azdo-internal

作者: microsoft

觸發、監控及驗證 Aspire 內部 Azure DevOps 管線(microsoft-aspire,定義 1602)變更的專業知識,涵蓋…

npx skills add https://github.com/microsoft/aspire --skill azdo-internal

Aspire AzDO Internal Pipeline

Expert knowledge for triggering, monitoring, and validating changes to the Aspire internal Azure DevOps pipeline (microsoft-aspire, definition 1602).

Overview

The Aspire repo (microsoft/aspire on GitHub) has an internal mirror at dnceng/internal/_git/microsoft-aspire on Azure DevOps. The internal pipeline (eng/pipelines/azure-pipelines.yml) runs builds including native CLI compilation and installer preparation (WinGet + Homebrew).

Shell note (Windows). Command examples below use bash. On Windows, run them in pwsh or Git Bash. az, git, and gh are cross-platform; only the shell glue differs. PowerShell equivalents for the constructs used here: cmd >/dev/null 2>&1 → cmd *> $null; cmd || echo "msg" → cmd; if (-not $?) { Write-Host "msg" }; git remote -v | grep internal → git remote -v | Select-String internal; "grep the job log" → pipe the log to Select-String.

Key Details

ItemValue
AzDO Org/Projectdnceng / internal
Pipeline Definition ID1602
Pipeline Namemicrosoft-aspire
Internal Git Repohttps://dev.azure.com/dnceng/internal/_git/microsoft-aspire
Git Remote Name*Whatever you've configured for the internal repo — the name is arbitrary. Discover it with `git remote -v
Pipeline URLhttps://dev.azure.com/dnceng/internal/_build?definitionId=1602
Pipeline YAMLeng/pipelines/azure-pipelines.yml

Related Pipelines

PipelineDefinition IDYAMLPurpose
microsoft-aspire (main)1602eng/pipelines/azure-pipelines.ymlOfficial internal build (PR + CI)
microsoft-aspire unofficial(discover — see note)eng/pipelines/azure-pipelines-unofficial.ymlUnofficial/dev builds

The definition ID for the unofficial pipeline isn't hardcoded here because it can change. Discover it with:

az pipelines list --organization https://dev.azure.com/dnceng --project internal \
  --query "[?contains(name,'aspire')].{name:name,id:id}" -o table

Prerequisites

Before triggering or querying builds, confirm the environment is set up. Fail early if any of these are missing rather than emitting commands that will error:

# 1. Azure CLI present
az version >/dev/null 2>&1 || echo "az CLI not installed"

# 2. The azure-devops extension (provides `az pipelines` / `az devops`)
az extension show --name azure-devops >/dev/null 2>&1 || az extension add --name azure-devops

# 3. Authenticated to the dnceng org (interactive login or AZURE_DEVOPS_EXT_PAT)
az devops project show --project internal --organization https://dev.azure.com/dnceng >/dev/null 2>&1 \
  || echo "Not authenticated to dnceng/internal — run 'az login' or set AZURE_DEVOPS_EXT_PAT"

If the user lacks access to dnceng/internal (it's a Microsoft-internal org), stop and tell them — the build cannot be triggered from outside. Don't loop on auth errors.

How to Push to the Internal Repo

The internal AzDO repo mirrors GitHub. To push a branch for a manual build:

# Push your local branch to the internal remote (see "Git Remote Name" above to find yours)
git push INTERNAL_REMOTE <local-branch>:<remote-branch-name>

# Example (use your own alias as the branch prefix):
git push INTERNAL_REMOTE fix-azdo-pr-build:<your-alias>/fix-azdo-pr-build

If no remote points at the internal repo yet, add one (pick any name you like):

git remote add INTERNAL_REMOTE https://dnceng@dev.azure.com/dnceng/internal/_git/microsoft-aspire

Branch rules (important). Only push/build personal branches, e.g. <your-alias>/<branch>. The internal mirror enforces branch policies:

  • main and release/* are policy-gated — direct/force pushes are rejected (they require a PR), so don't use them as your scratch validation branch.
  • A build won't start on a branch that isn't permitted by the pipeline's branch controls; if a triggered build never starts or the push is rejected, the branch (not your YAML) is usually the cause. Switch to a <your-alias>/... branch and retry.

Triggering the Pipeline

Via Azure CLI (preferred for manual triggers)

# Trigger a build on a specific branch
az pipelines run \
  --id 1602 \
  --organization https://dev.azure.com/dnceng \
  --project internal \
  --branch <branch-name>

# Example:
az pipelines run \
  --id 1602 \
  --organization https://dev.azure.com/dnceng \
  --project internal \
  --branch <your-alias>/fix-azdo-pr-build

The command returns JSON with the build details including id (build ID) and url.

Pipeline structure

Don't rely on a snapshot here — the stages, job conditions, and variables change. Read the current definition from the repo:

  • eng/pipelines/azure-pipelines.yml — stages, jobs, gating conditions
  • eng/pipelines/templates/ — per-job step templates
  • eng/pipelines/scripts/ — the scripts those steps run

To see why a stage/job ran or was skipped in a specific build, read the build timeline — az pipelines build show --id <BUILD_ID> returns only headline status/result, not per-task records. Query the timeline resource directly (az devops invoke --area build --resource Timeline --route-parameters project=internal buildId=<BUILD_ID> --org https://dev.azure.com/dnceng) or open the build's web UI; the condition that actually evaluated is in the YAML.

Monitoring a Build

Build Results URL

https://dev.azure.com/dnceng/internal/_build/results?buildId=<BUILD_ID>

Via Azure CLI

# Check build status
az pipelines build show \
  --id <BUILD_ID> \
  --organization https://dev.azure.com/dnceng \
  --project internal

# List recent builds for the pipeline
az pipelines build list \
  --definition-ids 1602 \
  --organization https://dev.azure.com/dnceng \
  --project internal \
  --top 5

# List builds for a specific branch
az pipelines build list \
  --definition-ids 1602 \
  --organization https://dev.azure.com/dnceng \
  --project internal \
  --branch refs/heads/<your-alias>/fix-azdo-pr-build \
  --top 5

Common Tasks

Test pipeline changes on a feature branch

# 1. Push to internal remote (personal branch only — see Branch rules)
git push INTERNAL_REMOTE my-branch:<your-alias>/my-branch

# 2. Trigger the build
az pipelines run --id 1602 --organization https://dev.azure.com/dnceng --project internal --branch <your-alias>/my-branch

# 3. Monitor (use build ID from step 2 output)
az pipelines build show --id <BUILD_ID> --organization https://dev.azure.com/dnceng --project internal --query "{status:status,result:result,sourceBranch:sourceBranch}"

Monitoring a long-running build

There is no az pipelines watch command. For a long build, don't poll in a foreground loop — run a single detached watcher that polls az pipelines build show periodically, writes status to a file, and notifies on completion. Poll the build's status/result fields rather than scraping logs.

Limits: what you can't fully validate on a personal branch

Some stages only run on main/release/* and will be skipped or fail on a <your-alias>/... branch, so they can't be exercised this way:

  • Publish / release stages (NuGet push, WinGet/Homebrew PR submission) run in the release pipeline, not on feature-branch CI.
  • Steps that read the publish-build-assets variable group fail on non-main/non-release/* branches — by Arcade convention that group is only pulled for non-PR official branches. A feature-branch build legitimately can't access it.

When validating pipeline changes, confirm up front whether the path you're testing is even reachable from a personal branch; if not, validate the mechanism safely (next section) rather than running it for real on a release branch.

Validating publish/release-only changes safely

When a change only runs on main/release/* (publish, NuGet push, WinGet/Homebrew PR submission, release notifications), validate the mechanism without real side effects. Never let a publishing or PR-submitting step run live during validation. In order of preference:

1. Run the release pipeline with DryRun=true. The release/publish pipeline (eng/pipelines/release-publish-nuget.yml) exposes a DryRun runtime parameter that defaults to false (live) — so you must pass it explicitly:

az pipelines run --id <RELEASE_PIPELINE_ID> \
  --organization https://dev.azure.com/dnceng --project internal \
  --branch <your-alias>/<branch> \
  --parameters DryRun=true

ESRP sign/publish, the gh release upload (publish-release-cli-assets.ps1), and the WinGet/npm publish steps are all gated on this flag, so the path runs end-to-end without pushing anything.

Always verify dry-run actually engaged — don't assume it. The scripts print it; grep the job log for:

DryRun: True        # publish-release-cli-assets.ps1
Dry Run: true       # release-publish-nuget.yml

If you don't see it, treat the step as having run live. (A malformed -DryRun argument has silently bound positionally before and run live against the wrong target — confirm from the log, don't trust the intent.)

2. Extract the step's script and run it locally with test inputs and -DryRun. Best when you're changing script logic (version compute, manifest/cask generation, notifications) rather than YAML wiring — no pipeline, no side effects, fastest loop.

3. Test gating, not effect. If the change only affects when a stage runs, trigger builds on representative branches and inspect which stages were scheduled vs skipped in the build timeline. The publish never needs to fire.

Do not validate by repointing publish targets at a personal fork / test feed / test repo — a misconfiguration hits the real target, which is the side effect you're trying to avoid.

Reduce the pipeline to one job (scratch worktree)

To iterate fast on a single job's mechanics, it's often easiest to temporarily strip the pipeline down to just that job (remove other stages, drop dependsOn, hardcode the gating condition to true). Do this on a throwaway branch in a separate worktree so the gutted YAML never reaches your real PR:

git worktree add ../azdo-scratch -b <your-alias>/azdo-scratch
# edit eng/pipelines/azure-pipelines.yml down to the one job, commit, then:
git push INTERNAL_REMOTE <your-alias>/azdo-scratch:<your-alias>/azdo-scratch
az pipelines run --id 1602 --organization https://dev.azure.com/dnceng --project internal --branch <your-alias>/azdo-scratch

Caveats:

  • A reduced job validates the job's own logic, not its integration. Stripping upstream stages removes the variables and artifacts it would normally receive, so passing here doesn't guarantee it passes in a full run — re-validate the wiring end-to-end before merging.
  • Keep the job's own setup steps (restore, etc.) when slimming.
  • Clean up afterward: remove the worktree and delete the scratch branch on the internal remote.
  • The reduced YAML isn't in your PR, so note in the PR description what you validated and link the build.

來自 microsoft 的更多技能

oss-growth
microsoft
開源增長駭客角色
agent-framework-azure-ai-py
microsoft
使用Microsoft Agent Framework Python SDK(agent-framework-azure-ai)构建Azure AI Foundry代理。适用于使用AzureAIAgentsProvider创建持久化代理、使用托管工具(代码解释器、文件搜索、网络搜索)、集成MCP服务器、管理对话线程或实现流式响应。涵盖函数工具、结构化输出和多工具代理。
development
airunway-aks-setup
microsoft
在AKS上設定AI Runway——從裸叢集到執行模型。涵蓋叢集驗證、控制器安裝、GPU評估、供應商設定及首次部署。時機:「設定AI Runway」、「上線AKS叢集」、「安裝AI Runway」、「airunway設定」、「部署模型至AKS」、「在AKS上進行GPU推論」、「在AKS上設定KAITO」、「在AKS上執行LLM」、「在AKS上使用vLLM」、「在AKS上設定模型服務」、「AI Runway控制器」。
devops
appinsights-instrumentation
microsoft
使用Azure Application Insights檢測Web應用程式的指南。提供遙測模式、SDK設定與組態參考。適用時機:如何檢測應用程式、App Insights SDK、遙測模式、什麼是App Insights、Application Insights指南、檢測範例、APM最佳實踐。
devops
applicationinsights-web-ts
microsoft
使用Application Insights JavaScript SDK(@microsoft/applicationinsights-web)為瀏覽器/Web應用程式進行檢測。適用於真實使用者監控(RUM)——頁面檢視、點擊、AJAX/fetch依賴、例外、自訂事件,以及與後端OpenTelemetry追蹤關聯的瀏覽器端GenAI代理追蹤。涵蓋SDK載入器指令碼與npm設定、框架擴充(React、React Native、Angular)、點擊分析、遙測初始化器,以及從瀏覽器發出的代理/工具/模型span的OTel GenAI語意慣例。
devops
azure-ai-anomalydetector-java
microsoft
使用適用於 Java 的 Azure AI 異常偵測器 SDK 建置異常偵測應用程式。在實作單變量/多變量異常偵測、時間序列分析或 AI 驅動監控時使用。
development
azure-ai-language-conversations-py
microsoft
使用 azure-ai-language-conversations Python SDK 實作對話語言理解(CLU)。當使用 ConversationAnalysisClient 分析對話意圖與實體、建置 NLP 功能,或將語言理解整合至應用程式時使用。
development
azure-ai-ml-py
microsoft
Azure Machine Learning SDK v2 for Python。用於機器學習工作區、作業、模型、資料集、計算資源與管線。 觸發詞:「azure-ai-ml」、「MLClient」、「workspace」、「model registry」、「training jobs」、「datasets」。
development