GitLab合并请求至main分支时流水线配置问题排查

1. 触发条件修正：仅针对向main分支的MR触发 #

原配置未限定MR的目标分支为main，导致所有涉及models变更的MR都可能触发任务。需调整规则，明确仅向main分支发起的MR才执行，同时优先排除production分支：

rules:
  # 永远跳过production分支的流水线
  - if: '$CI_COMMIT_REF_NAME == "production"'
    when: never
  # 仅在向main分支发起MR且models目录有变更时触发
  - if: $CI_PIPELINE_SOURCE == "merge_request_event" && $CI_MERGE_REQUEST_TARGET_BRANCH_NAME == "main"
    changes:
      paths:
        - 'models/**/*'  # 匹配models目录下所有层级的文件，比单*覆盖更全面

2. 解决changes仅检查当前提交的问题 #

在merge_request_event场景下，GitLab CI的changes默认会对比源分支与目标分支的所有差异，而非单个提交。但默认的浅克隆可能导致分支历史不完整，需在脚本中补充目标分支的fetch操作，确保能获取完整的分支快照。

3. 解决git diff报错问题 #

报错根源是：

• 脚本中先切换到目标分支，此时本地未拉取源分支，导致$CI_COMMIT_REF_NAME对应的源分支无法识别
• 错误使用了CI_DEFAULT_BRANCH变量，应直接用GitLab提供的MR专属变量区分源/目标分支

修正后的脚本部分：

script: 
  - echo 'linting models'
  - echo "Target branch: $CI_MERGE_REQUEST_TARGET_BRANCH_NAME"
  - echo "Source branch: $CI_MERGE_REQUEST_SOURCE_BRANCH_NAME"
  - pip install sqlfluff==1.3.2
  # 拉取目标分支最新快照
  - git fetch origin $CI_MERGE_REQUEST_TARGET_BRANCH_NAME --depth=1
  # 对比源分支相对于目标分支的所有models目录变更
  - CHANGED_FILES=$(git diff --name-only origin/$CI_MERGE_REQUEST_TARGET_BRANCH_NAME...$CI_MERGE_REQUEST_SOURCE_BRANCH_NAME | grep '^models/')
  - |
    if [ -z "$CHANGED_FILES" ]; then
      echo "No models files changed, skipping lint"
      exit 0
    fi
    for file in $CHANGED_FILES
    do
      echo "Linting $file"
      sqlfluff lint "$file"
    done

完整修正后的CI配置 #

lint_files:
  before_script:
    - apt-get update && apt-get install -y git  # 若python镜像已自带git可删除此行
  image: python:3.10
  tags:
    - gitlab-org-docker
  stage: merge_request_checks
  script: 
    - echo 'linting models'
    - echo "Target branch: $CI_MERGE_REQUEST_TARGET_BRANCH_NAME"
    - echo "Source branch: $CI_MERGE_REQUEST_SOURCE_BRANCH_NAME"
    - pip install sqlfluff==1.3.2
    - git fetch origin $CI_MERGE_REQUEST_TARGET_BRANCH_NAME --depth=1
    - CHANGED_FILES=$(git diff --name-only origin/$CI_MERGE_REQUEST_TARGET_BRANCH_NAME...$CI_MERGE_REQUEST_SOURCE_BRANCH_NAME | grep '^models/')
    - |
      if [ -z "$CHANGED_FILES" ]; then
        echo "No models files changed, skipping lint"
        exit 0
      fi
      for file in $CHANGED_FILES
      do
        echo "Linting $file"
        sqlfluff lint "$file"
      done
  rules:
    - if: '$CI_COMMIT_REF_NAME == "production"'
      when: never
    - if: $CI_PIPELINE_SOURCE == "merge_request_event" && $CI_MERGE_REQUEST_TARGET_BRANCH_NAME == "main"
      changes:
        paths:
          - 'models/**/*'

关键说明 #

• 直接使用GitLab预定义的CI_MERGE_REQUEST_TARGET_BRANCH_NAME和CI_MERGE_REQUEST_SOURCE_BRANCH_NAME，无需手动获取分支名，更可靠
• 用origin/<target_branch>...<source_branch>的格式对比分支差异，无需切换分支，避免本地分支缺失问题
• changes规则在MR事件中自动对比源/目标分支的全量变更，配合fetch操作可规避浅克隆带来的历史缺失问题

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