed application platforms.
Data Services pairs storage and the managed databases with Redshift, which gets a module of its own because a warehouse is designed differently from everything before it: distribution and sort keys decide whether a query reads its blocks locally or moves a table across the cluster. Messaging then connects the pieces already built: after outbound email come queues such as SQS, fan-out and event routing, streams through Kinesis, and Step Functions for workflows that have to remember where they are.
The plan ends on running what you assembled. CloudWatch and X-Ray tracing show what the system is doing, CloudTrail returns as the record of API activity, and cost management leads into the Well-Architected reviews that close the plan, since a bill is best read against an architecture you can already draw.
Which modules carry your preparation depends on the loop. A backend developer meeting AWS inside a design round gets most from the first four modules and messaging: be able to draw one request from DNS to a database and say which role each hop runs under. For that reader, cutting importance back to must know, with study order left on by frequency where the page opens, is a sound first pass. A cloud, DevOps or platform candidate should take every module, in by topics order, giving the extra time to policy evaluation, organization guardrails and cost management, where senior incident questions tend to start. A data engineer can begin at Data Services and the Kinesis streams and Firehose, then come back to IAM before the loop, because cross-account access to an S3 bucket is a common follow-up. The plan describes the services and how they behave, not how to declare them: it has no module on CloudFormation, the CDK or Terraform, and cloud rounds often end on how you would provision what you drew. The [Terraform roadmap](/roadmaps/terraform) covers that half. EKS appears only as a choice among the compute services; running workloads inside the cluster is the subject of the [Kubernetes roadmap](/roadmaps/kubernetes).
Which modules carry your preparation depends on the loop. A backend developer meeting AWS inside a design round gets most from the first four modules and messaging: be able to draw one request from DNS to a database and say which role each hop runs under. For that reader, cutting importance back to must know, with study order left on by frequency where the page opens, is a sound first pass. A cloud, DevOps or platform candidate should take every module, in by topics order, giving the extra time to policy evaluation, organization guardrails and cost management, where senior incident questions tend to start. A data engineer can begin at Data Services and the Kinesis streams and Firehose, then come back to IAM before the loop, because cross-account access to an S3 bucket is a common follow-up.
The plan describes the services and how they behave, not how to declare them: it has no module on CloudFormation, the CDK or Terraform, and cloud rounds often end on how you would provision what you drew. The Terraform roadmap covers that half. EKS appears only as a choice among the compute services; running workloads inside the cluster is the subject of the Kubernetes roadmap.