Kubernetes GPU环境搭建
Kubernetes GPU 环境搭建适应场景k8s集群v1.13.0以上版本,调用GPU显卡计算资源,以支持TensorFlow,Caffe和PyTorch等AI应用。准备工作1.centos7系统上安装v1.13.0以上版本k8s集群,且服务器有nvidia显卡;2.安装nvidia显卡驱动,并确保显卡驱动版本与nvidia library的版本一致;搭建环境docker运行时更新repo源d
·
K8S GPU 资源调度
适应场景
K8S集群v1.13.0以上版本,调用GPU显卡计算资源,以支持基于TensorFlow,Caffe和PyTorch框架下的AI应用,如视频转码、人脸识别、发票识别和内容审核等。
准备工作
- centos7.5+系统上安装v1.13.0以上版本k8s集群,且机器有nvidia显卡;
- 安装nvidia显卡驱动,并确保显卡驱动版本与nvidia library的版本一致。
搭建环境
docker运行时
- 更新repo源
distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-container-runtime/$distribution/nvidia-container-runtime.repo | \
sudo tee /etc/yum.repos.d/nvidia-container-runtime.repo
- 安装gpu运行时
yum install nvidia-container-runtime
- 修改docker
vim /etc/docker/daemon.json
// daemon.json文件示例:
{
"graph": "/data/docker/runtime",
"insecure-registries": ["hub.xxx.xxx.com"],
"default-runtime": "nvidia",
"max-concurrent-downloads": 10,
"max-concurrent-uploads": 5,
"tls": false,
"log-level": "info",
"exec-root": "/data/docker/exec",
"storage-driver": "overlay2",
"runtimes": {
"nvidia": {
"path": "/usr/bin/nvidia-container-runtime",
"runtimeArgs": []
}
},
"hosts": ["unix:///var/run/docker.sock","tcp://0.0.0.0:6372"]
}
- 重启组件
systemctl restart docker kubelet kube-proxy
- 测试gpu
// 进入容器后,执行nvidia-smi,若无报错,则GPU依赖正常
docker run -it registry.cn-hangzhou.aliyuncs.com/docker_learning_aliyun/caffe:v1 /bin/bash
K8S GPU调度插件
- 安装nvidia-device-plugin插件
kubectl apply -f nvidia-device-plugin.yaml
// nvidia-device-plugin.yaml文件示例
# Copyright (c) 2019, NVIDIA CORPORATION. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
apiVersion: apps/v1
kind: DaemonSet
metadata:
name: nvidia-device-plugin-daemonset
namespace: kube-system
spec:
selector:
matchLabels:
name: nvidia-device-plugin-ds
updateStrategy:
type: RollingUpdate
template:
metadata:
# This annotation is deprecated. Kept here for backward compatibility
# See https://kubernetes.io/docs/tasks/administer-cluster/guaranteed-scheduling-critical-addon-pods/
annotations:
scheduler.alpha.kubernetes.io/critical-pod: ""
labels:
name: nvidia-device-plugin-ds
spec:
tolerations:
# This toleration is deprecated. Kept here for backward compatibility
# See https://kubernetes.io/docs/tasks/administer-cluster/guaranteed-scheduling-critical-addon-pods/
- key: CriticalAddonsOnly
operator: Exists
- key: nvidia.com/gpu
operator: Exists
effect: NoSchedule
# Mark this pod as a critical add-on; when enabled, the critical add-on
# scheduler reserves resources for critical add-on pods so that they can
# be rescheduled after a failure.
# See https://kubernetes.io/docs/tasks/administer-cluster/guaranteed-scheduling-critical-addon-pods/
priorityClassName: "system-node-critical"
# added by X.L.Xia
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: accelerator
operator: Exists
containers:
- image: nvidia/k8s-device-plugin:1.0.0-beta4
name: nvidia-device-plugin-ctr
securityContext:
allowPrivilegeEscalation: false
capabilities:
drop: ["ALL"]
volumeMounts:
- name: device-plugin
mountPath: /var/lib/kubelet/device-plugins
volumes:
- name: device-plugin
hostPath:
path: /var/lib/kubelet/device-plugins
- 标记GPU k8s-worker节点
kubectl label node xxx.xxx.xxx.xx accelerator=nvidia-tesla-t4
- 测试k8s gpu应用
// transcode-dep.yaml文件示例
apiVersion: apps/v1beta2
kind: Deployment
metadata:
name: transcode-dep
spec:
replicas: 1
selector:
matchLabels:
app: transcode
template:
metadata:
labels:
app: transcode
spec:
containers:
- image: jstranscodeserver:1.8.5
name: transcode-container
ports:
- containerPort: 80
name: http
resources:
limits:
nvidia.com/gpu: 1
// transcode-svc文件示例
apiVersion: v1
kind: Service
metadata:
labels:
app: transcode
name: transcode
spec:
type: NodePort
ports:
- name: http
port: 80
nodePort: 35001
targetPort: http
selector:
app: transcode
参考资料
1. nvidia-container-runtime
2. Docker - 基于NVIDIA-Docker的Caffe-GPU环境搭建
3. 从零开始入门 K8s | GPU 管理和 Device Plugin 工作机制
更多推荐
已为社区贡献5条内容
所有评论(0)