deploying-to-azure
npx skills add https://github.com/microsoft-foundry/foundry-agent-webapp --skill deploying-to-azure
Agent 安装分布
Skill 文档
Deploying to Azure
Subagent Delegation for Deployment Analysis
Container logs and deployment output can be massive (1000+ lines). Delegate to subagent for:
- Analyzing full deployment logs
- Debugging container startup failures
- RBAC permission troubleshooting
- Multi-resource status checks
Delegation Pattern
runSubagent(
prompt: "ANALYSIS task - analyze deployment issue.
**Problem**: [describe the deployment failure]
**Run these commands**:
1. az containerapp logs show --name <app> --resource-group <rg> --tail 200
2. az containerapp show --name <app> --resource-group <rg> --query 'properties.provisioningState'
**Find**:
- Error messages or stack traces
- Resource provisioning failures
- Configuration mismatches
**Return** (max 15 lines):
- Root cause (1-2 sentences)
- Key error lines only (max 5)
- Suggested fix command
Do NOT include full log output.",
description: "Debug: [deployment issue]"
)
When to Delegate vs Inline
| Delegate to Subagent | Keep Inline |
|---|---|
| Full log analysis (100+ lines) | Quick status check |
| Multi-resource debugging | Single az command |
| RBAC permission audit | Container image query |
| Startup failure diagnosis | Provisioning state check |
Quick Commands
| Command | Purpose | Time |
|---|---|---|
azd up |
Full deployment (Entra app + infrastructure + container) | 10-12 min |
azd deploy |
Code-only deployment (Docker rebuild + push) | 3-5 min |
azd provision |
Re-run infrastructure + AI Foundry discovery | 5-7 min |
Deployment Phases
- preprovision â AI Foundry auto-discovery + tenant detection. For CI:
azd env set ENTRA_SERVICE_MANAGEMENT_REFERENCE "<guid>" - provision â Deploy Azure resources via Bicep (infrastructure + Entra app via Microsoft Graph Bicep extension + placeholder container image)
- postprovision â Sets
identifierUrion Entra app + updates redirect URIs + assigns RBAC to AI Foundry + generates local dev config - predeploy â Builds container (local Docker or ACR cloud build)
Implementation:
infra/entra-app.bicep(Entra app registration via Microsoft Graph Bicep extension)deployment/hooks/preprovision.ps1(AI Foundry discovery)deployment/hooks/postprovision.ps1(Entra config + RBAC + local config generation)deployment/hooks/predeploy.ps1(container build + push)deployment/hooks/modules/Get-AIFoundryAgents.ps1(agent discovery via REST)
Docker Multi-Stage Build
Build order: React â .NET â Runtime
- Frontend:
deployment/docker/frontend.Dockerfile - Backend:
deployment/docker/backend.Dockerfile - Custom npm registries: Add
.npmrctofrontend/directory
AI Foundry Resource Configuration
Auto-discovery (azd up): Searches subscription for AI Foundry resources â prompts to select if multiple â discovers agents via REST API â configures RBAC.
Change resource: Run azd provision to re-run discovery, or:
azd env set AI_FOUNDRY_RESOURCE_GROUP <rg>
azd provision
Container Infrastructure
- Health Probes: Liveness (
GET /api/healthevery 30s) and startup (GET /api/healthevery 10s, 5s initial delay) probes configured on the Container App - ACR Pull: Uses admin credentials via
listCredentials()â avoids chicken-and-egg with managed identity (Container App identity doesn’t exist until provisioned). Production apps should use a user-assigned managed identity created before the Container App. - Resource Defaults: 0.5 vCPU, 1GB RAM, 0-3 replicas (all parameterized in
container-app.bicep)
Troubleshooting
| Issue | Fix |
|---|---|
VITE_ENTRA_SPA_CLIENT_ID not set |
Run azd up to generate .env files |
AI_AGENT_ENDPOINT not configured |
Run azd provision to re-discover AI Foundry |
| No AI Foundry resources found | Create at https://ai.azure.com |
| Multiple AI Foundry resources | Run azd provision to select different resource |
| Container not updating | Check az containerapp logs show --name $app --resource-group $rg |
| Container fails health check | Verify /api/health endpoint returns 200 â check container logs for startup errors |
Useful Commands
# Check current container image
az containerapp show --name $app --resource-group $rg `
--query "properties.template.containers[0].image"
# View container logs
az containerapp logs show --name $app --resource-group $rg --tail 100
# Check RBAC assignments
$principalId = az containerapp show --name $app --resource-group $rg `
--query "identity.principalId" -o tsv
az role assignment list --assignee $principalId
Preprovision Hook Details
File: deployment/hooks/preprovision.ps1
What it does:
- Discovers AI Foundry resources in subscription (prompts if multiple)
- Discovers agents via REST API using
Get-AIFoundryAgents.ps1 - Auto-detects tenant ID
- Sets azd environment variables
Note: Entra app registration is handled by Bicep (infra/entra-app.bicep), not this hook.
Postprovision Hook Details
File: deployment/hooks/postprovision.ps1
What it does:
- Sets
identifierUri(api://{clientId}) on Entra app â can’t be done in Bicep because it references the auto-generatedappId - Updates Entra app redirect URIs (localhost + Container App FQDN)
- Assigns Cognitive Services User role to Container App’s managed identity on AI Foundry resource (via Azure CLI, not Bicep)
- Generates local dev config files (
.env.localfor frontend,.envfor backend)
Why RBAC via CLI?: Using Azure CLI for role assignment prevents azd from tracking the external AI Foundry resource group, avoiding accidental deletion on azd down.
Predeploy Hook Details
File: deployment/hooks/predeploy.ps1
What it does:
- Detects if Docker is available and running
- Uses local Docker build + push if available (~2 min)
- Falls back to ACR cloud build if Docker unavailable (~4-5 min)
- Updates Container App with new image (if it exists)
- Sets
SERVICE_WEB_IMAGE_NAMEenv var for Bicep
Dockerfile Example
File: deployment/docker/frontend.Dockerfile (production build)
# Stage 1: Build React Frontend
FROM node:22-alpine AS frontend-builder
ARG ENTRA_SPA_CLIENT_ID
ARG ENTRA_TENANT_ID
WORKDIR /app/frontend
COPY frontend/ ./
RUN npm ci
# Remove local .env files to prevent localhost config
RUN rm -f .env.local .env.development .env
ENV NODE_ENV=production
ENV VITE_ENTRA_SPA_CLIENT_ID=$ENTRA_SPA_CLIENT_ID
ENV VITE_ENTRA_TENANT_ID=$ENTRA_TENANT_ID
RUN npm run build
# Stage 2: Build .NET Backend
FROM mcr.microsoft.com/dotnet/sdk:9.0 AS backend-builder
WORKDIR /app
COPY backend/WebApp.sln ./
COPY backend/WebApp.Api/WebApp.Api.csproj ./backend/WebApp.Api/
COPY backend/WebApp.ServiceDefaults/WebApp.ServiceDefaults.csproj ./backend/WebApp.ServiceDefaults/
RUN dotnet restore backend/WebApp.Api/WebApp.Api.csproj
COPY backend/ ./backend/
RUN dotnet publish backend/WebApp.Api/WebApp.Api.csproj -c Release -o /app/publish
# Stage 3: Runtime - Single container serving API + static files
FROM mcr.microsoft.com/dotnet/aspnet:9.0-alpine
WORKDIR /app
COPY /app/publish ./
COPY /app/frontend/dist ./wwwroot
EXPOSE 8080
ENV ASPNETCORE_URLS=http://+:8080
ENV ASPNETCORE_ENVIRONMENT=Production
ENTRYPOINT ["dotnet", "WebApp.Api.dll"]
Related Skills
- writing-csharp-code – Backend coding patterns for Container App configuration
- writing-bicep-templates – Infrastructure templates for Azure resources
- troubleshooting-authentication – Entra ID and RBAC debugging
Official Documentation
| Topic | URL |
|---|---|
| Azure Container Apps | https://learn.microsoft.com/azure/container-apps/overview |
| Container Apps quickstart | https://learn.microsoft.com/azure/container-apps/quickstart-portal |
| Azure Developer CLI (azd) | https://learn.microsoft.com/azure/developer/azure-developer-cli/overview |
| azd templates | https://learn.microsoft.com/azure/developer/azure-developer-cli/azd-templates |
| Azure AI Foundry overview | https://learn.microsoft.com/azure/ai-foundry/what-is-ai-foundry |
| AI Foundry Agent Service | https://learn.microsoft.com/azure/ai-foundry/agents/overview |
| Managed Identity | https://learn.microsoft.com/entra/identity/managed-identities-azure-resources/overview |
| Cognitive Services RBAC | https://learn.microsoft.com/azure/ai-services/authentication |