非分阶段构建场景下，使用容器进行构建时，我们可以将容器中的缓存目录挂载到构建主机上，执行构建任务；然后将产物拷贝到运行镜像，制作应用镜像。但是在分阶段构建时，构建镜像和运行镜像在同一个 Dockerfile 中，这给优化第三方依赖的缓存带来了难度。
- 创建一个 Vue 实例项目
- 安装 Vue CLI
npm install -g @vue/cli --force
- 初始化示例项目
vue create hello-world
使用默认配置，创建示例项目: hello-world
- 运行项目
此时，项目已经包含全部依赖，可以直接运行项目:
npm run serve
- 删除依赖
依赖包通常不会提交到代码仓库，为了更好模拟构建情形，这里删除依赖，进行构建
rm -rf node_modules
- 项目中添加 Dockerfile 文件
进入项目目录:
cd hello-world
编辑并保存 Dockerfile 文件:
vim Dockerfile
FROM node:lts-alpine as builder
WORKDIR /
COPY package.json /
RUN npm install
COPY . .
RUN npm run build
FROM nginx:alpine
COPY --from=builder /dist/ /usr/share/nginx/html/
EXPOSE 80
- 构建镜像
执行命令：
docker build --no-cache -t shaowenchen/hello-world:v1 -f Dockerfile .
[+] Building 139.2s (13/13) FINISHED
=> [internal] load build definition from Dockerfile 2.6s
=> => transferring dockerfile: 228B 0.2s
=> [internal] load .dockerignore 3.4s
=> => transferring context: 2B 0.0s
=> [internal] load metadata for docker.io/library/nginx:alpine 4.2s
=> [internal] load metadata for docker.io/library/node:lts-alpine 4.3s
=> CACHED [builder 1/6] FROM docker.io/library/node:lts-alpine@sha256:2c6c59cf4d34d4f937ddfcf33bab9d8bbad8658d1b9de7b97622566a52167f2b 0.0s
=> [internal] load build context 1.8s
=> => transferring context: 5.03kB 0.4s
=> CACHED [stage-1 1/2] FROM docker.io/library/nginx:alpine@sha256:da9c94bec1da829ebd52431a84502ec471c8e548ffb2cedbf36260fd9bd1d4d3 0.0s
=> [builder 2/6] COPY package.json / 5.3s
=> [builder 3/6] RUN npm install 93.1s
=> [builder 4/6] COPY . . 5.9s
=> [builder 5/6] RUN npm run build 13.6s
=> [stage-1 2/2] COPY --from=builder /dist/ /usr/share/nginx/html/ 4.0s
=> exporting to image 4.0s
=> => exporting layers 2.3s
=> => writing image sha256:dc0f72b655eb95235b51d8fb30c430c3c1803c2d538d9948941f3e7afd23ab56 0.2s
=> => naming to docker.io/shaowenchen/hello-world:v1 0.2s
- 测试镜像
执行命令，创建容器:
docker run --rm -it -p 80:80 shaowenchen/hello-world:v1
- 访问服务
在本地打开: http://localhost, 可以看到页面
- 利用 Buildkit 挂载缓存优化
这种方式的思路是，将第三方包单独存储在一个缓存镜像中，当构建应用镜像时，将缓存镜像中的文件挂载到构建环境中。
2.1 开启 Buildkit
Buildkit 默认是关闭的。有两种方式打开 Buildkit:
- 第一种，在
/etc/docker/daemon.json中增加 buildkit 配置，{ "features": { "buildkit": true }}默认开启 buildkit 特性。 - 第二种，每次执行 docker 命令时，加上环境变量
DOCKER_BUILDKIT=1。
2.2 使用 Bind 的方式挂载缓存
- 准备缓存镜像的 Dockerfile
创建 Dockerfile 文件:
vim Dockerfile-Cache
FROM node:lts-alpine as builder
WORKDIR /
COPY . .
RUN npm install
RUN npm run build
这里有一个小细节就是，需要 npm run build编译第三方包。仅仅缓存第三方包，是不能获得很好的加速效果的。同时，预编译能减少 CPU 和内存的消耗。
- 编译包含第三方包的缓存镜像
docker build --no-cache -t shaowenchen/hello-world:cache -f Dockerfile-Cache .
[+] Building 111.9s (9/9) FINISHED
=> [internal] load build definition from Dockerfile-Cache 1.8s
=> => transferring dockerfile: 132B 0.0s
=> [internal] load .dockerignore 2.9s
=> => transferring context: 2B 0.0s
=> [internal] load metadata for docker.io/library/node:lts-alpine 4.2s
=> [internal] load build context 1.7s
=> => transferring context: 4.57kB 0.2s
=> CACHED [1/5] FROM docker.io/library/node:lts-alpine@sha256:2c6c59cf4d34d4f937ddfcf33bab9d8bbad8658d1b9de7b97622566a52167f2b 0.0s
=> [2/5] COPY . . 3.6s
=> [3/5] RUN npm install 69.2s
=> [4/5] RUN npm run build 14.5s
=> exporting to image 13.9s
=> => exporting layers 13.0s
=> => writing image sha256:e6ba7406f5d0c33d446ecc9a3c8e35fa593176ec9dedd899d39a1c00a14a5179 0.2s
=> => naming to docker.io/shaowenchen/hello-world:cache 0.2s
- 准备应用的构建 Dockerfile 文件
vim Dockerfile-Bind
FROM node:lts-alpine as builder
WORKDIR /
COPY . .
RUN --mount=type=bind,from=shaowenchen/hello-world:cache,source=/node_modules,target=/node_modules \
--mount=type=bind,from=shaowenchen/hello-world:cache,source=/root/.npm,target=/root/.npm npm install
RUN --mount=type=bind,from=shaowenchen/hello-world:cache,source=/node_modules,target=/node_modules \
--mount=type=bind,from=shaowenchen/hello-world:cache,source=/root/.npm,target=/root/.npm npm run build
FROM nginx:alpine
COPY --from=builder /dist/ /usr/share/nginx/html/
EXPOSE 80
在每个使用缓存的命令前面都需要加 --mount 。
- 编译应用镜像镜像
docker build --no-cache -t shaowenchen/hello-world:v1-bind -f Dockerfile-Bind .
[+] Building 55.3s (13/13) FINISHED
=> [internal] load build definition from Dockerfile-Bind 2.5s
=> => transferring dockerfile: 42B 0.0s
=> [internal] load .dockerignore 3.4s
=> => transferring context: 2B 0.0s
=> [internal] load metadata for docker.io/library/nginx:alpine 4.0s
=> [internal] load metadata for docker.io/library/node:lts-alpine 3.8s
=> [internal] load build context 2.4s
=> => transferring context: 4.47kB 0.2s
=> CACHED FROM docker.io/shaowenchen/hello-world:cache 0.3s
=> CACHED [stage-1 1/2] FROM docker.io/library/nginx:alpine@sha256:da9c94bec1da829ebd52431a84502ec471c8e548ffb2cedbf36260fd9bd1d4d3 0.0s
=> CACHED [builder 1/5] FROM docker.io/library/node:lts-alpine@sha256:2c6c59cf4d34d4f937ddfcf33bab9d8bbad8658d1b9de7b97622566a52167f2b 0.0s
=> [builder 2/5] COPY . . 4.2s
=> [builder 3/5] RUN --mount=type=bind,from=shaowenchen/hello-world:cache,source=/node_modules,target=/node_modules --mount=type=bind,from=shaowenchen/hello-world:cache 16.8s
=> [builder 4/5] RUN --mount=type=bind,from=shaowenchen/hello-world:cache,source=/node_modules,target=/node_modules --mount=type=bind,from=shaowenchen/hello-world:cache 13.2s
=> [stage-1 2/2] COPY --from=builder /dist/ /usr/share/nginx/html/ 3.7s
=> exporting to image 4.8s
=> => exporting layers 3.1s
=> => writing image sha256:de18663c5752a41cd61c23fb2cbbc1ac9c4c79cf5fdbe15ca16e806d0ce18d9d 0.2s
=> => naming to docker.io/shaowenchen/hello-world:v1-bind 0.1s
可以看到，加缓存之后，执行 install 和 build 总时长从 100 多秒降到了不到 30 秒。
- 利用 S3 存储缓存优化
3.1 快速部署一个 minio
参考文件: Jenkins 中的构建产物与缓存, https://www.chenshaowen.com/blog/artifacts-and-cache-in-jenkins.html
3.2 配置秘钥
在 hello-world 目录下创建凭证文件 .s3cfg 。
host_base = 1.1.1.1:9000
host_bucket = 1.1.1.1:9000
use_https = False
access_key = minio
secret_key = minio123
signature_v2 = False
3.3 改造 Dockerfile 适配 S3 缓存
这里主要的工作点在:
- 安装 s3cmd
- 获取并解压缓存，忽略错误（第一次为空）
- ... 安装依赖，进行构建
- 压缩并上传缓存
vim Dockerfile-S3
FROM node:lts-alpine as builder
ARG BUCKETNAME
ENV BUCKETNAME=$BUCKETNAME
RUN apk add python3 && ln -sf python3 /usr/bin/python && apk add py3-pip
RUN wget https://sourceforge.net/projects/s3tools/files/s3cmd/2.2.0/s3cmd-2.2.0.tar.gz \
&& mkdir -p /usr/local/s3cmd && tar -zxf s3cmd-2.2.0.tar.gz -C /usr/local/s3cmd \
&& ln -s /usr/local/s3cmd/s3cmd-2.2.0/s3cmd /usr/bin/s3cmd && pip3 install python-dateutil
WORKDIR /
# Get Cache
COPY .s3cfg /root/
RUN s3cmd get s3://$BUCKETNAME/node_modules.tar.gz && tar xf node_modules.tar.gz || exit 0
RUN s3cmd get s3://$BUCKETNAME/npm.tar.gz && tar xf npm.tar.gz || exit 0
COPY . .
RUN npm install
RUN npm run build
# Uploda Cache
RUN s3cmd del s3://$BUCKETNAME/node_modules.tar.gz || exit 0
RUN s3cmd del s3://$BUCKETNAME/npm.tar.gz || exit 0
RUN tar cvfz node_modules.tar.gz node_modules
RUN tar cvfz npm.tar.gz ~/.npm
RUN s3cmd put node_modules.tar.gz s3://$BUCKETNAME/
RUN s3cmd put npm.tar.gz s3://$BUCKETNAME/
FROM nginx:alpine
COPY --from=builder /dist/ /usr/share/nginx/html/
EXPOSE 80
- 首次使用 S3 缓存构建应用镜像
构建之前，需要提前创建一个名为 hello-world 的 Bucket。
docker build --no-cache --build-arg BUCKETNAME="hello-world" -t shaowenchen/hello-world:v1-s3 -f Dockerfile-S3 .
[+] Building 244.7s (23/23) FINISHED
=> [internal] load build definition from Dockerfile-S3 1.7s
=> => transferring dockerfile: 40B 0.1s
=> [internal] load .dockerignore 2.6s
=> => transferring context: 2B 0.1s
=> [internal] load metadata for docker.io/library/nginx:alpine 2.6s
=> [internal] load metadata for docker.io/library/node:lts-alpine 0.0s
=> CACHED [builder 1/16] FROM docker.io/library/node:lts-alpine 0.0s
=> [internal] load build context 2.2s
=> => transferring context: 4.53kB 0.1s
=> CACHED [stage-1 1/2] FROM docker.io/library/nginx:alpine@sha256:da9c94bec1da829ebd52431a84502ec471c8e548ffb2cedbf36260fd9bd1d4d3 0.0s
=> [builder 2/16] RUN apk add python3 && ln -sf python3 /usr/bin/python && apk add py3-pip 32.3s
=> [builder 3/16] RUN wget https://sourceforge.net/projects/s3tools/files/s3cmd/2.2.0/s3cmd-2.2.0.tar.gz && mkdir -p /usr/local/s3cmd && tar -zxf s3cmd-2.2.0.tar.g 12.8s
=> [builder 4/16] COPY .s3cfg /root/ 5.8s
=> [builder 5/16] RUN s3cmd get s3://hello-world/node_modules.tar.gz && tar xf node_modules.tar.gz || exit 0 6.7s
=> [builder 6/16] RUN s3cmd get s3://hello-world/npm.tar.gz && tar xf npm.tar.gz || exit 0 7.3s
=> [builder 7/16] COPY . . 5.7s
=> [builder 8/16] RUN npm install 71.3s
=> [builder 9/16] RUN npm run build 14.4s
=> [builder 10/16] RUN s3cmd del s3://hello-world/node_modules.tar.gz || exit 0 7.5s
=> [builder 11/16] RUN s3cmd del s3://hello-world/npm.tar.gz || exit 0 6.9s
=> [builder 12/16] RUN tar cvfz node_modules.tar.gz node_modules 11.3s
=> [builder 13/16] RUN tar cvfz npm.tar.gz ~/.npm 9.4s
=> [builder 14/16] RUN s3cmd put node_modules.tar.gz s3://hello-world/ 14.8s
=> [builder 15/16] RUN s3cmd put npm.tar.gz s3://hello-world/ 15.9s
=> [stage-1 2/2] COPY --from=builder /dist/ /usr/share/nginx/html/ 4.5s
=> exporting to image 3.9s
=> => exporting layers 2.5s
=> => writing image sha256:dceead698b2c5f3980bf17f246078fe967dda2d9b009c30d9fdb0c60263146e5 0.1s
=> => naming to docker.io/shaowenchen/hello-world:v1-s3 0.2s
在 Minio 的 UI 端可以看到相关的缓存文件：
- 再次使用 S3 缓存构建应用镜像
docker build --no-cache --build-arg BUCKETNAME="hello-world" -t shaowenchen/hello-world:v1-s3 -f Dockerfile-S3 .
[+] Building 213.8s (23/23) FINISHED
=> [internal] load build definition from Dockerfile-S3 2.0s
=> => transferring dockerfile: 40B 0.0s
=> [internal] load .dockerignore 2.7s
=> => transferring context: 2B 0.0s
=> [internal] load metadata for docker.io/library/nginx:alpine 4.6s
=> [internal] load metadata for docker.io/library/node:lts-alpine 0.0s
=> CACHED [builder 1/16] FROM docker.io/library/node:lts-alpine 0.0s
=> CACHED [stage-1 1/2] FROM docker.io/library/nginx:alpine@sha256:da9c94bec1da829ebd52431a84502ec471c8e548ffb2cedbf36260fd9bd1d4d3 0.0s
=> [internal] load build context 1.9s
=> => transferring context: 4.53kB 0.1s
=> [builder 2/16] RUN apk add python3 && ln -sf python3 /usr/bin/python && apk add py3-pip 30.9s
=> [builder 3/16] RUN wget https://sourceforge.net/projects/s3tools/files/s3cmd/2.2.0/s3cmd-2.2.0.tar.gz && mkdir -p /usr/local/s3cmd && tar -zxf s3cmd-2.2.0.tar.g 13.2s
=> [builder 4/16] COPY .s3cfg /root/ 5.5s
=> [builder 5/16] RUN s3cmd get s3://hello-world/node_modules.tar.gz && tar xf node_modules.tar.gz || exit 0 16.7s
=> [builder 6/16] RUN s3cmd get s3://hello-world/npm.tar.gz && tar xf npm.tar.gz || exit 0 15.3s
=> [builder 7/16] COPY . . 4.7s
=> [builder 8/16] RUN npm install 18.4s
=> [builder 9/16] RUN npm run build 13.6s
=> [builder 10/16] RUN s3cmd del s3://hello-world/node_modules.tar.gz || exit 0 7.4s
=> [builder 11/16] RUN s3cmd del s3://hello-world/npm.tar.gz || exit 0 7.9s
=> [builder 12/16] RUN tar cvfz node_modules.tar.gz node_modules 10.8s
=> [builder 13/16] RUN tar cvfz npm.tar.gz ~/.npm 10.0s
=> [builder 14/16] RUN s3cmd put node_modules.tar.gz s3://hello-world/ 17.9s
=> [builder 15/16] RUN s3cmd put npm.tar.gz s3://hello-world/ 16.3s
=> [stage-1 2/2] COPY --from=builder /dist/ /usr/share/nginx/html/ 5.0s
=> exporting to image 3.8s
=> => exporting layers 2.5s
=> => writing image sha256:a9c46eef6073b3ef8e6c4cd33cc1ed11c94dcebdb0883c89283883d9434de331 0.2s
=> => naming to docker.io/shaowenchen/hello-world:v1-s3 0.2s
可以看到，install 和 build 命令大约需要 80 秒，但是 S3 缓存相关的操作占用了大约 50 秒。
其中的 80 秒还可以优化的地方是，构建环境和 S3 服务之间网络限速为 1.2 MB/S 导致拉取和推送占用时间过长，就有较大优化空间。我认为在 30 秒以内，比较合理。
- 总结
缓存加速是 CI 产品的一个难点。用户使用的方式各不相同，我们能做的是针对用户的场景提供解决方案，而不能强制改变用户的使用习惯。在我之前开发的 CI 产品中，主要是将主机上的缓存挂载到构建环境中加速，无法适用分阶段构建的场景。这里主要提供了两个方案:
第一种，开启 Buildkit 特性，将第三方依赖包存储在缓存镜像。缓存镜像可以根据策略，定时进行更新。构建镜像时，挂载缓存镜像中的第三方包。
第二种，使用 S3 存储第三方依赖包，在构建时，使用 s3cmd 命令管理缓存。
以上两种方式，都不算很好。主要的原因是，它们都需要对 Dockerfile 进行修改，对业务的入侵较大。
- 参考