Monitoring and identity side by side: Azure Monitor and Entra ID vs AWS
Translate Azure Monitor, Application Insights, KQL and Log Analytics into CloudWatch, Logs Insights and OpenTelemetry, and map Entra ID, Active Directory and managed identities onto IAM Identity Center, Cognito, AWS Managed Microsoft AD and IAM roles, with four identity integration patterns for a bank.
Assumes: .NET 10 LTS · OpenTelemetry .NET · AWS Distro for OpenTelemetry (ADOT) · AWS SDK for .NET v4 · AWS CLI v2 · Region us-east-1. Verified against official sources on 2026-11-18.
Learning objectives
By the end of this lesson you can:
- Map Azure Monitor, Application Insights, Log Analytics and Activity Log onto CloudWatch, Logs Insights, X-Ray or OpenTelemetry, and CloudTrail.
- Translate three common KQL queries into CloudWatch Logs Insights queries.
- Replace the Application Insights SDK with OpenTelemetry in an ASP.NET Core service so telemetry is portable.
- Map Entra ID, Active Directory Domain Services, Conditional Access and managed identities to their AWS counterparts.
- Choose among four identity integration patterns for workforce and customer sign-in, and name the failure mode of each.
Prerequisites
- DevOps, IaC and secrets side by side
- Application Signals and Log aggregation and error monitoring
- IAM Identity Center, MFA and STS, Federation and segregation of duties and Cognito user pools
- Tools: .NET 10 SDK, AWS CLI v2 with
learn-sandbox
Why this matters in banking
Two things decide whether a migrated platform is operable: can the on-call engineer see what is happening, and can the right people (and only them) get in. At Harbourline, on-call engineers have years of KQL muscle memory and a workbook per service. If the first production incident on AWS finds them searching documentation for a query syntax, mean time to recovery doubles. Identity has the same risk with higher stakes: the bank's staff sign in with Entra ID, with Conditional Access rules, privileged role elevation and quarterly access reviews. A migration that leaves AWS access in a spreadsheet of IAM users fails the next access review and, worse, leaves leavers with working credentials.
This lesson shows how to carry observability habits and identity governance across.
The idea in plain language
Monitoring answers "is it healthy, and if not, why?" with three kinds of data: metrics (numbers over time), logs (event records) and traces (the path of one request across services). Identity answers "who is this, and what may they do?" for two populations: workforce (staff) and customers.
- On Azure, Azure Monitor collects metrics and logs, Log Analytics stores logs and runs KQL, and Application Insights adds application telemetry and tracing. Entra ID is the identity provider for staff, and managed identities are the identities of workloads.
- On AWS, CloudWatch holds metrics, logs, alarms and dashboards; X-Ray (backend) and OpenTelemetry carry traces; CloudTrail records control-plane activity. IAM Identity Center gives staff access to accounts; Cognito handles customer sign-in; IAM roles are workload identities.
Analogy: monitoring is the bank's CCTV and alarm panel; identity is the staff pass system. Moving buildings, you do not buy new cameras that only work with one vendor's recorders (use open standards: OpenTelemetry), and you do not issue new passes by hand (federate the existing staff directory).
What problem the AWS services solve
| Service | Problem | Limits and sharp edges |
|---|---|---|
| CloudWatch Metrics and Alarms | Numeric health signals with thresholds and anomaly detection | Custom metrics are billed; high-cardinality dimensions get expensive |
| CloudWatch Logs + Logs Insights | Central log storage with interactive queries | Billed by ingestion and storage; set retention, the default is never expire |
| CloudWatch Application Signals | Service-level SLOs and application maps built from OpenTelemetry data | See the Application Signals lesson |
| X-Ray | Trace storage and service map | The X-Ray SDKs and daemon entered maintenance mode on 2026-02-25; use OpenTelemetry (ADOT) and keep X-Ray as the backend |
| CloudTrail | Who called which AWS API, when | Management events on by default (90-day history); a trail to S3 is needed for long retention |
| IAM Identity Center | Workforce single sign-on and per-account permission sets | One instance per organisation; changing the identity source removes assignments, so export them first |
| Cognito user pools | Customer sign-up, sign-in, MFA, federation | Token shapes differ from Entra ID (see the code section) |
| AWS Managed Microsoft AD | A real AD (Windows Server domain controllers) in AWS for Windows workloads | Hourly cost per directory; a trust to on-premises AD is optional |
| AD Connector | Proxy that forwards sign-ins to on-premises AD | Depends on the connection to on-premises; not a directory itself |
Service availability note. AWS Directory Service Simple AD is in maintenance (announced 2026-06-30); do not plan new designs on it. Use AWS Managed Microsoft AD or AD Connector. ⚠ VERIFY on the AWS service lifecycle page before design sign-off.
Microsoft / Azure comparison
Monitoring
| You know (Azure) | AWS equivalent | What's different |
|---|---|---|
| Azure Monitor metrics | CloudWatch metrics | Namespaces and dimensions instead of resource scopes; standard metrics are free, custom are billed |
| Log Analytics workspace | CloudWatch log groups (and Logs Insights) | No single workspace: log groups per service; cross-account observability links accounts |
| KQL | Logs Insights query language; CloudWatch also supports OpenSearch PPL and SQL in Logs Insights | Three query languages; PPL and SQL were added in December 2024 (docs) |
| Alerts and action groups | CloudWatch alarms with SNS, EventBridge or Systems Manager actions | An alarm has states (OK, ALARM, INSUFFICIENT_DATA); add composite alarms for noisy services |
| Workbooks, dashboards | CloudWatch dashboards, Quick Sight for business reporting | Dashboards are JSON and can be code-managed |
| Application Insights (APM) | Application Signals, X-Ray traces, OpenTelemetry | CloudWatch also has a separate feature named "Application Insights" for detecting problems in common stacks; do not confuse them |
| Availability tests | CloudWatch Synthetics canaries | Canaries are Lambda-based scripts you pay for per run |
| Live metrics, Profiler | CloudWatch Application Signals, CodeGuru Profiler (check status) ⚠ VERIFY | Verify service status before design |
| Activity Log | CloudTrail | Ship to S3 and an organisation trail for evidence |
| Diagnostic settings | Log group settings, subscription filters, Firehose, S3 export | Many AWS services log to CloudWatch Logs or S3 when you enable it |
| Smart detection | Anomaly detection alarms | Train on a baseline of days to weeks |
Three KQL queries in Logs Insights
| Question | KQL | Logs Insights |
|---|---|---|
| Errors per 5 minutes | traces | where severityLevel >= 3 | summarize count() by bin(timestamp, 5m) | filter level = "Error" | stats count(*) by bin(5m) |
| Slowest endpoints | requests | summarize p95 = percentile(duration, 95) by name | top 5 by p95 desc | stats pct(durationMs, 95) as p95 by route | sort p95 desc | limit 5 |
| Find one request | union * | where operation_Id == "abc123" | fields @timestamp, @message | filter traceId = "abc123" | sort @timestamp asc |
Field names such as level and durationMs come from your JSON logs; Logs Insights discovers fields in JSON automatically, which is why the earlier lesson set up JSON console logging.
Identity
| You know (Microsoft) | AWS equivalent | What's different |
|---|---|---|
| Entra ID tenant (workforce) | IAM Identity Center with Entra ID as the external identity provider | Identity Center is the AWS front door; Entra stays the directory |
| Entra groups assigned to subscriptions | Groups assigned to permission sets per account | Assignments are per account and per permission set |
| Conditional Access (device, location, risk) | Enforced in Entra ID before the SAML sign-in; on AWS, IAM conditions and SCPs add network or tag limits | Keep device and location policy in Entra; AWS enforces what the session may do |
| PIM (just-in-time elevation) | Time-boxed permission set assignment through an access request workflow | Identity Center has no built-in approval workflow; many banks add a request tool ⚠ VERIFY current native options |
| Entra External ID / B2C (customers) | Cognito user pools | Different token claims and flows; plan app changes |
| App registration and service principal | Cognito app client; for workloads, an IAM role | Machine-to-machine on AWS means IAM roles, not client secrets |
| System-assigned managed identity | IAM role on EC2, ECS task or Lambda | You create and name the role |
| Active Directory Domain Services | AWS Managed Microsoft AD (or self-managed AD on EC2) | A real domain controller fleet managed by AWS across two AZs |
| Entra Domain Services | AWS Managed Microsoft AD | Similar managed-domain role |
| AD FS | Federation directly from Entra or the IdP to Identity Center or IAM SAML | AD FS is no longer required for AWS sign-in |
| Group Policy | Still GPO on Managed AD for Windows servers | Unchanged for domain-joined EC2 |
| Entra access reviews | Review Identity Center assignments and IAM Access Analyzer findings | Export assignments for the review; Access Analyzer finds unused access |
Architecture
Four identity integration patterns. Most banks use A for staff, B for Windows workloads, C during transition and D for customers.
Security-boundary view. In Pattern A, the strongest control sits in Entra: MFA and Conditional Access happen before AWS ever sees the user. AWS sees only a SAML assertion and issues a short session for a permission set. In Pattern B, domain controllers are in your private subnets, reachable only from your VPCs and through the trust link; the trust direction matters (see mistakes below).
Failure modes. A: SCIM drift leaves a leaver's access in place until the next sync; keep a revocation drill. B: if the link to on-premises drops, accounts present only on-premises cannot authenticate. C: Connector depends entirely on on-premises availability. D: token claim differences break APIs written for Entra tokens.
Step-by-step implementation
Inventory what you observe and who signs in
List dashboards, alerts, KQL queries in use, retention, and who reads them. For identity, list directories, groups, privileged roles, service principals, Conditional Access policies and access review cycles.
Choose the identity pattern
Apply the four patterns. Almost every Azure estate starts with Pattern A for staff; add B if Windows servers or RDS for SQL Server must use Windows authentication.
Connect Entra ID to Identity Center
Follow AWS's guide for SAML and SCIM with Entra ID (guide). Create the enterprise application, exchange metadata, enable automatic provisioning, assign groups, then create permission sets and assign groups to accounts. Check how nested groups and attribute removal behave before relying on them ⚠ VERIFY.
Instrument one service with OpenTelemetry
Replace the Application Insights SDK with OpenTelemetry in a single service, send OTLP to an ADOT collector (a sidecar or an agent), and compare traces with the old ones.
Rebuild alerts as code
Convert each critical Azure alert rule to a CloudWatch alarm in IaC with a documented runbook link and an owner. Set log retention in the same stack.
Parallel run
Keep Azure Monitor alerts on for the same workloads until AWS alarms have fired correctly in game-day tests.
AWS Console instructions
- Logs Insights: Console → CloudWatch → Logs → Logs Insights, select log groups, choose the query language (Logs Insights QL, OpenSearch PPL or SQL) and run.
- Alarm: Console → CloudWatch → Alarms → Create alarm → Select metric, set the threshold, then Add notification to an SNS topic.
- Identity Center: Console → IAM Identity Center → Settings → Identity source → Actions → Change identity source → External identity provider. Warning: changing the identity source (to or from Active Directory in particular) can remove existing assignments, so export them first.
- Directory: Console → Directory Service → Set up directory → AWS Managed Microsoft AD (design-only in the lab: it is billed hourly).
AWS CLI
# Run a Logs Insights query over the last hour (replace the log group). Returns a query ID first.
QID=$(aws logs start-query --log-group-name /hl/m13/rates-api \
--start-time $(date -d '-1 hour' +%s) --end-time $(date +%s) \
--query-string 'filter level = "Error" | stats count(*) by bin(5m)' \
--query queryId --output text --profile learn-sandbox --region us-east-1)
aws logs get-query-results --query-id "$QID" --profile learn-sandbox --region us-east-1
# Set retention so logs do not live (and bill) forever.
aws logs put-retention-policy --log-group-name /hl/m13/rates-api --retention-in-days 30 \
--profile learn-sandbox --region us-east-1
# A CloudWatch alarm on ALB 5xx errors, notifying an SNS topic (replace the dimension and topic ARN).
aws cloudwatch put-metric-alarm --alarm-name hl-m13-5xx --namespace AWS/ApplicationELB \
--metric-name HTTPCode_Target_5XX_Count --dimensions Name=LoadBalancer,Value=app/hl-m13/abc123 \
--statistic Sum --period 60 --evaluation-periods 3 --threshold 5 --comparison-operator GreaterThanThreshold \
--treat-missing-data notBreaching --alarm-actions arn:aws:sns:us-east-1:111122223333:hl-m13-oncall \
--profile learn-sandbox --region us-east-1
# Identity Center: which instance and identity store do I have? (management or delegated admin account)
aws sso-admin list-instances --profile learn-sandbox --region us-east-1The date -d syntax works on GNU date (Git Bash, Linux, CloudShell); on macOS use date -v-1H +%s. The account ID in the alarm ARN is a placeholder.
Infrastructure as Code
using Amazon.CDK;
using Amazon.CDK.AWS.CloudWatch;
using Amazon.CDK.AWS.CloudWatch.Actions;
using Amazon.CDK.AWS.Logs;
using Amazon.CDK.AWS.SNS;
using Constructs;
public sealed class ObservabilityStack : Stack
{
public ObservabilityStack(Construct scope, string id, IStackProps props) : base(scope, id, props)
{
// Finite retention is a cost and privacy control: logs may contain personal data.
var logGroup = new LogGroup(this, "Api", new LogGroupProps
{
LogGroupName = "/hl/m13/rates-api",
Retention = RetentionDays.ONE_MONTH,
RemovalPolicy = RemovalPolicy.DESTROY,
});
// Turn JSON error logs into a metric (the KQL "count errors" habit), then alarm on it.
var errors = new MetricFilter(this, "Errors", new MetricFilterProps
{
LogGroup = logGroup,
MetricNamespace = "Harbourline/RatesApi",
MetricName = "ErrorCount",
FilterPattern = FilterPattern.StringValue("$.level", "=", "Error"),
MetricValue = "1",
});
var topic = new Topic(this, "OnCall");
var alarm = new Alarm(this, "ErrorAlarm", new AlarmProps
{
Metric = errors.Metric(new MetricOptions { Statistic = "Sum", Period = Duration.Minutes(1) }),
Threshold = 5,
EvaluationPeriods = 3,
ComparisonOperator = ComparisonOperator.GREATER_THAN_THRESHOLD,
TreatMissingData = TreatMissingData.NOT_BREACHING,
AlarmDescription = "Rates API error burst. Runbook: wiki/runbooks/rates-api-errors",
});
alarm.AddAlarmAction(new SnsAction(topic));
Amazon.CDK.Tags.Of(this).Add("Project", "cea-lab");
Amazon.CDK.Tags.Of(this).Add("Lab", "m13-monitoring");
Amazon.CDK.Tags.Of(this).Add("Environment", "learn");
}
}Complete code
OpenTelemetry in ASP.NET Core, replacing AddApplicationInsightsTelemetry. Packages: OpenTelemetry.Extensions.Hosting, OpenTelemetry.Instrumentation.AspNetCore, OpenTelemetry.Instrumentation.Http, OpenTelemetry.Exporter.OpenTelemetryProtocol (use the current stable versions).
using OpenTelemetry;
using OpenTelemetry.Resources;
using OpenTelemetry.Trace;
var builder = WebApplication.CreateBuilder(args);
// Before: builder.Services.AddApplicationInsightsTelemetry(connectionString);
// After: OpenTelemetry, exported over OTLP to a collector (ADOT) at OTEL_EXPORTER_OTLP_ENDPOINT.
builder.Services.AddOpenTelemetry()
.ConfigureResource(r => r.AddService("harbourline-rates-api", serviceVersion: "1.0.0"))
.WithTracing(t => t
.AddAspNetCoreInstrumentation()
.AddHttpClientInstrumentation())
.UseOtlpExporter();
builder.Logging.ClearProviders();
builder.Logging.AddJsonConsole(); // fields become queryable in Logs Insights
var app = builder.Build();
app.MapGet("/rates/{from}/{to}", (string from, string to) => Results.Ok(new { from, to, rate = 1.17m }));
app.Run();Validating Cognito tokens in an API that used to validate Entra tokens. This is the claim difference that most often breaks a migration: Cognito access tokens carry client_id, not aud.
using System.Security.Claims;
using Microsoft.AspNetCore.Authentication.JwtBearer;
public static class CognitoAuth
{
public static IServiceCollection AddCognitoJwt(this IServiceCollection services, string region, string userPoolId, string appClientId)
{
services.AddAuthentication(JwtBearerDefaults.AuthenticationScheme)
.AddJwtBearer(options =>
{
// Metadata (and signing keys) are discovered from this issuer.
options.Authority = $"https://cognito-idp.{region}.amazonaws.com/{userPoolId}";
options.TokenValidationParameters.ValidateAudience = false; // access tokens have no "aud"
options.TokenValidationParameters.ValidateIssuer = true;
options.Events = new JwtBearerEvents
{
OnTokenValidated = ctx =>
{
// Entra: validate "aud". Cognito access token: validate "client_id" and "token_use".
var clientId = ctx.Principal?.FindFirstValue("client_id");
var use = ctx.Principal?.FindFirstValue("token_use");
if (clientId != appClientId || use != "access")
{
ctx.Fail("Token was not issued to this application.");
}
return Task.CompletedTask;
},
};
});
return services;
}
}Code explained line by line
AddOpenTelemetry().ConfigureResource(...)names the service; every span carries this name, which is how the service appears on the map.AddAspNetCoreInstrumentationandAddHttpClientInstrumentationcreate spans for incoming requests and outgoing HTTP calls with no code in your handlers.UseOtlpExporter()sends telemetry to the OTLP endpoint named byOTEL_EXPORTER_OTLP_ENDPOINT, normally an ADOT collector sidecar that forwards to CloudWatch and X-Ray. Check how your collector handles trace ID format ⚠ VERIFY.AddJsonConsole()writes structured logs that Logs Insights can query by field.- In
AddCognitoJwt,Authoritymakes the middleware download the user pool's signing keys.ValidateAudience = falseis intentional for Cognito access tokens, which omitaud. OnTokenValidatedreplaces the audience check withclient_idandtoken_use. Without it, a token for any app client in the pool would be accepted.
Security considerations
⚠ Learning shortcut — not for production
The examples send telemetry without redaction and treat any Cognito access token for one client as sufficient. Production adds log redaction (no account numbers, tokens or personal data in logs or span attributes), per-environment log groups with encryption by a customer managed key and finite retention, scope checks on tokens (scope claim per API), MFA enforced in the identity provider, and CloudTrail plus alarms on root and privileged role use.
- Observability data is personal data when it contains identifiers. Treat log retention and access as UK GDPR matters; involve your DPO.
- Least privilege for readers. Giving everyone
CloudWatchLogsFullAccesslets any developer read production customer data in logs. Use read-only permission sets per account. - Trust direction on Managed AD. A one-way trust lets on-premises users access AWS-hosted resources; the opposite direction exposes the AWS domain to on-premises. Choose the minimum and review with your security architects.
- Identity source change is destructive. Test the Entra connection in a sandbox organisation before touching production assignments.
- Revocation drill. Disable a test user in Entra, measure how long until AWS access stops (SCIM sync plus session length), and record it.
Banking use case
Harbourline's on-call rota needs four things on day one of the first production workload on AWS: a dashboard per service with the same four golden signals as before, an alarm per Azure alert rule that currently pages, KQL-to-Logs-Insights saved queries pinned in the runbooks, and staff sign-in through the existing Entra ID with Conditional Access. The SRE team converts its ten most-used KQL queries first, tests them against logs from a game day, and stores the Logs Insights equivalents in the runbook repository. For identity, staff use Pattern A with permission sets named after job functions (Payments-ReadOnly, Payments-Operator), Windows workloads use Pattern B for the SQL Server login, and customers are migrated to Cognito in a later wave once the token differences are handled.
Common mistakes
- Keeping the Application Insights SDK. It keeps sending to Azure. Switch to OpenTelemetry so you can choose the backend.
- Forgetting retention. CloudWatch log groups never expire by default; costs and personal data accumulate.
- Translating KQL line by line. Logs Insights has a smaller command set; restructure the question, and move heavy analytics to Athena over exported logs.
- Assuming Entra Conditional Access applies to API calls. It controls the sign-in; the session then works within its permissions. Add IAM conditions for sensitive actions.
- Choosing Simple AD. It is in maintenance; choose Managed Microsoft AD or AD Connector.
- Validating
audon Cognito access tokens. They have noaud, so every request is rejected, or the check gets removed entirely and security weakens.
Troubleshooting
| Symptom | Likely cause | Fix |
|---|---|---|
| Logs Insights returns no rows | Wrong log group, time range, or field name | Run fields @timestamp, @message | limit 20 to see the real field names |
Alarm stays INSUFFICIENT_DATA | Metric has no data points; treat-missing-data setting | Check the metric exists; set notBreaching for sparse metrics |
| No traces appear | Collector not reachable or wrong OTLP endpoint | Check OTEL_EXPORTER_OTLP_ENDPOINT and collector logs; confirm port 4317 (gRPC) or 4318 (HTTP) |
| Identity Center login loops back to Entra | SAML attribute or audience mismatch | Compare the ACS URL and entity ID in both consoles |
| User exists in Entra but not in AWS | Not in a group assigned to the SCIM app, or provisioning cycle pending | Assign the group; trigger provisioning on demand |
| 401 from API with a valid Cognito token | Audience validation on an access token | Disable ValidateAudience and validate client_id |
| Windows auth to RDS fails | No trust or the DNS cannot resolve the domain | Verify the Managed AD trust state and Route 53 Resolver rules |
Cost and cleanup
Cost label: free for the CLI, queries, CDK synthesis and OpenTelemetry code. CloudWatch has a free allowance for basic metrics, alarms and log ingestion; beyond it, log ingestion is billed per GB ⚠ VERIFY the current prices. AWS Managed Microsoft AD, Identity Center SAML federation with a real Entra tenant, and Cognito at scale are design-only here: Managed AD is billed hourly, and connecting a production Entra tenant belongs in a controlled change. Free alternative: do the identity exercises as diagrams and the Logs Insights exercises against logs from your own local runs sent to a sandbox log group.
cdk destroy ObservabilityStack --profile learn-sandbox
# or delete by hand
aws logs delete-log-group --log-group-name /hl/m13/rates-api --profile learn-sandbox --region us-east-1
aws cloudwatch delete-alarms --alarm-names hl-m13-5xx hl-m13-error-burst --profile learn-sandbox --region us-east-1
# Verify nothing tagged for the course remains
aws resourcegroupstaggingapi get-resources --tag-filters Key=Project,Values=cea-lab \
--profile learn-sandbox --region us-east-1 --query "ResourceTagMappingList[].ResourceARN"Exercises
- 🟢 Guided: translate the three KQL queries above to Logs Insights against a log group that holds JSON logs from the Rates API, and confirm the field names with
fields @timestamp, @message. - 🟡 Challenge: write a
Pattern chooseras a table for five Harbourline systems (staff portal, SQL Server on EC2, RDS for SQL Server with Windows auth, mobile banking app, a third-party SaaS) naming the pattern and its failure mode. - 🔴 Live problem: an auditor asks "how quickly does a leaver lose access to production on AWS?" Design the measurement, name every delay in the chain (Entra disable, SCIM sync, existing sessions, role session duration), and propose a control that shortens the longest one.
Assignment
Write a monitoring and identity migration plan for one service: current alerts and queries with their AWS equivalents, an OpenTelemetry instrumentation change, the identity pattern per user population, retention settings, and a revocation test. Commit it to your portfolio repository.
Solutions
Errors per five minutes: filter level = "Error" | stats count(*) by bin(5m). Slowest endpoints: stats pct(durationMs, 95) as p95 by route | sort p95 desc | limit 5. Trace lookup: fields @timestamp, @message | filter traceId = "abc123" | sort @timestamp asc. The why: Logs Insights queries are pipelines over JSON fields; if a field does not appear, the logs are not structured, and fixing the logger matters more than the query.
Staff portal: Pattern A (Entra to Identity Center), failure mode SCIM drift. SQL Server on EC2: Pattern B (Managed AD with trust), failure mode link outage. RDS for SQL Server with Windows auth: Pattern B, same. Mobile banking app: Pattern D (Cognito, possibly federated with Entra External ID), failure mode token claim differences. Third-party SaaS: stays on its own federation (Entra SAML), unaffected by the AWS move. The why: pick by who the user is and which resource must trust the identity.
Knowledge check
Answer: the native Logs Insights query language, OpenSearch PPL and OpenSearch SQL (the latter two added in December 2024). None of them is KQL, so queries must be rewritten.
Answer: the X-Ray SDKs and daemon entered maintenance mode on 2026-02-25, while OpenTelemetry is the vendor-neutral standard; X-Ray remains available as a trace backend through ADOT.
Answer: Entra ID is connected to IAM Identity Center as an external identity provider using SAML for sign-in and SCIM for provisioning users and groups.
Answer: an IAM role attached to the compute (EC2 instance profile, ECS task role, Lambda execution role), with no secret to store.
Answer: access tokens have no aud claim; they carry client_id. Validate client_id and token_use instead, or use ID tokens where an audience is present.
Interview questions
- "How do you keep Conditional Access when staff sign in to AWS?" Enforce MFA and device rules in Entra ID at sign-in, since the SAML assertion is only issued after they pass; AWS then limits what the session can do with permission sets and IAM conditions.
- "What would you check first if Logs Insights shows nothing?" The log group and time range, then whether the logs are structured JSON so fields exist.
- "When is Managed Microsoft AD needed?" When AWS-hosted Windows or SQL Server workloads need Kerberos or Windows authentication and Group Policy, typically with a trust to on-premises AD.
Certification practice
A bank keeps its workforce identities in Microsoft Entra ID and wants staff to sign in to many AWS accounts with MFA enforced by Entra, with groups kept in sync automatically. Which design is best?
A. Create an IAM user per employee with a password policy B. Connect Entra ID to IAM Identity Center as an external identity provider using SAML and SCIM, and assign groups to permission sets C. Share one IAM role and distribute the credentials D. Use Cognito user pools for staff
Answer: B. Federation keeps one source of truth, enforces Entra's MFA, and gives short sessions. A and C create standing credentials. D is meant for customer identities.
An operations team wants to be paged when more than five error log lines appear in three consecutive minutes. Their logs are JSON in CloudWatch Logs. What is the best approach?
A. Run a Logs Insights query manually each hour B. Create a metric filter on the error field and a CloudWatch alarm with an SNS notification C. Export logs to S3 daily D. Enable CloudTrail data events
Answer: B. A metric filter turns log lines into a metric; an alarm with three evaluation periods pages automatically. A is manual, C is delayed and D is for API auditing.
Completion checklist
- I can map Azure Monitor, Log Analytics and Application Insights to AWS services
- I translated three KQL queries to Logs Insights
- I can explain the four identity patterns and a failure mode of each
- I removed any log group or alarm I created and ran the tag query
Key takeaways
- Use OpenTelemetry so telemetry is portable; X-Ray is a backend, not an SDK choice.
- KQL does not run on AWS: rewrite queries for Logs Insights, PPL or SQL, and keep logs structured.
- Entra ID stays the directory; Identity Center is the AWS front door via SAML and SCIM.
- Managed AD is for Windows workloads; Simple AD is in maintenance.
- Cognito tokens differ from Entra's; fix audience validation first.
Further reading
Independent learning resource — not affiliated with, endorsed or sponsored by Amazon Web Services. AWS and related marks are trademarks of Amazon.com, Inc. or its affiliates. All banking data in examples is fictional. Sample architectures are for learning and are not automatically compliant with GDPR, PCI DSS or any banking regulation.
DevOps, IaC and secrets side by side: Azure DevOps, Bicep and Key Vault vs AWS
Translate Azure DevOps pipelines, service connections, ARM and Bicep deployments and Key Vault into GitHub Actions or CodePipeline, OIDC-assumed roles, CloudFormation or CDK change sets, and Secrets Manager with KMS, and decide what to migrate first and what to leave in place.
The 7 Rs of migration
Classify every application in a bank's estate as retire, retain, relocate, repurchase, rehost, replatform or refactor using an ordered decision rule, justify each choice with evidence, group chatty components into waves, and test the logic in a small C# classifier.