GitLab CI/CD多镜像场景下npm依赖缓存失效问题求解

方案1：统一缓存Key并管控缓存策略 #

GitLab CI默认会根据job、分支等生成差异化缓存Key，导致不同job无法共享同一缓存。你可以指定基于依赖锁文件的固定Key，确保只有依赖变更时才更新缓存，同时给不需要修改缓存的job设置拉取模式：

stages:
    - build
    - quality

cache:
    key:
        files:
            - package-lock.json  # 或npm-shrinkwrap.json，根据你的项目选择
        prefix: npm-shared-cache
    paths:
        - node_modules/
    policy: pull-push  # 默认策略，build阶段推送缓存，后续job拉取

build-one:
  image: node:latest
  stage: build
  script:
    - npm install <some package>

build-two:
  image: foo_image:latest
  stage: build
  script:
    - some cmd
  cache:
    policy: pull  # 该job无需修改缓存，仅拉取避免覆盖

quality:
  image: node:latest
  stage: quality
  script:
    - <some cmd using the previously installed package>

方案2：用Artifacts传递依赖 #

如果缓存机制仍有问题，可以改用Artifacts在流水线job间传递node_modules目录，Artifacts不受runner或镜像差异影响：

stages:
    - build
    - quality

build-one:
  image: node:latest
  stage: build
  script:
    - npm install <some package>
  artifacts:
    paths:
        - node_modules/
    expire_in: 1h  # 设置过期时间，避免占用过多存储

build-two:
  image: foo_image:latest
  stage: build
  script:
    - some cmd

quality:
  image: node:latest
  stage: quality
  dependencies:
    - build-one  # 明确拉取build-one生成的Artifacts
  script:
    - <some cmd using the previously installed package>

方案3：统一缓存目录权限 #

不同Docker镜像的运行用户可能不同（比如node:latest默认用node用户，其他镜像可能用root），会导致缓存目录权限不匹配无法读取。可以在build阶段放宽目录权限：

build-one:
  image: node:latest
  stage: build
  script:
    - npm install <some package>
    - chmod -R 777 node_modules/  # 确保后续job能读取缓存目录

或者在quality job中切换到对应执行用户：

quality:
  image: node:latest
  stage: quality
  before_script:
    - su node  # 切换到node用户执行脚本，匹配缓存目录权限
  script:
    - <some cmd using the previously installed package>

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