Azure DevOps Pipeline跨阶段传递变量失败求助

问题现象 #

在ImageDeployment阶段生成的NEW_VERSION变量，传递到Cleanup阶段后输出为空。已尝试相关解决方案但未解决，流水线配置如下：

stages:
  - stage: ImageDeployment
    jobs:
      - job: NewImageDeployment
        timeoutInMinutes: 180
        pool:
          vmImage: 'ubuntu-latest'
        steps:
          - task: AzureKeyVault@2
            displayName: 'Fetch Secrets'
            inputs:
              connectedServiceName: '$(GLOBAL_SERVICE_CONNECTION)'
              keyVaultName: $(GLOBAL_KV_NAME)
              secretsFilter: "Packer-Secret"
              runAsPreJob: false

          - task: AzureCLI@2
            displayName: 'Check image version'
            retryCountOnTaskFailure: 0
            inputs:
              azureSubscription: '$(GLOBAL_SERVICE_CONNECTION)'
              scriptType: 'bash'
              scriptLocation: 'inlineScript'
              inlineScript: |
                while IFS=$'\t' read -a arr_image_versions; do
                  version=${arr_image_versions[0]}
                  date=${arr_image_versions[1]}
                  if [[ $date > $old_date ]]; then
                    last_version=$version
                  fi
                  old_date=$date
                done <<< "$(az sig image-version list --gallery-image-definition xxxxxxx --gallery-name xxxxxxxx --resource-group xxxxxxxxx --query "[].[name, publishingProfile.publishedDate]" -o tsv)"
                echo "Last image version: $last_version"
                echo "##vso[task.setvariable variable=last_version]$last_version"
            
          - task: Bash@3
            name: CheckImageVersion
            displayName: 'Increment image version'
            inputs:
              targetType: 'inline'
              script: |
                IFS='.' read -ra VERSION_PARTS <<< "$(last_version)"
                PATCH_VERSION=${VERSION_PARTS[2]}
                PATCH_VERSION=$((PATCH_VERSION + 1))
                NEW_VERSION="${VERSION_PARTS[0]}.${VERSION_PARTS[1]}.$PATCH_VERSION"
                echo "New image version: $NEW_VERSION"
                echo "##vso[task.setvariable variable=NEW_VERSION;isOutput=true]$NEW_VERSION"


  - template: tests/build_functionapp_testcase2.yaml@self

  - template: tests/datafactory_testcase3.yaml@self

  - stage: Cleanup
    condition: failed()
    jobs:
      - job: DeleteImage
        pool:
          vmImage: 'ubuntu-latest'
        variables:
          NEW_VERSION: $[ stageDependencies.ImageDeployment.NewImageDeployment.outputs['CheckImageVersion.NEW_VERSION'] ]
        steps:
          - script: |
              echo "NEW_VERSION in Cleanup stage: $(NEW_VERSION)"
          - task: AzureCLI@2
            displayName: 'Delete Packer Image'
            inputs:
              azureSubscription: '$(GLOBAL_SERVICE_CONNECTION)'
              scriptType: 'bash'
              scriptLocation: 'inlineScript'
              inlineScript: |
                echo "Deleting image version: $(NEW_VERSION)"
                az sig image-version delete --gallery-image-definition xxxxxxxxxxxxxxxxxxxxxxx

排查与修复方案 #

1. 调整Cleanup阶段的触发条件 #

当前Cleanup阶段仅用condition: failed()触发，若ImageDeployment阶段本身执行失败（比如生成NEW_VERSION的步骤未完成），变量根本未被设置，自然为空。需确保ImageDeployment成功执行后，才在后续阶段失败时触发清理：

- stage: Cleanup
  condition: and(failed(), succeeded('ImageDeployment'))
  # 其他配置不变

2. 确认跨阶段变量引用语法 #

检查stageDependencies层级是否完全匹配：阶段名ImageDeployment → 作业名NewImageDeployment → 任务名CheckImageVersion → 输出变量名NEW_VERSION。原引用语法本身正确，但需确保模板引入的阶段未打乱依赖顺序。

3. 确保输出变量的作用域有效性 #

在设置NEW_VERSION时，明确变量的输出属性，可补充isReadOnly避免后续步骤意外修改：

echo "##vso[task.setvariable variable=NEW_VERSION;isOutput=true;isReadOnly=true;]$NEW_VERSION"

4. 验证前置变量last_version的正确性 #

若Check image version步骤未正确获取到last_version，后续生成NEW_VERSION会失败。在该步骤末尾添加调试打印：

echo "##vso[task.setvariable variable=last_version]$last_version"
echo "DEBUG: last_version = $last_version"

同时检查Azure CLI命令az sig image-version list的返回格式是否符合预期，确保能正确解析版本号和日期。

5. 明确Cleanup阶段的依赖关系 #

流水线引入的两个模板阶段可能影响依赖链，需明确Cleanup依赖所有前置阶段：

- stage: Cleanup
  dependsOn:
    - ImageDeployment
    - build_functionapp_testcase2  # 替换为模板内定义的实际阶段名
    - datafactory_testcase3        # 替换为模板内定义的实际阶段名
  condition: and(failed(), succeeded('ImageDeployment'))
  # 其他配置不变

内容的提问来源于stack exchange，提问作者Sumit Bhardwaj