cert-bank

Guides

이 글을 한국어로 읽기

SOA-C03 Study Guide: Detect, Analyze, Act. The Verb in the Question Picks the Service

The AWS Certified CloudOps Engineer – Associate exam (SOA-C03) is about keeping systems running, not designing them. A question describes something that is happening in an account, and asks you to notice it, find out why, or fix it, ideally so that it fixes itself next time. Those three verbs map to three groups of services. If you classify the question first and only then read the options, half of them drop away. This guide is built on that loop: detect, analyze, act, and then prevent.

Study guide · AWS Certified CloudOps Engineer – Associate (SOA-C03)

SOA-C03 Study Guide: Detect, Analyze, Act. The Verb in the Question Picks the Service

The exam

65 questions: 50 scored and 15 unscored, which are not marked; scaled score from 100 to 1,000, passing score 720. Every question is multiple choice (one correct answer out of four options) or multiple response (two or more correct answers out of five or more); the SOA-C03 exam guide lists no other question types, so there are no exam labs. Five domains with unusually even weights: Monitoring, Logging, Analysis, Remediation, and Performance Optimization (22%); Reliability and Business Continuity (22%); Deployment, Provisioning, and Automation (22%); Networking and Content Delivery (18%); Security and Compliance (16%). Because no domain dominates, there is no safe area to skip.

In the 395 SOA-C03 questions in this question bank, 55 stems (about 14%) explicitly ask for something to happen automatically, and 81 questions (21%) are troubleshooting scenarios. The service tagged most often is CloudWatch (38 questions, about one in ten), followed by Systems Manager, S3 and VPC (30 each), EC2 (29) and CloudFormation (28). Expect "what is the least manual way" far more than "which service is this".

SOA-C03: the operations loop this guide is built on
1
Detect
Classify the signal: CloudWatch alarm, metric filter, EventBridge rule, Config rule, CloudTrail
2
Analyze
Logs Insights in place; CloudTrail for who, Config for before and after; Flow Logs for traffic
3
Act
Sized to the fix: EC2 action on the alarm, Run Command, Automation runbook, Config remediation
4
Prevent
CloudFormation change sets and drift detection, StackSets, State Manager, Patch Manager
Place the question at one step of the loop first; only then read the options.

Step 1: classify the signal

Most monitoring questions come down to what kind of signal is described. Each signal type has its own detector:

The stem mentionsDetector
A number crossing a threshold (CPU, latency, queue depth)CloudWatch alarm; a composite alarm to combine several and cut noise
A specific line appearing in logsMetric filter (to alarm on it) or subscription filter (to stream it to Lambda, Kinesis or Firehose)
A discrete event: an instance stopped, a finding, a state changeEventBridge rule
A resource whose settings break a policyAWS Config rule
Who did it, and whenCloudTrail (event history, or a trail queried at scale)
Is this the right size?Compute Optimizer or Trusted Advisor: advisory, not real-time

Step 2: analyze

Analysis questions ask where to look. CloudWatch Logs Insights queries one or many log groups in place, with no export. CloudTrail answers who changed what; Config answers what a resource looked like before and after. For networks, VPC Flow Logs record what actually happened to traffic (ACCEPT or REJECT), Reachability Analyzer checks whether the configuration allows a path without sending a packet, and Traffic Mirroring copies real packets when payloads must be inspected. Work through connectivity in layers: security groups, then network ACLs, then route tables, then Flow Logs to confirm.

Step 3: act

The remediation tool depends on how big the fix is:

  • One action on one instance (stop, reboot, recover): an EC2 action on the CloudWatch alarm itself.
  • A command across a fleet: Systems Manager Run Command.
  • Several steps with conditions or approvals: a Systems Manager Automation runbook, often started by an EventBridge rule.
  • Configuration drift: a Config rule with an attached remediation action.
  • Interactive access to fix something by hand: Session Manager, with no open SSH port and no bastion host.

Beginner tip: Systems Manager questions are hard to picture until you have seen a managed instance. Launch one small Amazon Linux instance with an instance profile that includes the AmazonSSMManagedInstanceCore policy, in a subnet that can reach the Systems Manager endpoints, and with no inbound rules at all. Open a shell with Session Manager, then run a Run Command against it. Then remove the instance profile and watch the node's status in Fleet Manager change to Connection lost (unless Default Host Management Configuration is turned on in the account). You will have met the three prerequisites (agent, role, network path) that troubleshooting questions ask about.

Step 4: prevent it next time

The Deployment, Provisioning, and Automation domain is about making the right state the default. CloudFormation with change sets previews updates; drift detection finds manual changes; StackSets roll the same stack into many accounts and Regions. State Manager keeps re-applying a configuration on a schedule, which is different from Run Command's one-off execution. Patch Manager applies a patch baseline inside a maintenance window and reports compliance. Golden AMIs built with EC2 Image Builder and launch templates keep new instances consistent.

Reliability: availability is not recoverability

Keep two questions apart. Will it stay up? That is Multi-AZ, load balancer health checks, Auto Scaling health checks and scaling policies. Can we get it back? That is backups, point-in-time recovery and a disaster recovery pattern sized to the RTO and RPO. RDS Multi-AZ fails over automatically; a read replica scales reads and helps with regional recovery but must be promoted. A point-in-time restore creates a new instance with a new endpoint, so the runbook needs a step to repoint the application. AWS Backup centralizes plans across services.

Going deeper: why Vault Lock in compliance mode stops even an administrator

Ordinary protection against deleting backups is an IAM decision, and whoever can change IAM can undo it; an attacker with administrator access or a careless broad policy gets through. AWS Backup Vault Lock moves the rule into the vault itself. In governance mode, users with a specific permission can still change or remove the lock, which is useful for testing. In compliance mode, once the grace period after locking has passed, the lock becomes immutable: no user, including the root user, can delete recovery points before their retention ends or shorten the retention. That is why "must survive a compromised administrator account" or "ransomware" in a stem points to compliance mode, and "we want to test the policy first" points to governance mode.

Networking: the operations view

This domain is about running networks, not designing them: public and private subnets, NAT gateways for outbound-only access, gateway and interface endpoints to keep AWS traffic off the NAT, Route 53 hosted zones and Resolver endpoints for hybrid DNS, health checks with failover routing, and CloudFront with Origin Access Control, cache behaviors and invalidations. Remember that DNS answers are cached for the record's TTL, so lower the TTL before a planned change, not after.

Going deeper: why the zone apex needs an alias record

The DNS standard does not allow a CNAME record to sit next to any other record with the same name. The apex of a zone (example.com itself) must always hold the zone's SOA and NS records, so a CNAME there is impossible in any DNS service. Route 53 alias records solve this inside Route 53: the name server looks up the current addresses of the target (a load balancer, a CloudFront distribution, an S3 website endpoint) and answers with ordinary A or AAAA records. The client never sees an alias, the target's changing IP addresses are followed automatically, and alias queries to AWS resources are not charged. So "point example.com at the load balancer" has one answer: an alias record.

Security and compliance, as an operator

Expect the prevent, detect, correct pattern again. Prevent: least-privilege IAM, permissions boundaries for delegated administrators, encryption enforced by policy and by default (S3 default encryption, EBS encryption by default). Detect: GuardDuty for threats, Security Hub CSPM for posture, IAM Access Analyzer for external access. Correct: Config remediation and EventBridge-driven runbooks. An existing unencrypted EBS volume cannot be encrypted in place: snapshot it, copy the snapshot with encryption, and create a new volume.

Study order

  1. CloudWatch, Logs, EventBridge, CloudTrail and Config: the detect and analyze tools.
  2. Systems Manager in depth: Run Command, Automation, Session Manager, State Manager, Patch Manager.
  3. Auto Scaling, load balancing, RDS resilience, AWS Backup and DR patterns.
  4. CloudFormation, StackSets, AMIs and containers; then VPC, Route 53 and CloudFront; then security tooling; then timed mixed sets.

Practise

Practise in the question bank: every question explains each option, right and wrong. Start with the domain you just studied, then take a timed mock exam.

Domains and weights as listed in the AWS exam guide for SOA-C03. Question types and exam rules change; check the official exam page before you book.