镜像加速服务

基本使用方式

对于学术训练常用的开源镜像仓库，英博云在每个可用区内部署了镜像加速服务，将原始镜像镜像地址替换为加速地址，可以实现镜像拉取加速，具体规则如下：

GCR

Google Container Registry

替换规则：将gcr.io替换为：registry-{region-label}-internal.ebcloud.com/gcr.io

示例（以华北一区为例）：

# 原始命令：
docker pull gcr.io/cadvisor/cadvisor:v0.39.3

# 替换为：
docker pull registry-cn-huabei1-internal.ebcloud.com/gcr.io/cadvisor/cadvisor:v0.39.3

GHCR

Github Container Registry

替换规则：将ghcr.io替换为：registry-{region-label}-internal.ebcloud.com/ghcr.io

示例（以华北一区为例）：

# 原始命令：
docker pull ghcr.io/github/super-linter:latest

# 替换为：
docker pull registry-cn-huabei1-internal.ebcloud.com/ghcr.io/github/super-linter:latest

Quay

Quay Container Registry

替换规则：将quay.io替换为：registry-{region-label}-internal.ebcloud.com/quay.io

示例（以华北一区为例）：

# 原始命令：
docker pull quay.io/prometheus/prometheus:latest

# 替换为：
docker pull registry-cn-huabei1-internal.ebcloud.com/quay.io/prometheus/prometheus:latest

K8S

Kubernetes's container image registry

替换规则：将registry.k8s.io替换为：registry-{region-label}-internal.ebcloud.com/registry.k8s.io

示例（以华北一区为例）：

# 原始命令：
docker pull registry.k8s.io/pause:3.10

# 替换为：
docker pull registry-cn-huabei1-internal.ebcloud.com/registry.k8s.io/pause:3.10

NGC

NVIDIA NGC Catalog

替换规则：将nvcr.io替换为：registry-{region-label}-internal.ebcloud.com/nvcr.io

示例（以华北一区为例）：

# 原始命令：
docker pull nvcr.io/nvidia/cuda

# 替换为：
docker pull registry-cn-huabei1-internal.ebcloud.com/nvcr.io/nvidia/cuda

提交K8S工作负载自动加速

用yaml文件提交工作负载时，若命中了上述开源加速镜像，英博云会自动进行镜像地址替换，用户无需关注。

例如，若用户在华北一区提交如下工作负载：

# distroless.yaml
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: distroless-demo
spec:
  replicas: 1
  selector:
    matchLabels:
      app: distroless-demo
  template:
    metadata:
      labels:
        app: distroless-demo
    spec:
      containers:
      - name: demo
        image: gcr.io/distroless/static-debian12:debug-nonroot
        command: ["/busybox/sleep", "3600"]

英博云会自动将镜像地址：gcr.io/distroless/static-debian12:debug-nonroot，替换为：registry-cn-huabei1-internal.ebcloud.com/gcr.io/distroless/static-debian12:debug-nonroot

部署上述deployment后，kubectl describe pod查看pod事件，可以看到如下输出：

kubectl apply -f distroless.yaml

kubectl get pods
NAME                               READY   STATUS    RESTARTS   AGE
distroless-demo-5cff56ffbb-d7tp8   1/1     Running   0          6s

kubectl describe pod distroless-demo-5cff56ffbb-d7tp8
···
  Normal   Pulling    3m25s  kubelet     Pulling image "registry-cn-huabei1-internal.ebcloud.com/gcr.io/distroless/static-debian12:debug-nonroot"
  Normal   Pulled     3m18s  kubelet     Successfully pulled image "registry-cn-huabei1-internal.ebcloud.com/gcr.io/distroless/static-debian12:debug-nonroot" in 7.192s (7.192s including waiting). Image size: 3254726 bytes.
  Normal   Created    3m18s  kubelet     Created container demo
  Normal   Started    3m18s  kubelet     Started container demo 

可以看到镜像地址：gcr.io已经完成了加速替换。