CertifyDeck
All cheat sheets
DOP-C02 · CHEAT SHEET

AWS Certified DevOps Engineer - Professional

Key concepts grouped by official exam weight.

Download as PDF (free, no sign-up)

1. SDLC Automation

22–22%

Implement CI/CD pipelines

  • CodeArtifact recommends one repository per domain holding the external connection to a public registry, with the other repositories using it as an upstream; cross-account access needs both a domain policy and a repository policy. In Amazon ECR, lifecycle policies expire images by rule priority, cross-account replication needs only a registry policy in the destination, and pull through cache rules sync registries such as Docker Hub.
  • CodePipeline V2 pipelines with AWS CodeConnections connections support push triggers (branches, file paths and Git tags) and pull request triggers; filters inside one event type are ANDed and excludes take precedence. The CODEBUILD_CLONE_REF output format hands CodeBuild a full Git clone, and CloudFormation Git sync keeps a stack in sync with its template in the repository.
  • CodeBuild runs the install, pre_build, build and post_build phases of the buildspec; Amazon S3 caching is shared across build hosts and supports dynamic keys, while local caching serves only the same host and is not supported with a VPC. VPC builds use private subnets with a NAT gateway, Docker builds need privileged mode, and exported-variables passes values to later CodePipeline actions.
  • CodeBuild retrieves secrets with the parameter-store and secrets-manager buildspec sections or PARAMETER_STORE and SECRETS_MANAGER variables, never PLAINTEXT. CloudFormation uses secretsmanager dynamic references (other accounts by full ARN) and ssm-secure (same account only). Cross-account access needs a resource policy and a customer managed KMS key, and ECS injects a secret when the task starts.
  • CodeDeploy supports in-place and blue/green deployments with predefined canary, linear and all-at-once configurations for ECS and Lambda, and rolls back automatically on failure or when a CloudWatch alarm fires. ECS has native blue/green deployments with test traffic, lifecycle hooks and bake time; Elastic Beanstalk offers rolling, immutable and traffic-splitting policies, and AWS SAM automates gradual Lambda deployments.

Integrate automated testing into CI/CD pipelines

  • CodeBuild starts builds from GitHub, GitLab and Bitbucket webhooks filtered by pull request events, BASE_REF, HEAD_REF or FILE_PATH, and the pull request build policy can require comment approval by a trusted contributor. CodePipeline V2 supports pull request triggers with the PARALLEL execution mode, and for CodeCommit, EventBridge rules on pull request events plus approval rule templates control testing and merges.
  • Distributed Load Testing on AWS runs JMeter, k6 or Locust load tests in Fargate containers across Regions without servers to manage. AWS FIS runs experiments with actions, targets selected by tags and stop conditions based on CloudWatch alarms; CloudWatch Synthetics canaries measure availability and latency on a schedule, and CodeBuild batch builds split tests across isolated environments.
  • In ECS, a container health check runs a command whose exit code 0 means success and any non-zero code means failure; only health checks defined in the task definition are monitored, and CMD-SHELL enables shell operators. cfn-signal sends the cfn-init exit code to a creation policy with -e $?, and a Lambda action in CodePipeline reports its result with PutJobSuccessResult or PutJobFailureResult.
  • CodeBuild creates test reports (JUnit XML, NUnit, TestNG, Cucumber JSON, Visual Studio TRX) and coverage reports (JaCoCo XML, SimpleCov, Clover, Cobertura, LCOV) from the buildspec reports section. A group name creates a group prefixed with the project name and an ARN reuses an existing one; the role needs BatchPutTestCases and BatchPutCodeCoverages, and raw files can be exported to S3.
  • CodePipeline invokes AWS services for testing through managed actions: the Lambda action must close the job with PutJobSuccessResult or PutJobFailureResult (with a continuationToken for long work), the Step Functions action starts a state machine and, for Standard workflows, returns its output as an artifact, and InspectorScan fails on severity thresholds. CodeBuild publishes test and coverage reports from the reports section.

Build and manage artifacts

  • In CodeArtifact the recommended layout is one repository per domain with the external connection to the public registry and the others using it as an upstream; package origin controls (Publish/Upstream) prevent dependency substitution. In ECR, cross-account replication needs a registry policy in the destination, destination settings come from repository creation templates, and pull through cache syncs registries such as Docker Hub.
  • CodeBuild produces artifacts as defined in the buildspec: artifacts and secondary-artifacts for several outputs, exported-variables to pass values to CodePipeline, and cache with dynamic keys in S3. Batch builds (graph, list, matrix, fanout) use their own batch role. Lambda compute starts faster but does not support Docker, privileged mode, VPC access or caching.
  • EC2 Image Builder automates images with recipes (AMI, or container with a Dockerfile and an ECR repository), build and test components, temporary infrastructure, and distribution to Regions and accounts, including launch template updates. Scheduled pipelines can run only when dependencies change, Amazon Inspector scans the images, and lifecycle policies retire old ones.

Implement deployment strategies for instance, container, and serverless environments

  • Access to artifact repositories combines IAM and resource policies: CodeArtifact needs codeartifact:GetAuthorizationToken and sts:GetServiceBearerToken on the identity, plus a domain and a repository policy for other accounts, which accept aws:PrincipalOrgID. In ECS, the task execution role pulls images from ECR or from private registries with repositoryCredentials. Cross-account artifacts need a customer managed KMS key.
  • The CodeDeploy agent is needed only on the EC2/On-Premises platform, not for ECS or Lambda. It is installed and updated with its Systems Manager Distributor package through State Manager; on-premises instances register with an IAM session ARN and temporary credentials. Its configuration file controls :max_revisions:, :proxy_uri: and :enable_auth_policy:, required with VPC endpoints.
  • Deployment failures are diagnosed from the events and logs of each lifecycle event. In CodeDeploy, ApplicationStop uses the scripts of the last successful revision (--ignore-application-stop-failures unblocks it), long-running processes must be redirected to /dev/null, and DownloadBundle fails on permissions or a bucket in another Region. In ECS, the deployment circuit breaker rolls back and emits EventBridge events.
  • Each platform has its own methods: Elastic Beanstalk offers all at once, rolling, rolling with additional batch, immutable, traffic splitting and blue/green with a CNAME swap; CodeDeploy does in-place or blue/green on EC2 and canary, linear or all-at-once on Lambda and ECS; ECS has native blue/green, linear and canary with bake time; and API Gateway runs canary releases on a stage.

2. Configuration Management and IaC

17–17%

Define cloud infrastructure and reusable components to provision and manage systems throughout their lifecycle

  • AWS SAM templates use the AWS::Serverless-2016-10-31 transform, events that generate the API, and least-privilege policy templates; package uploads local artifacts to S3. In CodePipeline, CHANGE_SET_REPLACE and CHANGE_SET_EXECUTE allow a review before changes apply. Imported exports cannot be modified, and CDK needs bootstrapping with --trust in each target environment.
  • A StackSet deploys the same template across accounts and Regions. With service-managed permissions, StackSets creates the roles and can deploy automatically to new accounts in an OU, but never to the management account; with self-managed permissions you create the administration and execution roles. Concurrency, failure tolerance and Region order control rollout risk.
  • AWS AppConfig changes configuration and feature flags without redeploying code, with validators, gradual rollout and alarm-based rollback. State Manager keeps nodes in a desired state and can run Ansible playbooks or Chef recipes without a server of your own. AWS Config records the configuration history of resources and evaluates their compliance with rules.
  • Service Catalog publishes approved products: a launch constraint sets the role the service assumes, and sharing through AWS Organizations avoids imports. CloudFormation modules are registered as versioned private extensions in each account and Region, and the local default version is used. In the AWS CDK, aspects validate every construct in a scope and can make cdk synth fail.

Deploy automation to create, onboard, and secure AWS accounts in a multi-account or multi-Region environment

  • Control Tower offers several ways to standardize accounts: Account Factory Customization applies Service Catalog blueprints without pipelines of your own, CfCT deploys templates, SCPs and RCPs in step with lifecycle events, and AFT provisions and customizes accounts with Terraform and GitOps. The Organizations CreateAccount API is asynchronous: wait for SUCCEEDED before using the account.
  • To govern existing accounts, registering an OU in Control Tower enrolls all of its accounts; enrolling one account needs the AWSControlTowerExecution role and no existing AWS Config resources, unless auto-enrollment is on. Organizations creates OrganizationAccountAccessRole only in accounts it creates, not in invited ones, and enabling all features requires every invited account to accept the request.
  • SCPs never grant permissions: an action must be allowed at every level from the root to the account, a Deny at any level wins, and they do not affect the management account. RCPs limit external principals' access to resources in member accounts, and aws:PrincipalOrgID avoids listing accounts. Cross-account access needs the role's trust policy and the sts:AssumeRole permission on the caller.
  • Controls scale through delegated administrators: GuardDuty is Regional and is turned on in every account with the ALL preference; Security Hub CSPM central configuration applies policies from a home Region to linked Regions. Organization conformance packs need an AWS Config recorder, aggregators are read-only, and Control Tower proactive controls block resources before they are provisioned.

Design and build automated solutions for complex tasks and large-scale environments

  • Quick Setup patch policies run scans and installs on separate schedules across all accounts and Regions of an organization; predefined baselines cannot be edited, so you copy them to assign compliance levels. Inventory accepts custom metadata, resource data sync centralizes it in S3 for Athena, and AWS Config keeps the inventory history.
  • Step Functions calls almost any AWS API through SDK integrations, with no intermediate Lambda functions; Standard workflows support task token callbacks (.waitForTaskToken) and the workflow type cannot be changed. A Distributed Map processes millions of S3 objects in child workflows, and Retry with BackoffRate and Catch handle errors. For asynchronous Lambda, on-failure destinations keep the event and the response.
  • State Manager keeps nodes in a desired state with associations: an SSM document, targets (tags, resource groups or all nodes) and a schedule; new nodes that get the tag are configured right away. AWS-ApplyAnsiblePlaybooks, AWS-ApplyChefRecipes and AWS-ApplyDSCMofs reuse playbooks, recipes or MOF files, and concurrency and error thresholds control the rollout.
  • Patch Manager applies patch baselines with approval rules and rejected patch lists, and a Quick Setup patch policy covers every account and Region of the organization. Compliance shows patch and association status; EventBridge can remediate with AWS-RefreshAssociation or AWS-RunPatchBaseline, and resource data sync moves the data to S3 for Athena.

3. Resilient Cloud Solutions

15–15%

Implement highly available solutions to meet resilience and business requirements

  • A business impact analysis sets how much data can be lost (RPO) and how long recovery may take (RTO); those targets drive the choice between backup and restore, pilot light, warm standby or multi-site active/active, from lowest to highest cost. AWS Resilience Hub stores the targets in resiliency policies and assesses whether an application meets them, also from the pipeline.
  • A single point of failure is any component that exists in only one instance or one zone: a shared NAT gateway, a cache node without replicas, a Lambda function with one subnet or files on one EBS volume. It is fixed with one resource per zone, replicas with Multi-AZ, subnets in several zones, regional storage such as Amazon EFS and Auto Scaling groups across zones.
  • Several services replicate across Regions as a managed feature: DynamoDB global tables accept reads and writes in every replica, RDS supports cross-Region read replicas that can be promoted to standalone instances, and S3 replicates new objects with versioning enabled and existing ones with Batch Replication. CloudFront fails over to a secondary origin, but only for GET, HEAD and OPTIONS requests.
  • A load balancer only sends traffic to enabled zones, each with its own node. Cross-zone load balancing is always on at the Application Load Balancer level, can be turned off per target group and is off by default on Network Load Balancers. Target group health thresholds remove a zone from DNS, and ELB health checks on the Auto Scaling group replace unhealthy instances.
  • To minimize downtime the mechanism depends on the case: an Aurora Global Database switchover changes Region with no data loss, RDS blue/green deployments upgrade the engine without application changes, Route 53 fails over between Regions with failover records and Evaluate target health, and ARC zonal autoshift moves traffic away from a zone with pre-scaled capacity and practice runs.

Implement solutions that are scalable to meet business requirements

  • Scaling issues usually come from the metric or from boot time: instances with very long boot times use a warm pool, the default instance warmup stops new instances from skewing the metric, and SQS workers scale on backlog per instance built with metric math. In Lambda, reserved concurrency is both a minimum and a maximum, so a low value causes throttling.
  • Predictive scaling gets ahead of daily or weekly patterns, and scheduled actions cover events at a known time and work alongside target tracking within the minimum and maximum. For caching, DAX gives microsecond reads with the DynamoDB API, and the API Gateway stage cache needs cache keys, such as query strings, to keep responses apart.
  • In ECS, an Auto Scaling group capacity provider with managed scaling adds instances for the capacity that tasks request, and Service Auto Scaling adjusts the task count; FARGATE and FARGATE_SPOT are combined with a base and weights. In EKS, the Horizontal Pod Autoscaler needs the Metrics Server, and EKS Auto Mode creates and consolidates nodes without running Karpenter yourself.
  • To scale across Regions, Route 53 latency records send each user to the fastest Region, ECR replication copies new images while repository creation templates apply the settings, S3 Multi-Region Access Points give one global endpoint, and Aurora write forwarding lets secondary Regions write through the primary cluster.
  • In Lambda, provisioned concurrency pre-initializes environments on an alias or version and can be scheduled with Application Auto Scaling; reserved concurrency only reserves and caps, and for SQS the maximum concurrency is set per event source. API Gateway throttles per client with usage plans and API keys and cuts backend calls with a stage cache. ECS on Fargate can scale on backlog per task with metric math.

Implement automated recovery processes to meet RTO and RPO requirements

  • Failover tests are automated with AWS FIS (reboot-db-instances with forceFailover for RDS Multi-AZ, failover-db-cluster for clusters) and stop conditions on alarms. In Aurora Global Database, switchover is for drills with RPO 0. CloudFront fails over between origins only for GET, HEAD and OPTIONS, and ARC routing controls shift traffic between Regions without editing DNS records.
  • AWS Backup copies recovery points to other Regions and accounts with copy rules; Organizations backup policies apply immutable plans to whole OUs, and cross-account copies need shared customer managed keys. Restore testing proves that backups can be restored. In S3, Replication Time Control gives SLA-backed replication time and Batch Replication copies the objects that already exist.
  • An ALB routes only to targets that pass the target group health check and fails open when all targets fail. Auto Scaling replaces instances on ELB failures only when ELB health checks are on, with a suitable grace period. Target group health thresholds drive DNS failover with Route 53 Evaluate Target Health, and Automatic Target Weights shifts traffic away from targets that return 5xx errors.

4. Monitoring and Logging

15–15%

Configure the collection, aggregation, and storage of logs and metrics

  • CloudWatch Logs always encrypts data and supports symmetric KMS keys per log group, with the log group ARN as encryption context in the key policy. Data protection policies mask sensitive data at ingestion, and only logs:Unmask reveals it. In S3, Object Lock in compliance mode blocks deletion even by the root user, and CloudTrail proves log integrity with signed digest files.
  • A metric filter turns log events that match a pattern into a CloudWatch metric with a fixed value or a value taken from the event, and can add low-cardinality dimensions taken from JSON or space-delimited fields. It does not process events from before it was created, a default value avoids spotty metrics but cannot be combined with dimensions, and it works only in Standard class log groups.
  • A metric stream continuously sends CloudWatch metrics to Amazon Data Firehose (and from there to S3, a data lake or partner tools) in JSON or OpenTelemetry format. The Firehose stream must be in the same account and Region; filters are either include or exclude, never both, and additional statistics such as p99 are configured per metric. In a monitoring account a stream can include source accounts.
  • The CloudWatch agent collects in-guest metrics such as memory and disk, process metrics with procstat, and application metrics through StatsD or collectd; it is installed and configured with Systems Manager from a configuration stored in Parameter Store, and aggregation_dimensions rolls metrics up per Auto Scaling group. In Lambda, embedded metric format publishes metrics from logs without PutMetricData or a collector.
  • CloudWatch Logs keeps log events indefinitely unless a log group retention or an account-level retention policy is set; the account policy applies by name prefix and can be retroactive. Continuous archiving uses Firehose subscriptions to S3, with S3 Lifecycle rules that transition objects to Glacier storage classes and expire them; versioned buckets also need NoncurrentVersionExpiration.

Audit, monitor, and analyze logs and metrics to detect issues

  • CloudWatch dashboards can combine data from many accounts through cross-account observability, with a monitoring account linked to the organization, and can be shared with people without AWS accounts by email, public link or SAML SSO through Amazon Cognito. AWS Quick Sight visualizes Athena data once the buckets and KMS key are authorized, and SPICE supports incremental refreshes.
  • A metric alarm watches a standard or custom metric, or the result of a math expression, and runs SNS, EC2 or Auto Scaling actions when its state changes. For custom metrics the alarm must include every dimension. High-resolution metrics support 10- or 30-second alarms, and composite alarms reduce noise but cannot run EC2 or Auto Scaling actions.
  • X-Ray is enabled with Lambda active tracing (with AWSXRayDaemonWriteAccess on the execution role), tracing on API Gateway REST API stages and, for ECS containers, an AWS Distro for OpenTelemetry sidecar with a task role. Sampling is decided by the first traced service, and groups with filter expressions publish metrics to CloudWatch.
  • CloudWatch Logs subscription filters send events in near real time to Kinesis Data Streams, also across accounts through a destination with an access policy. Random distribution helps avoid throttling. Lambda consumers scale with ParallelizationFactor or enhanced fan-out, and Amazon Managed Service for Apache Flink computes continuous aggregations without managing servers.
  • CloudWatch Logs Insights interactively queries existing logs across several log groups, with commands such as filter, stats, pattern and diff. Athena runs SQL over the logs that services deliver to S3, such as CloudTrail, ALB or VPC Flow Logs, and partition projection avoids managing partitions by hand. Metric filters are not retroactive.

Automate monitoring and event management of complex environments

  • DynamoDB scales provisioned capacity with Application Auto Scaling target tracking, with a separate policy for each global secondary index. RDS storage autoscaling grows storage up to a maximum threshold. In ECS, Service Auto Scaling adjusts tasks and a capacity provider with managed scaling adjusts instances according to targetCapacity; a value below 100 percent keeps spare capacity.
  • Metric filters turn CloudWatch Logs events into numeric metrics with case-sensitive patterns; they are not retroactive and a default value cannot be used with dimensions. An alarm on the metric notifies an SNS topic. The embedded metric format lets Lambda publish custom metrics by writing structured logs.
  • S3 Event Notifications deliver events to Lambda, standard SQS, SNS or EventBridge; FIFO queues are reached only through EventBridge. The Lambda function needs a resource-based policy that lets S3 invoke it, and writing to the bucket that triggers it creates a loop. OpenSearch Ingestion indexes new S3 files in near real time by reading an SQS queue.
  • An EventBridge rule sends events that match its pattern to targets such as SNS; pattern values go in arrays and support operators such as prefix, wildcard or anything-but. CloudTrail events need an active trail, arrive only on the default bus, and read-only ones need the ENABLED_WITH_ALL_CLOUDTRAIL_MANAGEMENT_EVENTS rule state. Input transformers format the message.
  • SSM Agent makes EC2 instances managed nodes through an instance profile or Default Host Management Configuration, which requires IMDSv2; on-premises servers register with a hybrid activation. The CloudWatch agent is installed with AWS-ConfigureAWSPackage or with Quick Setup across the organization, needs CloudWatchAgentServerPolicy and, without internet access, interface endpoints.

5. Incident and Event Response

14–14%

Manage event sources to process, notify, and take action in response to events

  • AWS Health, CloudTrail and services such as ECS send events to EventBridge, where rules filter by pattern and route to SNS, Lambda or Automation. Global Health events arrive in N. Virginia; data events must be enabled on the trail, and read-only events need the ENABLED_WITH_ALL_CLOUDTRAIL_MANAGEMENT_EVENTS state. Cross-account routing needs a policy on the receiving bus.
  • Event workflows combine SQS, SNS, Kinesis, Lambda and Step Functions. For SQS, ReportBatchItemFailures avoids reprocessing successes and dead-letter queues with redrive set aside and recover messages; for Kinesis, bisecting and an on-failure destination isolate bad records. SNS with filter policies fans out to several queues, and Step Functions Standard workflows support long waits.

Implement configuration changes in response to events

  • Configuration changes are applied with Run Command (with concurrency and error controls), scheduled State Manager associations, also on on-premises servers registered through a hybrid activation, and Automation across accounts and Regions from a delegated administrator. AWS AppConfig validates data before deployment and rolls out gradually with automatic rollback when an alarm fires.
  • Events can change infrastructure without human action: an alarm or a Parameter Store change triggers EventBridge rules that run runbooks such as AWS-ResizeInstance or update launch templates; instance refresh with auto rollback needs numbered versions. Auto Scaling lifecycle hooks, the ECS deployment circuit breaker and EventBridge Scheduler complete the toolkit.
  • Returning a system to its desired state combines State Manager associations that reapply configuration, EventBridge rules that run Automation runbooks (for example to isolate instances or turn CloudTrail logging back on), OpsCenter OpsItems with associated runbooks, Auto Scaling custom health checks, and Retry and Catch error handling in Step Functions.

Troubleshoot system and application failures

  • In CodeDeploy, ApplicationStop scripts come from the last successful revision, and an agent 'Missing credentials' error points to the instance profile. A stack in UPDATE_ROLLBACK_FAILED cannot be updated: fix the cause (for example with signal-resource) and continue the rollback. CodeBuild and CodePipeline failures often trace to service role permissions, and a canary in a VPC without internet needs endpoints for CloudWatch and S3.
  • Auto Scaling activity history explains failed launches and terminations; a grace period shorter than startup causes replacement loops. In ECS, service events, minimumHealthyPercent and maximumPercent and the circuit breaker with rollback explain stuck deployments, and Fargate in private subnets needs ECR and S3 endpoints. In EKS, the Horizontal Pod Autoscaler needs the Metrics Server.

6. Security and Compliance

17–17%

Implement techniques for identity and access management at scale

  • IAM Access Analyzer generates policies from CloudTrail activity, although it leaves out iam:PassRole and data event actions; its custom policy checks catch new access in a pipeline, and unused access analyzers flag unused roles and permissions across the organization. Permissions boundaries cap the permissions of delegated roles, and last accessed data for an OU helps refine its SCPs.
  • RBAC assigns policies per job function, such as IAM Identity Center permission sets that become roles in each account. ABAC compares principal and resource tags (aws:PrincipalTag, aws:ResourceTag, aws:RequestTag) and scales without policy edits; IdP attributes arrive as session tags, which require sts:TagSession in the role trust policy.
  • Secrets Manager rotates secrets with managed rotation (for example the RDS or Aurora master password with ManageMasterUserPassword) or with a Lambda function; the alternating users strategy keeps valid credentials during rotation. A function in a VPC without internet needs a Secrets Manager endpoint, and IAM Roles Anywhere replaces access keys on on-premises servers with temporary credentials.
  • MFA for API access is enforced with aws:MultiFactorAuthPresent, in the role trust policy for cross-account access and with BoolIfExists to also deny long-term keys; the user passes the device and code to AssumeRole or GetSessionToken. Machine identities use temporary credentials: instance profiles on EC2, task roles in ECS and EKS Pod Identity in Kubernetes.

Apply automation for security controls and data protection

  • Security Hub CSPM central configuration applies standards and controls across accounts and Regions with configuration policies and prevents drift; its automation rules update findings as they arrive. Control Tower has preventive (SCP), detective (Config rules) and proactive (CloudFormation hooks) controls. Organization conformance packs are deployed per Region, and multi-account Automation needs administration and execution roles.
  • Defense in depth combines layers: stateful security groups with allow rules only, stateless network ACLs with subnet-level denies, AWS WAF on ALB, CloudFront or API Gateway with rate-based rules and managed rule groups, and Network Firewall at the VPC perimeter with domain filtering. GuardDuty detects, EventBridge automates the response, Detective investigates, and ACM renews only the certificates it issues.
  • Amazon Macie automates sensitive data discovery in S3 in two ways: automated sensitive data discovery, which continually samples objects across the organization from a delegated administrator, and discovery jobs, periodic or one-time, whose bucket criteria are evaluated at each run. Custom data identifiers and allow lists refine what Macie detects and what it ignores.
  • For encryption in transit, ACM automatically renews Amazon-issued certificates that use DNS validation (imported ones are not renewed), and an aws:SecureTransport condition in the bucket policy rejects HTTP requests. At rest, KMS rotates key material without changing the logical key, a CloudHSM key store adds single-tenant HSMs, and sharing encrypted snapshots needs a customer managed key.

Implement security monitoring and auditing solutions

  • Robust auditing starts with an organization trail, which covers every account and cannot be modified by members, with log file integrity validation through signed digest files and S3 Object Lock in compliance mode to keep logs in WORM storage. AWS Config deploys organization conformance packs from a delegated administrator, and IAM Access Analyzer reports unused access.
  • Security alerting combines CloudTrail Insights, which detects deviations from normal API call or error rates, with EventBridge rules on CloudTrail events and GuardDuty findings, which support numeric matching and need a special state to match read-only events. Count thresholds use metric filters with alarms, and a central event bus can receive events from the whole organization.
  • CloudTrail records management events and, only when configured with advanced event selectors, data events such as S3 object operations. In CloudWatch Logs, destinations and account-level subscription filter policies centralize logs in near real time, a symmetric KMS key associated with the log group encrypts the data, and data protection policies mask sensitive values at ingestion.
  • To analyze logs, Athena with partition projection queries CloudTrail logs in S3 without partition maintenance, and Logs Insights queries log groups over any past period, whereas metric filters are not retroactive. Detective investigates GuardDuty findings with behavior graphs, AWS Security Hub correlates signals into exposure findings, and AWS Config advanced queries against an aggregator span the organization.