无变更时CI/CD流水线NuGet间歇性恢复失败问题求助

未做任何变更的情况下，CI/CD流水线中的NuGet突然出现间歇性失败：执行dotnet test时触发错误NU1301，提示无法加载源https://api.nuget.org/v3/index.json的服务索引，同时反复重试FindPackagesByIdAsync操作，报错Name or service not known (api.nuget.org:443)。

我们使用Sonatype Nexus作为私有源，其他用户流水线均正常，但该问题时好时坏：允许v3源通过时，有时流水线能正常执行（仍会弹出上述报错），有时直接失败。目前未修改任何配置，求可行解决办法。

附带流水线配置：

variables:
  # Update this for new version builds. - TODO: This will be removed at some point once Jim figures out automatic versioning based on git tags.
  VERSION: "1.0.$CI_PIPELINE_IID"
  VERSION_SUFFIX: "$CI_COMMIT_REF_SLUG-$CI_PIPELINE_IID-$CI_COMMIT_SHORT_SHA"
  # NOTE: If you change your package name, you should also change this to match.
  PACKAGE_NAME: "Reporting.HoldingQueueReport"

image: $CICD_IMAGE

stages:
  - test
  - deploy
  - deploy_runner

Test:
  stage: test
  before_script:
    - dotnet nuget add source $REPORTS_NUGET_SOURCE -n Nexus
    - "echo | dotnet --version" # must be v 2.2.300 or later for reporter tool to work
    - dotnet --info
    - dotnet restore --verbosity detailed
    - dotnet tool install dotnet-reportgenerator-globaltool --tool-path tools
  script:
    - dotnet test --logger "junit" --collect:"XPlat Code Coverage"
    # each reporttype must be a separate line, doesn't work together ; separated
    - ./tools/reportgenerator "-reports:${CI_PROJECT_DIR}/**/TestResults/*/coverage.cobertura.xml" "-targetdir:Reports_Coverage" -reportTypes:TextSummary;
    - ./tools/reportgenerator "-reports:${CI_PROJECT_DIR}/**/TestResults/*/coverage.cobertura.xml" "-targetdir:Reports_Coverage" -reportTypes:Html;
    - cat ./Reports_Coverage/Summary.txt
    - echo 'End Summary'
  coverage: /Line coverage[\s\S].+%/
  artifacts:
    paths:
      - Reports_Coverage/
    reports:
      junit:
        - "*.Test/TestResults/TestResults.xml"
        - "*.Tests/TestResults/TestResults.xml"

NuGet Publish:
  stage: deploy
  before_script:
    # Add our custom NuGet repository group so nuget restore works
    - dotnet nuget add source $REPORTS_NUGET_SOURCE -n Nexus
  script:
    # Pack the project into a NuGet package
    - dotnet pack -c Release -p:PackageVersion=${VERSION} -o ./build/packages --version-suffix "$VERSION_SUFFIX" --include-symbols

    - dotnet nuget push ./build/packages/**/*.symbols.nupkg -k ${NEXUS_NUGET_KEY} -s $REPORTS_DEPLOY_NUGET_SOURCE
  only:
    - master

1. 禁用默认公共NuGet源，仅使用私有Nexus源 #

既然已用Nexus作为私有源且其他用户正常，说明Nexus应已代理所需公共包。在流水线的Test和NuGet Publish阶段的before_script中添加移除默认源的命令，避免CI环境尝试访问公共源：

- dotnet nuget remove source "nuget.org" || true

（|| true用于避免源不存在时命令报错中断流水线）

2. 修复CI环境的DNS解析问题 #

报错Name or service not known指向DNS解析失败，间歇性出现说明CI runner的DNS服务不稳定：

• 临时替换CI环境DNS为公共DNS，在before_script中添加：

- echo "nameserver 8.8.8.8" > /etc/resolv.conf

• 联系运维检查CI runner所在网络的DNS配置，排查是否存在解析超时或缓存异常。

3. 优化NuGet重试与超时配置 #

修改NuGet配置，增加重试次数和超时时间，缓解间歇性网络问题：

- dotnet nuget config -set retryCount=5
- dotnet nuget config -set timeout=300

4. 确认Nexus已正确代理公共NuGet源 #

检查Nexus配置，确认是否创建了nuget.org的代理仓库，并将其加入到你们使用的仓库组中。这样所有包请求都会走Nexus，无需直接访问公共源，从根源避免外部网络问题。

5. 分离restore与test的依赖处理 #

当前dotnet test会自动触发restore，可能和之前的dotnet restore冲突。修改dotnet test命令，跳过自动restore：

- dotnet test --no-restore --logger "junit" --collect:"XPlat Code Coverage"

确保只使用之前restore好的包，避免重复请求源。

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