SOA-C03 · CHEAT SHEET
AWS Certified CloudOps Engineer - Associate
Key concepts grouped by official exam weight.
Download as PDF (free, no sign-up)1. Monitoring, Logging, Analysis, Remediation, and Performance Optimization
22–22%Implement metrics, alarms, and filters by using AWS monitoring and logging services
- Each workload has its tool: Lambda Insights gives Lambda system metrics and cold starts, Bedrock model invocation logging stores requests and responses, and Amazon Managed Service for Prometheus managed collectors scrape EKS metrics without agents. CloudTrail does not log data events (such as S3 DeleteObject) unless you turn them on, and metric filters turn log events into metrics you can alarm on.
- The CloudWatch agent collects in-guest metrics (memory, disk) and logs, using a role with CloudWatchAgentServerPolicy. For fleets, the configuration is stored in Parameter Store and applied with AmazonCloudWatch-ManageAgent; after editing it you must run fetch-config. On ECS with EC2 it runs as a daemon service, and on EKS through the amazon-cloudwatch-observability add-on with EKS Pod Identity.
- Alarms can notify SNS, invoke Lambda (with a resource policy for lambda.alarms.cloudwatch.amazonaws.com), run EC2 and Auto Scaling actions and create OpsItems; other targets use their state change events in EventBridge. Composite alarms combine states with AND, OR and NOT, cannot trigger Auto Scaling and support suppressor alarms. TreatMissingData decides whether missing data counts as OK or ALARM.
- CloudWatch dashboards can show metrics and alarms from several accounts and Regions (accountId and region properties) when each account shares its data with the monitoring account. For cross-account logs and alarms within a Region, use cross-account observability. Variables let one dashboard switch, for example, between Lambda functions, and a dashboard can be shared by email, publicly or through SSO.
- Many services publish to SNS: RDS event subscriptions, Auto Scaling notifications (EC2_INSTANCE_LAUNCH_ERROR), S3 event notifications and the Amazon Managed Service for Prometheus alert manager. The topic needs an access policy that lets the service principal publish and, when encrypted, a customer managed key that grants it kms:GenerateDataKey* and kms:Decrypt.
Identify and remediate issues by using monitoring and availability metrics
- CloudWatch anomaly detection compares a metric with its expected band based on seasonality. EC2 alarm actions reboot an instance on instance status check failures and recover it only with StatusCheckFailed_System. AWS Config fixes resources with Automation runbooks, and EventBridge can start a runbook when its required parameters are set. AWS DevOps Agent investigates incidents and proposes mitigations.
- EventBridge routes events with rules per bus; CloudTrail events reach only the default bus, global service events such as IAM only us-east-1, and read-only ones need the ENABLED_WITH_ALL_CLOUDTRAIL_MANAGEMENT_EVENTS state. Input transformers reshape the event, Pipes enriches it by calling API Gateway, Lambda or Step Functions, and cross-account delivery needs a bus policy and an IAM role.
- Systems Manager Automation runs predefined runbooks (such as AWS-RestartEC2Instance) or custom ones, with tag targets and rate controls (concurrency and error threshold). aws:executeScript runs Python or PowerShell and aws:executeAwsApi calls APIs with automatic retries; aws:approve requests approval through SNS. It can be started from the CLI or SDKs, EventBridge, AWS Config, or across accounts and Regions.
Implement performance optimization strategies for compute, storage, and database resources
- AWS Compute Optimizer analyzes CloudWatch metrics and specifications of EC2, Auto Scaling groups, Lambda, EBS or ECS on Fargate to find under- or over-provisioned resources. Memory requires the CloudWatch agent with the InstanceId dimension, and enhanced metrics extend the lookback period. Recommendations can be filtered by tags; on T instances, a CPUCreditBalance of zero in standard mode caps CPU at baseline.
- SSD volumes (gp3, gp2, io2) suit IOPS-bound transactional workloads and HDD volumes (st1, sc1) suit large sequential throughput workloads; gp3 has no bursting and provisions IOPS and throughput separately. Elastic Volumes changes type and performance without stopping the instance. Key metrics: BurstBalance, VolumeQueueLength and the IOPS and throughput exceeded checks; fast snapshot restore avoids first-access latency.
- S3 Transfer Acceleration uses edge locations for long-distance uploads through the s3-accelerate endpoint; multipart upload sends parts in parallel and retries only failed ones. DataSync moves data from NFS, SMB or other clouds to S3 with integrity validation and bandwidth limits. S3 Lifecycle transitions or expires objects and aborts incomplete uploads; Intelligent-Tiering adapts to unpredictable access.
- Amazon EFS provides shared NFS storage with Elastic, Provisioned or Bursting throughput, and lifecycle policies that move files to IA and Archive and can return them to Standard on first access. FSx for Windows File Server provides SMB with Active Directory, FSx for Lustre gives high-performance POSIX storage linked to S3, and S3 Files mounts a bucket as a synchronized file system.
- CloudWatch Database Insights measures DB load in average active sessions and breaks it down by SQL, waits, hosts or users; with paid retention it produces proactive recommendations. Enhanced Monitoring shows OS processes. RDS Proxy pools and reuses connections (useful with Lambda) and attaches only to the writer instance. Read replicas offload reads, and ReplicaLag measures their delay.
2. Reliability and Business Continuity
22–22%Implement scalability and elasticity
- Amazon EC2 Auto Scaling offers target tracking (on metrics that change with the number of instances, such as ALBRequestCountPerTarget or SQS backlog per instance), step scaling, scheduled actions and predictive scaling for cyclical traffic. Warm pools and instance warmup soften long boot times, and lifecycle hooks let you act before an instance launches or terminates.
- In CloudFront, the cache policy defines the cache key and TTLs, and the origin request policy sends data to the origin without changing that key; Origin Shield adds a layer that consolidates requests. ElastiCache uses strategies such as lazy loading, write-through and TTL, and scales replicas or shards with Application Auto Scaling; DAX caches eventually consistent DynamoDB reads.
- RDS grows storage automatically up to a maximum threshold and scales reads with read replicas; RDS Proxy pools connections during surges. Aurora Auto Scaling adds or removes replicas (it does not act on the writer) and Aurora Serverless scales the compute of each instance within a range. DynamoDB offers on-demand mode or provisioned mode with auto scaling for tables and global secondary indexes.
Implement highly available and resilient environments
- An Application Load Balancer sends traffic only to targets that pass its health check; reason codes (Target.Timeout, Target.ResponseCodeMismatch) guide troubleshooting, and security groups and network ACLs must allow the health check port. Route 53 checks endpoints, other health checks (calculated) or CloudWatch alarms, and Evaluate Target Health makes alias records honor health.
- An RDS Multi-AZ DB instance deployment keeps a synchronous standby that does not serve reads; a Multi-AZ DB cluster adds two readable readers. Aurora promotes replicas in other zones and its cluster endpoint follows the primary. ElastiCache Multi-AZ needs replicas in another zone, Regional EFS stores data across zones, and ARC zonal autoshift moves traffic away from an impaired zone.
Implement backup and restore strategies
- AWS Backup centralizes backups of EC2, EBS, RDS, DynamoDB, EFS, FSx, or S3 with plans that set schedule, retention, and copies to other Regions or accounts, and it assigns resources by tags as long as the service is opted in. Vault Lock in compliance mode makes backups immutable, and Data Lifecycle Manager automates only EBS snapshots and AMIs.
- Point-in-time restore for RDS, Aurora, and DynamoDB always creates a new instance, cluster, or table, with parameter groups and settings to review afterward. Aurora MySQL backtracking rewinds the same cluster in minutes, an Aurora clone gives a fast, space-efficient copy, and cross-Region replication of RDS automated backups allows point-in-time restore in another Region.
- With S3 Versioning, an overwrite creates a new version and a simple DELETE adds a delete marker; a versioned bucket can only be suspended, and NoncurrentVersionExpiration rules clean up old versions. On FSx for Windows File Server, shadow copies must be turned on, and on FSx for ONTAP users recover files from the .snapshot directory.
- Backup and restore needs infrastructure as code to meet the RTO; pilot light keeps data replicated but servers must be switched on and scaled out; warm standby only needs scaling up; active/active serves traffic from several Regions. For failover, Route 53 and Global Accelerator use health checks, and ARC adds routing controls, zonal shift, and Region switch.
3. Deployment, Provisioning, and Automation
22–22%Provision and maintain cloud resources
- EC2 Image Builder automates AMIs and container images with scheduled pipelines, recipes, build and test components, infrastructure configuration, and distribution to Regions, accounts, and launch templates; its lifecycle policies retire old versions. In EC2 an AMI can be deprecated or protected from deregistration, and ECR offers immutable tags and scan on push.
- Change sets show before execution whether a resource will be modified or replaced; termination protection blocks stack deletion, a stack policy protects resources during updates and DeletionPolicy decides what survives deletion. In the AWS CDK, cdk diff compares templates and cdk drift finds changes made outside CloudFormation.
- Deployment failures are diagnosed from the stack event status reason: UPDATE_ROLLBACK_FAILED is fixed by correcting the cause (for example, sending signals with signal-resource) and continuing the rollback, and custom IAM names require CAPABILITY_NAMED_IAM. Every subnet reserves five addresses, and with awsvpc or EKS each task or Pod uses an IP, so small subnets run out.
- StackSets deploys one template to many accounts and Regions: service-managed permissions use AWS Organizations, create the roles and support automatic deployments but skip the management account; self-managed permissions need administration and execution roles. AWS RAM is Regional and shares resources such as subnets or transit gateways without invitations inside the organization.
- Deployment strategies range from in-place or rolling updates to blue/green, canary and linear. In ECS, minimumHealthyPercent and maximumPercent set how many tasks stop or start and the circuit breaker rolls back when tasks fail to start; in Auto Scaling, instance refresh supports checkpoints and auto rollback on alarms; CodeDeploy and weighted aliases shift Lambda traffic.
Automate the management of existing resources
- Systems Manager automates operations: State Manager keeps a defined state and covers new nodes, Maintenance Windows runs prioritized tasks within a schedule, Run Command acts at scale with max-concurrency and max-errors, and Automation runs runbooks with an assume role, approvals, alarm monitoring and multi-account, multi-Region execution. Quick Setup patch policies cover the whole organization.
- S3 Event Notifications send events to SNS, standard SQS, Lambda or EventBridge; destinations need a policy that lets S3 publish, and filters cannot overlap. EventBridge rules filter with patterns, transform input, use dead-letter queues and receive events from other accounts. AWS DevOps Agent investigates incidents automatically and proposes the root cause and mitigation.
4. Security and Compliance
16–16%Implement and manage security and compliance tools and policies
- The password policy enforces new length rules at the next password change, but expiration immediately. aws:MultiFactorAuthPresent exists only with temporary credentials (GetSessionToken or AssumeRole). Cross-account roles avoid duplicate users, sts:ExternalId prevents the confused deputy problem, aws:PrincipalOrgID covers a whole organization, and IAM Identity Center federates with permission sets.
- CloudTrail event history shows recent management events without a trail. Access denied messages name the policy type that denied the request, and decode-authorization-message decodes encoded messages. The policy simulator evaluates SCPs but not RCPs, and IAM Access Analyzer finds external access per Region, generates policies from CloudTrail and checks whether a change grants new access.
- AWS Organizations SCPs set the maximum permissions of member accounts and never grant permissions; a Deny at the root or an OU is inherited downwards and does not affect the management account. Region restrictions use aws:RequestedRegion with NotAction for global services. IAM Identity Center assigns permission sets to groups across many accounts and supports a delegated administrator in a member account.
- Trusted Advisor security checks flag risks such as ports open to 0.0.0.0/0, public EBS snapshots, exposed access keys or a root user without MFA in red or yellow, and each check gives a recommended action. Its events reach EventBridge in us-east-1 and can start automatic remediation, and the organizational view aggregates the results of all accounts.
- A conformance pack bundles AWS Config rules and remediation actions that are deployed as one entity in an account or across the whole organization; it requires Config recording and provides a compliance score. Aggregators give a read-only view of many accounts and Regions, and Control Tower restricts Regions with its Region deny controls for the landing zone or per OU.
Implement strategies to protect data and infrastructure
- A classification scheme defines sensitivity levels and the controls for each one; it is applied with standardized tags through AWS Organizations tag policies and with accounts separated by level. Amazon Macie finds sensitive data in S3 through automated discovery or jobs, managed or custom data identifiers and allow lists, and publishes its findings to EventBridge to automate the response.
- Encryption at rest relies on AWS KMS keys: EBS encryption by default is enabled per Region, an unencrypted RDS instance is encrypted by restoring an encrypted copy of its snapshot, and S3 Bucket Keys reduce SSE-KMS requests to KMS for new objects. Key policies and conditions such as kms:ViaService control usage, and deleting a key requires a waiting period.
- ACM issues public certificates and automatically renews DNS-validated ones while they are in use and the CNAME record remains; imported certificates are not renewed, and CloudFront needs the certificate in us-east-1. Encryption in transit is enforced with HTTPS or TLS listeners, an NLB TCP listener when targets must decrypt, aws:SecureTransport on S3 and rds.force_ssl on RDS.
- AWS Secrets Manager stores and rotates secrets: RDS and Aurora offer managed rotation without Lambda for the master password, and other secrets use a rotation function that must reach the service, for example through an interface VPC endpoint. Secrets replicate across Regions, cross-account access requires a customer managed key, and ECS injects values when each task starts.
- AWS Config remediates noncompliant resources with Systems Manager Automation documents, manually or automatically. GuardDuty publishes findings to EventBridge and its suppression rules archive them without sending them to Security Hub CSPM; Security Hub CSPM automation rules change severity or status at ingestion, and Amazon Inspector exports CSV or JSON reports to S3 encrypted with a KMS key.
5. Networking and Content Delivery
18–18%Implement and optimize networking features and connectivity
- Security groups are stateful and support only allow rules; network ACLs are stateless, are evaluated in ascending rule order and need rules for the response's ephemeral ports. A subnet is public when its route table points to the internet gateway; a public NAT gateway gives private subnets outbound IPv4, and an egress-only internet gateway does the same for IPv6.
- VPC peering is one to one, not transitive, and needs routes in both VPCs. Gateway endpoints (S3 and DynamoDB only) add prefix-list routes; interface endpoints create network interfaces whose security group must allow the traffic. With PrivateLink, the provider publishes a service behind a Network Load Balancer, allows specific principals and accepts connection requests.
- DNS Firewall lets you test rules with the Alert action, publishes metrics in AWS/Route53Resolver and fails closed unless fail open is enabled. AWS WAF logs go to destinations whose names start with aws-waf-logs-; Shield Advanced publishes DDoSDetected in AWS/DDoSProtection (CloudFront in us-east-1), and Network Firewall logs only traffic that reaches its stateful engine.
- Gateway endpoints for S3 and DynamoDB have no additional charge and take that traffic off the NAT gateway, but work only in their Region and not from on premises; interface endpoints have hourly and per-GB fees. A NAT gateway per zone avoids cross-zone transfer, CloudFront cuts data transfer with edge caching, and IPAM Public IP insights helps release unused Elastic IPs.
Configure domains, DNS services, and content delivery
- Route 53 VPC Resolver answers at the VPC plus two address; inbound endpoints receive queries from the on-premises network, and outbound endpoints, with per-domain forwarding rules, send queries to it. A forwarding rule takes precedence over a private hosted zone for the same domain, a system rule excludes subdomains, and private zones require enableDnsHostnames and enableDnsSupport.
- Route 53 offers failover (active-passive), weighted (zero-weight records are used only when the others fail), latency, geolocation (with a default record for unmapped locations) and multivalue answer routing with health checks. Public query logging sends logs to CloudWatch Logs in us-east-1, and Resolver query logging records queries from VPCs.
- CloudFront distributes HTTP content with edge caching: it protects S3 origins with origin access control (OAC) and fails over to a secondary origin only for GET, HEAD and OPTIONS. Global Accelerator provides static anycast IP addresses for TCP and UDP, per-Region traffic dials, per-endpoint weights and client affinity by source IP.
Troubleshoot network connectivity issues
- Network ACL rules are evaluated from the lowest number, and security groups support only allow rules. A public NAT gateway needs a subnet with a route to the internet gateway and times out connections idle for 350 seconds. After attaching a VPC to a transit gateway you must add subnet routes; Reachability Analyzer names the blocking component.
- An ACCEPT followed by a REJECT in VPC flow logs points to a stateless network ACL, and pkt-dstaddr shows the real destination behind a NAT gateway. In ALB access logs, a 504 with target_processing_time -1 means the target did not answer in time; in AWS WAF logs, terminatingRuleId names the blocking rule. On Fargate, container logs need awslogs and the task execution role.
- The cache policy defines the cache key: include only the query strings, cookies and headers that change the response, and use an origin request policy to send values without splitting the cache. With a minimum TTL above 0, CloudFront ignores no-cache and no-store. Versioned file names are recommended for files that change often; invalidations are case sensitive and must cover query strings.
- With Site-to-Site VPN, the route table needs a route to the virtual private gateway, which propagation adds automatically, and a static route with the same CIDR takes priority. Gateway endpoints do not serve on-premises traffic and do not support aws:SourceIp; interface endpoint private DNS needs DNS hostnames and resolution in the VPC, and Route 53 VPC Resolver forwards on-premises domains through an outbound endpoint.
- CloudWatch offers Network Flow Monitor (agents that measure TCP RTT and retransmissions with an AWS network health indicator), Internet Monitor (performance per city-network and events in EventBridge) and Network Synthetic Monitor (agentless probes to on-premises destinations). Metrics also include VPN TunnelState, NAT ErrorPortAllocation, transit gateway PacketDropCountBlackhole and the agent's ethtool counters.