前言

在使用 Docker 和 Kubernetes 时,我们经常需要访问 gcr.ioquay.io 镜像仓库,由于众所周知的原因,这些镜像仓库在中国都无法访问,唯一能访问的是 Docker Hub,但速度也是奇慢无比。gcr.azk8s.cngcr.io 镜像仓库的代理站点,原来可以通过 gcr.azk8s.cn 访问 gcr.io 仓库里的镜像,但是目前 *.azk8s.cn 已经仅限于 Azure 中国的 IP 使用,不再对外提供服务了。国内其他的镜像加速方案大多都是采用定时同步的方式来缓存,这种方法是有一定延迟的,不能保证及时更新,ustc 和七牛云等镜像加速器我都试过了,非常不靠谱,很多镜像都没有。

为了能够顺利访问 gcr.io 等镜像仓库,我们需要在墙外自己搭建一个类似于 gcr.azk8s.cn 的镜像仓库代理站点。利用 Docker 的开源项目 registry[1] 就可以实现这个需求,registry 不仅可以作为本地私有镜像仓库,还可以作为上游镜像仓库的缓存,也就是  pull through cache

先来感受下速度:



gcr镜像 gcr镜像站_etcd

gcr镜像 gcr镜像站_gcr镜像_02

gcr镜像 gcr镜像站_docker_03

gcr镜像 gcr镜像站_gcr镜像_04

1. 前提条件

  • 一台能够施展魔法的服务器(你懂得,可以直接访问 gcr.io)
  • 一个域名和域名相关的 SSL 证书(docker pull 镜像时需要验证域名证书),一般用 Let's Encrypt[2] 就够了。

2. 核心思路

registry 可以通过设置参数 remoteurl 将其作为远端仓库的缓存仓库,这样当你通过这个私有仓库的地址拉取镜像时,regiistry 会先将镜像缓存到本地存储,然后再提供给拉取的客户端(有可能这两个步骤是同时的,我也不太清楚)。我们可以先部署一个私有 registry,然后将 remoteurl 设为需要加速的镜像仓库地址,基本上就可以了。

3. 定制 registry

为了能够支持缓存 docker.iogcr.iok8s.gcr.ioquay.ioghcr.io 等常见的公共镜像仓库,我们需要对 registry 的配置文件进行定制,Dockerfile 如下:

FROM registry:2.6
LABEL maintainer="registry-proxy Docker Maintainers https://fuckcloudnative.io"
ENV PROXY_REMOTE_URL="" \
    DELETE_ENABLED=""
COPY entrypoint.sh /entrypoint.sh

其中 entrypoint.sh 用来将环境变量传入配置文件:

entrypoint.sh

#!/bin/sh

set -e

CONFIG_YML=/etc/docker/registry/config.yml

if [ -n "$PROXY_REMOTE_URL" -a `grep -c "$PROXY_REMOTE_URL" $CONFIG_YML` -eq 0 ]; then
    echo "proxy:" >> $CONFIG_YML
    echo "  remoteurl: $PROXY_REMOTE_URL" >> $CONFIG_YML
    echo "  username: $PROXY_USERNAME" >> $CONFIG_YML
    echo "  password: $PROXY_PASSWORD" >> $CONFIG_YML
    echo "------ Enabled proxy to remote: $PROXY_REMOTE_URL ------"
elif [ $DELETE_ENABLED = true -a `grep -c "delete:" $CONFIG_YML` -eq 0 ]; then
    sed -i '/rootdirectory/a\  delete:' $CONFIG_YML
    sed -i '/delete/a\    enabled: true' $CONFIG_YML
    echo "------ Enabled local storage delete -----"
fi

sed -i "/headers/a\    Access-Control-Allow-Origin: ['*']" $CONFIG_YML
sed -i "/headers/a\    Access-Control-Allow-Methods: ['HEAD', 'GET', 'OPTIONS', 'DELETE']" $CONFIG_YML
sed -i "/headers/a\    Access-Control-Expose-Headers: ['Docker-Content-Digest']" $CONFIG_YML

case "$1" in
    *.yaml|*.yml) set -- registry serve "$@" ;;
    serve|garbage-collect|help|-*) set -- registry "$@" ;;
esac

exec "$@"

4. 启动缓存服务

构建好 Docker 镜像之后,就可以启动服务了。如果你不想自己构建,可以直接用我的镜像:yangchuansheng/registry-proxy

一般来说,即使你要同时缓存 docker.iogcr.iok8s.gcr.ioquay.ioghcr.io,一台 1C 2G 的云主机也足够了(前提是你不在上面跑其他的服务)。我的博客、评论服务和其他一堆乱七八糟的服务都要跑在云主机上,所以一台是不满足我的需求的,我直接买了两台腾讯云香港轻量级服务器。



gcr镜像 gcr镜像站_docker_05

既然买了两台,肯定得????组个 k3s 集群啦,看主机名就知道我是用来干啥的。其中 2C 4G 作为 master 节点,1C 2G 作为 node 节点。

docker.io 为例,创建资源清单:

dockerhub.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: dockerhub
  labels:
    app: dockerhub
spec:
  replicas: 1
  selector:
    matchLabels:
      app: dockerhub
  template:
    metadata:
      labels:
        app: dockerhub
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: app
                  operator: In
                  values:
                  - dockerhub
              topologyKey: kubernetes.io/hostname
            weight: 1
      dnsPolicy: None
      dnsConfig:
        nameservers:
          - 8.8.8.8
          - 8.8.4.4
      containers:
      - name: dockerhub
        image: yangchuansheng/registry-proxy:latest
        env:
        - name: PROXY_REMOTE_URL
          value: https://registry-1.docker.io
        - name: PROXY_USERNAME
          value: yangchuansheng
        - name: PROXY_PASSWORD
          value: ********
        ports:
        - containerPort: 5000
          protocol: TCP
        volumeMounts:
        - mountPath: /etc/localtime
          name: localtime
        - mountPath: /var/lib/registry
          name: registry
      volumes:
      - name: localtime
        hostPath:
          path: /etc/localtime
      - name: registry
        hostPath:
          path: /var/lib/registry
---
apiVersion: v1
kind: Service
metadata:
  name: dockerhub
  labels:
    app: dockerhub
spec:
  selector:
    app: dockerhub
  ports:
    - protocol: TCP
      name: http
      port: 5000
      targetPort: 5000

使用资源清单创建对应的服务:

????  → kubectl apply -f dockerhub.yaml

如果你只有一台主机,可以使用 docker-compose 来编排容器,配置文件可以自己参考 k8s 的配置修改,本文就不赘述了。

5. 代理选择

如果只缓存 docker.io,可以直接将 registry-proxy 的端口改成 443,并添加 SSL 证书配置。如果要缓存多个公共镜像仓库,就不太推荐这么做了,因为 443 端口只有一个,多个 registry-proxy 服务不能共用一个端口,合理的做法是使用边缘代理服务根据域名来转发请求到不同的 registry-proxy 服务。

对于 Kubernetes 集群来说,Ingress Controller 即边缘代理,常见的 Ingress Controller 基本上都是由 Nginx 或者 Envoy 来实现。Envoy 虽为代理界新秀,但生而逢时,它的很多特性都是原生为云准备的,是真正意义上的 Cloud Native L7 代理和通信总线。比如它的服务发现和动态配置功能,与 Nginx 等代理的热加载不同,Envoy 可以通过 API 来实现其控制平面,控制平面可以集中服务发现,并通过 API 接口动态更新数据平面的配置,不需要重启数据平面的代理。不仅如此,控制平面还可以通过 API 将配置进行分层,然后逐层更新。

目前使用 Envoy 实现的 Ingress Controller 有 ????Contour 和 Gloo[3] 等,如果你对 Envoy 比较感兴趣,并且想使用 Ingress Controller 作为边缘代理,可以试试 ????Contour。Ingress Controller 对底层做了抽象,屏蔽了很多细节,无法顾及到所有细节的配置,必然不会支持底层代理所有的配置项,所以我选择使用原生的 Envoy 来作为边缘代理。如果你是单机跑的 registry-proxy 服务,也可以试试 Envoy。

6. 代理配置

首先创建 Envoy 的资源清单:

envoy.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: envoy
  namespace: kube-system
  labels:
    app: envoy
spec:
  replicas: 2
  selector:
    matchLabels:
      app: envoy
  strategy:
    rollingUpdate:
      maxSurge: 0
      maxUnavailable: 1
    type: RollingUpdate
  template:
    metadata:
      labels:
        app: envoy
    spec:
      hostNetwork: true
      dnsPolicy: ClusterFirstWithHostNet
      containers:
      - name: envoy
        image: envoyproxy/envoy:v1.17-latest
        imagePullPolicy: IfNotPresent
        command:
        - envoy
        - /etc/envoy/envoy.yaml
        ports:
        - containerPort: 443
          name: https
        - containerPort: 80
          name: http
        - containerPort: 15001
          name: http-metrics
        volumeMounts:
        - mountPath: /etc/localtime
          name: localtime
        - mountPath: /etc/envoy
          name: envoy
        - mountPath: /root/.acme.sh/fuckcloudnative.io
          name: ssl
      volumes:
      - name: localtime
        hostPath:
          path: /etc/localtime
      - name: ssl
        hostPath:
          path: /root/.acme.sh/fuckcloudnative.io
      - name: envoy
        hostPath:
          path: /etc/envoy

使用资源清单创建对应的服务:

????  → kubectl apply -f envoy.yaml

这里选择使用 hostPath 将 envoy 的配置挂载到容器中,来看下 Envoy 的配置,先进入 /etc/envoy 目录。

bootstrap 配置:

envoy.yaml

node:
  id: node0
  cluster: cluster0
dynamic_resources:
  lds_config:
    path: /etc/envoy/lds.yaml
  cds_config:
    path: /etc/envoy/cds.yaml
admin:
  access_log_path: "/dev/stdout"
  address:
    socket_address:
      address: "0.0.0.0"
      port_value: 15001

LDS 的配置:

lds.yaml

version_info: "0"
resources:
- "@type": type.googleapis.com/envoy.config.listener.v3.Listener
  name: listener_http
  address:
    socket_address:
      address: 0.0.0.0
      port_value: 80
  filter_chains:
  - filters:
    - name: envoy.filters.network.http_connection_manager
      typed_config:
        "@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
        stat_prefix: ingress_http
        codec_type: AUTO
        access_log:
          name: envoy.access_loggers.file
          typed_config:
            "@type": type.googleapis.com/envoy.extensions.access_loggers.file.v3.FileAccessLog
            path: /dev/stdout
        route_config:
          name: http_route
          virtual_hosts:
          - name: default
            domains:
            - "*"
            routes:
            - match:
                prefix: "/"
              redirect:
                https_redirect: true
                port_redirect: 443
                response_code: "FOUND"
        http_filters:
        - name: envoy.filters.http.router
- "@type": type.googleapis.com/envoy.config.listener.v3.Listener
  name: listener_https
  address:
    socket_address:
      address: 0.0.0.0
      port_value: 443
  listener_filters:
  - name: "envoy.filters.listener.tls_inspector"
    typed_config: {}
  filter_chains:
  - transport_socket:
      name: envoy.transport_sockets.tls
      typed_config:
        "@type": type.googleapis.com/envoy.extensions.transport_sockets.tls.v3.DownstreamTlsContext
        common_tls_context:
          alpn_protocols: h2,http/1.1
          tls_certificates:
          - certificate_chain:
              filename: "/root/.acme.sh/fuckcloudnative.io/fullchain.cer"
            private_key:
              filename: "/root/.acme.sh/fuckcloudnative.io/fuckcloudnative.io.key"
    filters:
    - name: envoy.filters.network.http_connection_manager
      typed_config:
        "@type": type.googleapis.com/envoy.extensions.filters.network.http_connection_manager.v3.HttpConnectionManager
        stat_prefix: ingress_https
        codec_type: AUTO
        use_remote_address: true
        access_log:
          name: envoy.access_loggers.file
          typed_config:
            "@type": type.googleapis.com/envoy.extensions.access_loggers.file.v3.FileAccessLog
            path: /dev/stdout
        route_config:
          name: https_route
          response_headers_to_add:
          - header:
              key: Strict-Transport-Security
              value: "max-age=15552000; includeSubdomains; preload"
          virtual_hosts:
          - name: docker
            domains:
            - docker.fuckcloudnative.io
            routes:
            - match:
                prefix: "/"
              route:
                cluster: dockerhub
                timeout: 600s
        http_filters:
        - name: envoy.filters.http.router
          typed_config:
            "@type": type.googleapis.com/envoy.extensions.filters.http.router.v3.Router

CDS 的配置:

cds.yaml

version_info: "0"
resources:
- "@type": type.googleapis.com/envoy.config.cluster.v3.Cluster
  name: dockerhub
  connect_timeout: 15s
  type: strict_dns
  dns_lookup_family: V4_ONLY
  lb_policy: ROUND_ROBIN
  load_assignment:
    cluster_name: dockerhub
    endpoints:
    - lb_endpoints:
      - endpoint:
          address:
            socket_address:
              address: dockerhub.default
              port_value: 5000

这里的 address 使用的是 Kubernetes 集群内部域名,其他部署方式请自己斟酌。

配置好了 Envoy 之后,就可以通过代理服务器拉取 docker.io 的镜像了。

7. 验证加速效果

现在你就可以通过代理服务器来拉取公共镜像了。比如你想拉取 nginx:alpine 镜像,可以使用下面的命令:

????  → docker pull docker.fuckcloudnative.io/library/nginx:alpine

alpine: Pulling from library/nginx
801bfaa63ef2: Pull complete
b1242e25d284: Pull complete
7453d3e6b909: Pull complete
07ce7418c4f8: Pull complete
e295e0624aa3: Pull complete
Digest: sha256:c2ce58e024275728b00a554ac25628af25c54782865b3487b11c21cafb7fabda
Status: Downloaded newer image for docker.fuckcloudnative.io/library/nginx:alpine
docker.fuckcloudnative.io/library/nginx:alpine

8. 缓存所有镜像仓库

前面的示例只是缓存了 docker.io,如果要缓存所有的公共镜像仓库,可以参考 4-6 节的内容。以 k8s.gcr.io 为例,先准备一个资源清单:

gcr-k8s.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: gcr-k8s
  labels:
    app: gcr-k8s
spec:
  replicas: 1
  selector:
    matchLabels:
      app: gcr-k8s
  template:
    metadata:
      labels:
        app: gcr-k8s
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: app
                  operator: In
                  values:
                  - gcr-k8s
              topologyKey: kubernetes.io/hostname
            weight: 1
      dnsPolicy: None
      dnsConfig:
        nameservers:
          - 8.8.8.8
          - 8.8.4.4
      containers:
      - name: gcr-k8s
        image: yangchuansheng/registry-proxy:latest
        env:
        - name: PROXY_REMOTE_URL
          value: https://k8s.gcr.io
        ports:
        - containerPort: 5000
          protocol: TCP
        volumeMounts:
        - mountPath: /etc/localtime
          name: localtime
        - mountPath: /var/lib/registry
          name: registry
      volumes:
      - name: localtime
        hostPath:
          path: /etc/localtime
      - name: registry
        hostPath:
          path: /var/lib/registry
---
apiVersion: v1
kind: Service
metadata:
  name: gcr-k8s
  labels:
    app: gcr-k8s
spec:
  selector:
    app: gcr-k8s
  ports:
    - protocol: TCP
      name: http
      port: 5000
      targetPort: 5000

将其部署到 Kubernetes 集群中:

????  → kubectl apply -f gcr-k8s.yaml

lds.yaml 中添加相关配置:

virtual_hosts:
          - name: docker
            ...
            ...
          - name: k8s
            domains:
            - k8s.fuckcloudnative.io
            routes:
            - match:
                prefix: "/"
              route:
                cluster: gcr-k8s
                timeout: 600s

cds.yaml 中添加相关配置:

- "@type": type.googleapis.com/envoy.config.cluster.v3.Cluster
  name: gcr-k8s
  connect_timeout: 1s
  type: strict_dns
  dns_lookup_family: V4_ONLY
  lb_policy: ROUND_ROBIN
  load_assignment:
    cluster_name: gcr-k8s
    endpoints:
    - lb_endpoints:
      - endpoint:
          address:
            socket_address:
              address: gcr-k8s.default
              port_value: 5000

其他镜像仓库可照搬上述步骤,以下是我自己跑的所有缓存服务容器:

????  → kubectl get pod -o wide

gcr-8647ffb586-67c6g                     1/1     Running   0          21h     10.42.1.52    blog-k3s02
ghcr-7765f6788b-hxxvc                    1/1     Running   0          21h     10.42.1.55    blog-k3s01
dockerhub-94bbb7497-x4zwg                1/1     Running   0          21h     10.42.1.54    blog-k3s02
gcr-k8s-644db84879-7xssb                 1/1     Running   0          21h     10.42.1.53    blog-k3s01
quay-559b65848b-ljclb                    1/1     Running   0          21h     10.42.0.154   blog-k3s01

9. 容器运行时配置

配置好所有的缓存服务后,就可以通过代理来拉取公共镜像了,只需按照下面的列表替换镜像地址中的字段就行了:

原 URL

替换后的 URL

docker.io/xxx/xxx 或 xxx/xxx

docker.fuckcloudnative.io/xxx/xxx

docker.io/library/xxx 或 xxx

docker.fuckcloudnative.io/library/xxx

gcr.io/xxx/xxx

gcr.fuckcloudnative.io/xxx/xxx

k8s.gcr.io/xxx/xxx

k8s.fuckcloudnative.io/xxx/xxx

quay.io/xxx/xxx

quay.fuckcloudnative.io/xxx/xxx

ghcr.io/xxx/xxx

ghcr.fuckcloudnative.io/xxx/xxx

当然,最好的方式还是直接配置 registry mirror,Docker 只支持配置 docker.io 的 registry mirror,ContainerdPodman 支持配置所有镜像仓库的 registry mirror。

Docker

Docker 可以修改配置文件 /etc/docker/daemon.json,添加下面的内容:

{
    "registry-mirrors": [
        "https://docker.fuckcloudnative.io"
    ]
}

然后重启 Docker 服务,就可以直接拉取 docker.io 的镜像了,不需要显示指定代理服务器的地址,Docker 服务本身会自动通过代理服务器去拉取镜像。比如:

???? → docker pull nginx:alpine
???? → docker pull docker.io/library/nginx:alpine

Containerd

Containerd 就比较简单了,它支持任意 registry 的 mirror,只需要修改配置文件 /etc/containerd/config.toml,添加如下的配置:

[plugins."io.containerd.grpc.v1.cri".registry]
      [plugins."io.containerd.grpc.v1.cri".registry.mirrors]
        [plugins."io.containerd.grpc.v1.cri".registry.mirrors."docker.io"]
          endpoint = ["https://docker.fuckcloudnative.io"]
        [plugins."io.containerd.grpc.v1.cri".registry.mirrors."k8s.gcr.io"]
          endpoint = ["https://k8s.fuckcloudnative.io"]
        [plugins."io.containerd.grpc.v1.cri".registry.mirrors."gcr.io"]
          endpoint = ["https://gcr.fuckcloudnative.io"]
        [plugins."io.containerd.grpc.v1.cri".registry.mirrors."ghcr.io"]
          endpoint = ["https://ghcr.fuckcloudnative.io"]
        [plugins."io.containerd.grpc.v1.cri".registry.mirrors."quay.io"]
          endpoint = ["https://quay.fuckcloudnative.io"]

重启 Containerd 服务后,就可以直接拉取所有镜像了,不需要修改任何前缀,Containerd 会根据配置自动选择相应的代理 URL 拉取镜像。

Podman

Podman 也支持任意 registry 的 mirror,只需要修改配置文件 /etc/containers/registries.conf,添加如下的配置:

unqualified-search-registries = ['docker.io', 'k8s.gcr.io', 'gcr.io', 'ghcr.io', 'quay.io']

[[registry]]
prefix = "docker.io"
insecure = true
location = "registry-1.docker.io"

[[registry.mirror]]
location = "docker.fuckcloudnative.io"

[[registry]]
prefix = "k8s.gcr.io"
insecure = true
location = "k8s.gcr.io"

[[registry.mirror]]
location = "k8s.fuckcloudnative.io"

[[registry]]
prefix = "gcr.io"
insecure = true
location = "gcr.io"

[[registry.mirror]]
location = "gcr.fuckcloudnative.io"

[[registry]]
prefix = "ghcr.io"
insecure = true
location = "ghcr.io"

[[registry.mirror]]
location = "ghcr.fuckcloudnative.io"

[[registry]]
prefix = "quay.io"
insecure = true
location = "quay.io"

[[registry.mirror]]
location = "quay.fuckcloudnative.io"

然后就可以直接拉取所有镜像了,不需要修改任何前缀,Podman 会根据配置自动选择相应的代理 URL 拉取镜像。而且 Podman 还有 fallback 机制,上面的配置表示先尝试通过 registry.mirrorlocation 字段的 URL 来拉取镜像,如果失败就会尝试通过 registry 中 location 字段的 URL 来拉取。

10. 清理缓存

缓存服务会将拉取的镜像缓存到本地,所以需要消耗磁盘容量。一般云主机的磁盘容量都不是很大,OSS 和 s3 存储都比较贵,不太划算。

为了解决这个问题,我推荐定期删除缓存到本地磁盘的部分镜像,或者删除所有镜像。方法也比较简单,单独再部署一个 registry,共用其他 registry 的存储,并启用 delete 功能,然后再通过 API 或者 Dashboard 进行删除。

先准备一个资源清单:

reg-local.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: reg-local
  labels:
    app: reg-local
spec:
  replicas: 1
  selector:
    matchLabels:
      app: reg-local
  template:
    metadata:
      labels:
        app: reg-local
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: app
                  operator: In
                  values:
                  - reg-local
              topologyKey: kubernetes.io/hostname
            weight: 1
      containers:
      - name: reg-local
        image: yangchuansheng/registry-proxy:latest
        env:
        - name: DELETE_ENABLED
          value: "true"
        ports:
        - containerPort: 5000
          protocol: TCP
        volumeMounts:
        - mountPath: /etc/localtime
          name: localtime
        - mountPath: /var/lib/registry
          name: registry
      volumes:
      - name: localtime
        hostPath:
          path: /etc/localtime
      - name: registry
        hostPath:
          path: /var/lib/registry
---
apiVersion: v1
kind: Service
metadata:
  name: reg-local
  labels:
    app: reg-local
spec:
  selector:
    app: reg-local
  ports:
    - protocol: TCP
      name: http
      port: 5000
      targetPort: 5000

将其部署到 Kubernetes 集群中:

????  → kubectl apply -f reg-local.yaml

再准备一个 Docker Registry UI 的资源清单:

registry-ui

apiVersion: apps/v1
kind: Deployment
metadata:
  name: registry-ui
  labels:
    app: registry-ui
spec:
  replicas: 1
  selector:
    matchLabels:
      app: registry-ui
  template:
    metadata:
      labels:
        app: registry-ui
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: app
                  operator: In
                  values:
                  - registry-ui
              topologyKey: kubernetes.io/hostname
            weight: 1
      tolerations:
      - key: node-role.kubernetes.io/ingress
        operator: Exists
        effect: NoSchedule
      containers:
      - name: registry-ui
        image: joxit/docker-registry-ui:static
        env:
        - name: REGISTRY_TITLE
          value: My Private Docker Registry
        - name: REGISTRY_URL
          value: "http://reg-local:5000"
        - name: DELETE_IMAGES
          value: "true"
        ports:
        - containerPort: 80
          protocol: TCP
        volumeMounts:
        - mountPath: /etc/localtime
          name: localtime
      volumes:
      - name: localtime
        hostPath:
          path: /etc/localtime
---
apiVersion: v1
kind: Service
metadata:
  name: registry-ui
  labels:
    app: registry-ui
spec:
  selector:
    app: registry-ui
  ports:
    - protocol: TCP
      name: http
      port: 80
      targetPort: 80

将其部署到 Kubernetes 集群中:

????  → kubectl apply -f registry-ui.yaml

这样就可以通过 Dashboard 来清理镜像释放空间了。



gcr镜像 gcr镜像站_docker_06

或者直接简单粗暴,定时删除整个存储目录的内容。例如,执行命令 crontab -e,添加如下内容:

* * */2 * * /usr/bin/rm -rf /var/lib/registry/* &>/dev/null

表示每过两天清理一次 /var/lib/registry/ 目录。

11. 防白嫖认证

最后还有一个问题,我把缓存服务的域名全部公开了,如果大家都来白嫖,我的云主机肯定承受不住。为了防止白嫖,我得给 registry-proxy 加个认证,最简单的方法就是使用 basic auth,用 htpasswd 来存储密码。

  1. 为用户 admin 创建一个密码文件,密码为 admin
???? → docker run \
  --entrypoint htpasswd \
  registry:2.6 -Bbn admin admin > htpasswd
  1. 创建 Secret:
???? → kubectl create secret generic registry-auth --from-file=htpasswd
  1. 修改资源清单的配置,以 docker.io 为例:
apiVersion: apps/v1
kind: Deployment
metadata:
  name: dockerhub
  labels:
    app: dockerhub
spec:
  replicas: 1
  selector:
    matchLabels:
      app: dockerhub
  template:
    metadata:
      labels:
        app: dockerhub
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: app
                  operator: In
                  values:
                  - dockerhub
              topologyKey: kubernetes.io/hostname
            weight: 1
      dnsPolicy: None
      dnsConfig:
        nameservers:
          - 8.8.8.8
          - 8.8.4.4
      containers:
      - name: dockerhub
        image: yangchuansheng/registry-proxy:latest
        env:
        - name: PROXY_REMOTE_URL
          value: https://registry-1.docker.io
        - name: PROXY_USERNAME
          value: yangchuansheng
        - name: PROXY_PASSWORD
          value: ********
+       - name: REGISTRY_AUTH_HTPASSWD_REALM
+         value: Registry Realm
+       - name: REGISTRY_AUTH_HTPASSWD_PATH
+         value: /auth/htpasswd 
        ports:
        - containerPort: 5000
          protocol: TCP
        volumeMounts:
        - mountPath: /etc/localtime
          name: localtime
        - mountPath: /var/lib/registry
          name: registry
+       - mountPath: /auth
+         name: auth
      volumes:
      - name: localtime
        hostPath:
          path: /etc/localtime
      - name: registry
        hostPath:
          path: /var/lib/registry
+     - name: auth
+       secret:
+         secretName: registry-auth

apply 使其生效:

???? → kubectl apply -f dockerhub.yaml
  1. 尝试拉取镜像:
???? → docker pull docker.fuckcloudnative.io/library/nginx:latest
   
Error response from daemon: Get https://docker.fuckcloudnative.io/v2/library/nginx/manifests/latest: no basic auth credentials
  1. 登录镜像仓库:
???? → docker login docker.fuckcloudnative.io
Username: admin
Password:
WARNING! Your password will be stored unencrypted in /root/.docker/config.json.
Configure a credential helper to remove this warning. See
https://docs.docker.com/engine/reference/commandline/login/#credentials-store

Login Succeeded

现在就可以正常拉取镜像了。

如果你想更细粒度地控制权限,可以使用 Token 的方式来进行认证,具体可以参考 docker_auth[4] 这个项目。