Enterprise reference design

Central security operations across Azure application subscriptions

A central Sentinel design with explicit cloud connectors, agent collection, telemetry boundaries, response permissions, cost controls and failure tests.

Security architectsSOC engineersPlatform engineers

The enterprise use case

Design for the actual operating context.

An enterprise SOC needs to investigate identity, Microsoft 365 and Azure workload incidents across independently operated application subscriptions, while keeping operational access, sensitive payloads and automated response privileges under control.

Required outcomes

  • Connect each source through its supported collection path
  • Protect security telemetry and response permissions
  • Detect collection failures independently of incidents
  • Measure ingestion and retention before committing spend

Enterprise architecture

Explore the design, one view at a time.

Application and host collection

Follow separately selected host, security and operational events into the management subscription.

Central security operations — Application and host collectionFollow separately selected host, security and operational events into the management subscription.. Official Microsoft service icons; explicit resource and network boundaries. Green numbered steps match the F-numbered flow register below the diagram. Central security operations Application and host collection Follow separately selected host, security and operational events into the management subscription. Application subscriptions Host agents and resource diagnostic settings Management subscription Regional workspaces with separate access scopes F3: Host events. The host produces events through its audit policy; AMA collects the selected local events. F4: DCR assignment. Configuration flow: reviewed DCR associations select event streams and destinations; this arrow is not event transport. F5: Selected host logs. The Sentinel Windows Security Events via AMA path writes SecurityEvent. Generic Windows event DCRs can write Event instead. F6: Categories. Configuration flow: each App Service and Key Vault resource receives its own diagnostic settings; subscriptions receive Activity Log export settings. The shared node is not a forwarding service. F7: Security categories. Selected PaaS resource logs and subscription Activity Logs are delivered directly by Azure to the security workspace. F8: Operational categories. Approved operational categories go to the separate operations workspace; document and cost any duplication. Azure virtual machines Azure virtual machines Azure Monitor Agent Azure Monitor Agent Data collection rules Data collection rules App Service App Service Diagnostic settings Diagnostic settings Security workspace Security workspace Operations workspace Operations workspace Host events 3 DCR assignment 4 Selected host logs 5 Categories 6 Security categories 7 Operational categories 8 Collection is source-specific AMA and DCRs collect supported host events. Diagnostic settings route supported resource log categories; neither replaces cloud connectors. Boundary to verify Managed App Service and workspaces remain outside customer VNets. Select the correct table plan, region and access scope before enabling collection. Reference design · Microsoft product icons used under Microsoft’s published terms · Numbered steps refer to the accompanying flow register
7 components · 6 numbered connections

Green step numbers match the F-numbered flow register below. Solid boundaries show ownership; dashed blue boundaries show networks. Logical groups are labeled explicitly. Microsoft service icons identify the components.

Cloud identity and optional partner ingestion

Cloud services use their own connectors; partner Event Hubs ingestion needs a separately owned adapter.

Central security operations — Cloud identity and optional partner ingestionCloud services use their own connectors; partner Event Hubs ingestion needs a separately owned adapter.. Official Microsoft service icons; explicit resource and network boundaries. Green numbered steps match the F-numbered flow register below the diagram. Central security operations Cloud identity and optional partner ingestion Cloud services use their own connectors; partner Event Hubs ingestion needs a separately owned adapter. Microsoft Entra tenant Service-specific cloud sources Optional external source Separate ingestion contract Management subscription Cloud connectors and explicitly engineered ingestion F1: Cloud enrollment. The group represents three separate Entra, Microsoft 365 and Defender XDR integrations. Cloud feeds do not pass through AMA or the optional Event Hubs branch. F2: Cloud events. Each connector writes its supported security tables after licensing, tenant and category checks. F9: Optional stream. A partner publishes an approved event schema to the optional Event Hubs namespace. F10: Consume / validate. An owned adapter checkpoints, validates, retries and isolates malformed events. It is not the preview direct Event Hubs-to-Logs integration. F11: API / DCR ingestion. The adapter uses scoped identity and supported Logs Ingestion API tables and DCR schema. Preserve event identifiers for duplicate analysis. Microsoft Entra ID Microsoft Entra ID Microsoft 365 audit Microsoft 365 audit Defender XDR Defender XDR Cloud connectors. Service-specific APIs Cloud connectors Service-specific APIs Security workspace Security workspace Partner stream Partner stream Event Hubs Event Hubs Ingestion adapter Ingestion adapter Cloud enrollment 1 Cloud events 2 Optional stream 9 Consume / validate 10 API / DCR ingestion 11 Cloud sources are separate integrations Entra, Microsoft 365 and Defender XDR are tenant services, not VM-agent routes. Their connector permissions, licensing and available tables differ. Optional means engineered Event Hubs is a transport in this pattern. An owned consumer validates schemas, handles retries and writes through a supported ingestion API and DCR. Reference design · Microsoft product icons used under Microsoft’s published terms · Numbered steps refer to the accompanying flow register
8 components · 5 numbered connections

Green step numbers match the F-numbered flow register below. Solid boundaries show ownership; dashed blue boundaries show networks. Logical groups are labeled explicitly. Microsoft service icons identify the components.

Detection, response and collection health

Keep incident automation least privileged, and alert independently when expected collection stops.

Central security operations — Detection, response and collection healthKeep incident automation least privileged, and alert independently when expected collection stops.. Official Microsoft service icons; explicit resource and network boundaries. Green numbered steps match the F-numbered flow register below the diagram. Central security operations Detection, response and collection health Keep incident automation least privileged, and alert independently when expected collection stops. Management subscription Microsoft Sentinel enabled on the security workspace Collection monitoring Independent freshness and failure checks Operational escalation Owned on-call notification route F12: Detection inputs. Sentinel analytics query approved security tables with documented plan, timing and entity requirements. F13: Incident handling. Reviewed incident automation begins with routing and enrichment, with disruptive actions gated. F14: Invoke workflow. Sentinel invocation authority is scoped to the intended playbook resource group and separate from workflow target permissions. F15: Authorized action. The workflow's own identity authenticates to the target API with least privilege and an approval step where the action is disruptive. F16: Collection failures. Test source freshness, ingestion lag, collector health and workflow failures. Include a notification path that does not depend exclusively on querying the affected workspace. Security workspace Security workspace Sentinel analytics Sentinel analytics Automation rules Automation rules Logic Apps playbooks Logic Apps playbooks Approved response Approved response Microsoft Sentinel. Workspace-enabled SIEM Microsoft Sentinel Workspace-enabled SIEM Collection health Collection health On-call notification On-call notification Detection inputs 12 Incident handling 13 Invoke workflow 14 Authorized action 15 Collection failures 16 Response is a separate privilege boundary Sentinel automation invokes approved Logic Apps. Each playbook identity receives only the permissions needed on its named response targets. Operate the collector Monitor connector health, ingestion latency, transformation failures and data freshness. A quiet detection rule does not prove that telemetry is arriving. Reference design · Microsoft product icons used under Microsoft’s published terms · Numbered steps refer to the accompanying flow register
8 components · 5 numbered connections

Green step numbers match the F-numbered flow register below. Solid boundaries show ownership; dashed blue boundaries show networks. Logical groups are labeled explicitly. Microsoft service icons identify the components.

Follow the numbered connections

Flow register

Paths shown in the diagram have the following scope.

FlowFrom → toPurpose and control
F1Identity & controlMicrosoft Entra ID Separate cloud connectorsCloud enrollment

The group represents three separate Entra, Microsoft 365 and Defender XDR integrations. Cloud feeds do not pass through AMA or the optional Event Hubs branch.

F2MonitoringSeparate cloud connectors Security Log AnalyticsCloud events

Each connector writes its supported security tables after licensing, tenant and category checks.

F3MonitoringAzure virtual machines Azure Monitor AgentHost events

The host produces events through its audit policy; AMA collects the selected local events.

F4Identity & controlDCRs and associations Azure Monitor AgentDCR assignment

Configuration flow: reviewed DCR associations select event streams and destinations; this arrow is not event transport.

F5MonitoringAzure Monitor Agent Security Log AnalyticsSelected host logs

The Sentinel Windows Security Events via AMA path writes SecurityEvent. Generic Windows event DCRs can write Event instead.

F6Identity & controlApp Service resources Azure diagnostic settingsCategory selection

Configuration flow: each App Service and Key Vault resource receives its own diagnostic settings; subscriptions receive Activity Log export settings. The shared node is not a forwarding service.

F7MonitoringAzure diagnostic settings Security Log AnalyticsSecurity categories

Selected PaaS resource logs and subscription Activity Logs are delivered directly by Azure to the security workspace.

F8MonitoringAzure diagnostic settings Operational Log AnalyticsOperational categories

Approved operational categories go to the separate operations workspace; document and cost any duplication.

F9MonitoringOptional partner stream Optional Event HubsOptional stream

A partner publishes an approved event schema to the optional Event Hubs namespace.

F10MonitoringOptional Event Hubs Owned ingestion adapterConsume and validate

An owned adapter checkpoints, validates, retries and isolates malformed events. It is not the preview direct Event Hubs-to-Logs integration.

F11MonitoringOwned ingestion adapter Security Log AnalyticsAPI / DCR ingestion

The adapter uses scoped identity and supported Logs Ingestion API tables and DCR schema. Preserve event identifiers for duplicate analysis.

F12MonitoringSecurity Log Analytics Analytics and incidentsDetection inputs

Sentinel analytics query approved security tables with documented plan, timing and entity requirements.

F13RecoveryAnalytics and incidents Automation rulesIncident handling

Reviewed incident automation begins with routing and enrichment, with disruptive actions gated.

F14Identity & controlAutomation rules Logic Apps playbooksInvoke workflow

Sentinel invocation authority is scoped to the intended playbook resource group and separate from workflow target permissions.

F15Identity & controlLogic Apps playbooks Approved response targetAuthorized action

The workflow's own identity authenticates to the target API with least privilege and an approval step where the action is disruptive.

F16MonitoringCollection health Operational notificationCollection failures

Test source freshness, ingestion lag, collector health and workflow failures. Include a notification path that does not depend exclusively on querying the affected workspace.

Assumptions to confirm

  • One Microsoft Entra tenant, with additional tenants requiring separate connector planning
  • This consolidated variant shares one platform Management subscription for security and operations
  • West Europe is the example workspace region, subject to residency approval
  • Entra, Microsoft 365 and Defender licenses are validated before connector onboarding
  • This is a documentation-reviewed reference design, not a deployed or benchmarked environment

Use case and boundary decisions

A shared SOC supports several application teams, including an internal AI assistant and a customer portal. Investigators need a correlated timeline without giving every developer access to tenant sign-ins or every automation identity authority to disable users. The design below is an explicit consolidated-subscription variant, with service constraints supported by the linked documentation.

This variant places the security workspace, Sentinel resources, response automation and operational workspace in the platform Management subscription, separate from application subscriptions. It reduces subscription administration but increases coupling between security and operations administrators: broad inherited subscription permissions can affect both monitoring environments. Use separate resource groups and narrowly scoped roles, and accept that shared administrative boundary explicitly. The enterprise landing-zone design in this library remains the recommended enterprise separation: a Security subscription for Sentinel and security evidence, with operational telemetry in Management. Moving to that split preserves the collection paths described here while requiring destination permissions and deployment scopes to be updated. A subscription separates ownership and permissions; it is not a network boundary. Azure PaaS and Microsoft cloud connectors are not appliances inside application VNets.

Use one security workspace and one operational workspace here because the teams have different access and retention requirements. This is an explicit tradeoff, not a requirement to split every installation. Fewer workspaces simplify correlation and commitment-tier economics; table and resource access controls can remove the need for some splits. Microsoft workspace design guidance.

Components and control ownership

Component Selected design Accountable owner and control
Security workspace Regional Log Analytics workspace with Sentinel; Analytics plan for initial detection tables SOC platform team; tightly scoped reader, responder and engineering roles
Operational workspace Separate application diagnostics with only approved overlap Platform operations; resource-context access where appropriate
Cloud sources Dedicated Entra, Microsoft 365 and Defender XDR integrations Identity and collaboration owners approve categories and licenses
Azure virtual machines AMA with reviewed DCR associations in the application subscriptions Workload owner maintains agent and OS audit policy
PaaS resources Per-resource diagnostic settings Platform policy deploys settings; application owner proves events arrive
Optional streaming branch Event Hubs Standard plus a maintained ingestion adapter Integration team owns consumer checkpoints, retries and schema
Response Incident handling first; isolated Logic Apps playbook resource group SOC approves each action and its target permissions
Configuration Versioned infrastructure, rules, transforms and connector inventory Platform and detection teams review changes together

The collection view uses Azure virtual machines as its host source. An extension to on-premises or other-cloud servers requires separate Azure Arc onboarding and approved connectivity from those external networks to the required management and monitoring endpoints. Arc gives an external machine an Azure resource representation; it does not place that machine inside the application VNet. Validate its agent, DCR association and network path separately. Azure Arc-enabled servers scope.

West Europe is an example, not a residency determination. Document where each source, workspace and processing feature operates. Built-in Microsoft and Azure SaaS connectors require a workspace in the source tenant; a second tenant needs its own design. Cross-tenant management through Lighthouse does not erase that restriction. Sentinel workspace and tenant constraints.

Collection contract before onboarding

Record the source, category, destination table, expected latency, owner and retention decision before enabling a connector. Verify a generated event, not just a green configuration screen.

Source Actual route and expected table Validation detail
Entra sign-ins and audit Entra integration to SigninLogs and AuditLogs Sign-in ingestion requires the documented P1/P2 entitlement; workload risk feeds have additional licensing
Microsoft 365 audit Microsoft 365 connector to OfficeActivity Confirm selected workloads, audit availability and realistic delivery latency
Defender XDR XDR integration with selected incidents, alerts and event tables Reconcile overlapping connectors and incident creation rules
Azure control plane Subscription Activity Log diagnostic export to AzureActivity Use the Azure Activity connector policy; prove a benign administrative change arrives
Windows security events Sentinel Windows Security Events via AMA to SecurityEvent Generic AMA Windows collection instead writes Event; forwarded-event collection has another table
Key Vault AuditEvent diagnostic category to AzureDiagnostics Preserve operation, result, identity and resource context
App Service AppServiceHTTPLogs category to AppServiceHTTPLogs Enable and verify the category on the actual app; do not assume every OS/configuration emits identical logs

These mappings follow the Entra prerequisites, Microsoft 365 connector catalog, Defender integration, Sentinel Windows connector, generic Windows collection, and the current Key Vault and App Service log references.

Activity Log export is a subscription-level collection path, distinct from resource diagnostics. Enroll every workload subscription and the Management subscription so changes to collection and response infrastructure are visible. Azure Activity connector setup.

Numbered data and response flows

  1. F1 — Cloud source enrollment. The cloud-connector group represents separate Entra, Microsoft 365 and Defender integrations. It is not one agent or a shared network tunnel. In the Defender portal, account for automatically configured XDR integration before adding older connector paths.
  2. F2 — Cloud delivery. Approved categories arrive in their security-workspace tables. Measure event time versus ingestion time independently for each feed.
  3. F3 — Host collection. AMA reads selected local Windows events from onboarded machines. OS audit policy must produce those events first.
  4. F4 — Collection configuration. DCR associations define the approved event selection and destinations. Change review prevents duplicate collection and accidental removal of security events.
  5. F5 — Agent delivery. AMA sends selected events through its supported Azure Monitor endpoints. Permit required connectivity; the diagram does not imply that all cloud logs traverse these agents.
  6. F6 — Export selection. App Service and Key Vault each receive their own diagnostic setting; subscriptions receive Activity Log export settings. The Diagnostic settings node in the collection view represents configuration, not an intermediate collector; the illustrated App Service path does not replace other sources’ own settings.
  7. F7 — Security export. Required resource and Activity Logs go directly to the security workspace. Diagnostic logs are not enabled merely because a resource exists.
  8. F8 — Operational export. Only approved operational categories go to the operations workspace. Document intentional duplication and its ingestion charge.
  9. F9 — Optional external stream. A partner source writes a defined event contract to a dedicated Event Hubs namespace and consumer group.
  10. F10 — Consumer processing. An owned adapter reads checkpoints, validates records, isolates malformed messages and retries safely. This component is an integration responsibility, not an assumed built-in connector.
  11. F11 — Custom ingestion. The adapter uses a supported Logs Ingestion API/DCR schema and scoped identity. Preserve source identifiers to recognize retries and duplicates.
  12. F12 — Detection. Sentinel analytics query the security tables. Rule owners document required tables, scheduling, lookback, entity mapping and ingestion-delay tolerance.
  13. F13 — Incident automation. Approved analytics produce incidents and automation rules; begin with routing and enrichment.
  14. F14 — Workflow invocation. Sentinel receives permission to invoke playbooks in the isolated automation resource group.
  15. F15 — Controlled response. A playbook authenticates separately to its target API. Destructive containment requires an explicit approval step and a rollback procedure.
  16. F16 — Independent notification. Collector-health monitoring raises a service notification through an independently tested operational path; a workspace outage must not rely exclusively on querying that workspace.

Diagnostic delivery to a regional Event Hub or storage account requires a destination in the source resource’s region. Firewall-protected destinations have documented trusted-service requirements. These constraints do not mean every Log Analytics workspace must share every source region. Diagnostic settings restrictions.

Direct Event Hubs-to-Azure Monitor ingestion is a separate preview option with commitment-tier or dedicated-cluster prerequisites and regional capacity constraints. It is not the baseline adapter in F10–F11, nor a reason to put SaaS connectors behind Event Hubs. Direct Event Hubs ingestion prerequisites.

Sensitive data and response permissions

Exclude access tokens, authorization headers, raw prompts, document bodies and model responses at the producing application. Retain correlation identifiers and security-relevant metadata. Transformation support varies by connector and table; a workspace transform is neither universal nor retroactive. Test filters against actual sample events and check for another export retaining the original payload. Supported transformation paths.

Separate three authorities: the engineer who edits a workflow, Sentinel’s permission to invoke it, and the workflow identity’s permission on the target. Grant Sentinel Automation Contributor only on the intended playbook resource group. A managed identity that updates incidents needs the corresponding Sentinel write role; reading incidents does not require Contributor. Grant target API permissions independently and avoid subscription Owner. The Sentinel connector’s managed-identity authentication is currently documented as preview: approve that dependency explicitly, or keep that automated path disabled pending an accepted authentication design. Invocation permissions and connector authentication.

Acceptance and adversarial tests

Test Required evidence before production
Remove one DCR association in a test subscription Missing-host telemetry is noticed within the agreed window and attributed to the correct owner
Disable a test diagnostic setting Change telemetry and source-silence checks identify the gap; configuration is restored
Emit a canary secret-shaped value Value is excluded at source; correlation metadata remains useful
Query as an application developer Operational access works; tenant-wide security data remains unavailable
Replay a duplicate partner event Adapter behavior and detection deduplication are documented
Trigger a playbook without target authority Action fails safely and the incident shows the failure
Delay a connector feed Rule lookback avoids a silent coverage gap or an explained limitation is accepted

The related hunts are starting points for ingestion and configuration monitoring, not proof that the full architecture or any threat detection has been tested. In particular, ingestion-source-silence does not run unchanged against this baseline: it requires AZKVAuditLogs and StorageBlobLogs, while this design sends Key Vault AuditEvent to AzureDiagnostics and does not onboard Blob logs. Adapt the query to the actual collected tables, filter the appropriate resource categories, and validate its resource-identifier mapping and expected activity before use. Do not treat a missing table as evidence that a source is healthy.

Operations, recovery and cost

Enable Sentinel health and audit monitoring, inspect SentinelHealth and SentinelAudit, and monitor Logic Apps run outcomes separately. A successful trigger does not prove every response action succeeded. Track AMA heartbeat, source event freshness, ingestion delay, adapter backlog and rejected records. Sentinel health monitoring.

Keep critical detection tables on Analytics initially. Basic and Auxiliary/Lake have different query, alerting and feature support; they are not interchangeable cheap storage for an existing rule. Confirm each table and rule combination before changing plan. Choose table retention from investigation and contractual needs, then measure daily volume, duplication and query cost before selecting a commitment tier. A daily cap can create a detection blind spot. Azure Monitor table capabilities.

Roll out one application subscription and one cloud feed first. The platform team proves configuration redeployment; the SOC proves canary detection and incident handling; source owners sign off completeness. Rebuild a lost workspace from reviewed configuration, then reauthorize connectors and validate ingestion. Historical log recovery and regional continuity need a separately tested retention/export or replication design; infrastructure code alone does not restore evidence. Exercise that recovery quarterly and record measured gaps rather than promising a universal recovery time.

Trace the design decisions

Microsoft architecture & product guidance

  1. Design a Log Analytics workspace architecture ↗
  2. Extend Microsoft Sentinel across workspaces and tenants ↗
  3. Azure Monitor diagnostic settings ↗
  4. Sentinel diagnostic settings-based connections ↗
  5. Connect Windows-based services through Azure Monitor Agent ↗
  6. Collect Windows events with Azure Monitor Agent ↗
  7. Azure Arc-enabled servers overview ↗
  8. Connect Microsoft Entra ID logs to Sentinel ↗
  9. Connect Microsoft Defender XDR to Sentinel ↗
  10. Microsoft Sentinel connector reference ↗
  11. Key Vault supported resource logs ↗
  12. App Service supported resource logs ↗
  13. Azure Monitor table plans ↗
  14. Transform and filter data in Sentinel ↗
  15. Ingest events from Event Hubs into Azure Monitor Logs ↗
  16. Authenticate playbooks to Microsoft Sentinel ↗
  17. Automate and run Microsoft Sentinel playbooks ↗
  18. Audit and monitor Microsoft Sentinel health ↗

Inspect operational evidence

Supporting Sentinel queries

These are contextual hunting starters, not end-to-end validation of the architecture. Check each query's source tables, collection scope and limitations.

Query library →