容器服务Docker&Kubernetes 关注
手机版

Kubeflow实战系列: 利用TFJob运行分布式TensorFlow

  1. 云栖社区>
  2. 容器服务Docker&Kubernetes>
  3. 博客>
  4. 正文

Kubeflow实战系列: 利用TFJob运行分布式TensorFlow

必嘫 2018-06-14 08:13:46 浏览2207 评论0

摘要: TensorFlow作为现在最为流行的深度学习代码库,在数据科学家中间非常流行,特别是可以明显加速训练效率的分布式训练更是杀手级的特性。但是如何真正部署和运行大规模的分布式模型训练,却成了新的挑战。

介绍

本系列将介绍如何在阿里云容器服务上运行Kubeflow, 本文介绍如何使用TfJob运行分布式模型训练。

TensorFlow分布式训练和Kubernetes

TensorFlow作为现在最为流行的深度学习代码库,在数据科学家中间非常流行,特别是可以明显加速训练效率的分布式训练更是杀手级的特性。但是如何真正部署和运行大规模的分布式模型训练,却成了新的挑战。 实际分布式TensorFLow的使用者需要关心3件事情。

  1. 寻找足够运行训练的资源,通常一个分布式训练需要若干数量的worker(运算服务器)和ps(参数服务器),而这些运算成员都需要使用计算资源。
  2. 安装和配置支撑程序运算的软件和应用
  3. 根据分布式TensorFlow的设计,需要配置ClusterSpec。这个json格式的ClusterSpec是用来描述整个分布式训练集群的架构,比如需要使用两个worker和ps,ClusterSpec应该长成下面的样子,并且分布式训练中每个成员都需要利用这个ClusterSpec初始化tf.train.ClusterSpec对象,建立集群内部通信
cluster = tf.train.ClusterSpec({"worker": ["<VM_1>:2222",
                                           "<VM_2>:2222"],
                                "ps": ["<IP_VM_1>:2223",
                                       "<IP_VM_2>:2223"]})

其中第一件事情是Kubernetes资源调度非常擅长的事情,无论CPU和GPU调度,都是直接可以使用;而第二件事情是Docker擅长的,固化和可重复的操作保存到容器镜像。而自动化的构建ClusterSpecTFJob解决的问题,让用户通过简单的集中式配置,完成TensorFlow分布式集群拓扑的构建。

应该说烦恼了数据科学家很久的分布式训练问题,通过Kubernetes+TFJob的方案可以得到比较好的解决。

利用Kubernetes和TFJob部署分布式训练

  1. 修改TensorFlow分布式训练代码

之前在阿里云上小试TFJob一文中已经介绍了TFJob的定义,这里就不再赘述了。可以知道TFJob里有的角色类型为MASTER, WORKERPS

举个现实的例子,假设从事分布式训练的TFJob叫做distributed-mnist, 其中节点有1个MASTER, 2个WORKERS和2个PS,ClusterSpec对应的格式如下所示:

{  
    "master":[  
        "distributed-mnist-master-0:2222"
    ],
    "ps":[  
        "distributed-mnist-ps-0:2222",
        "distributed-mnist-ps-1:2222"
    ],
    "worker":[  
        "distributed-mnist-worker-0:2222",
        "distributed-mnist-worker-1:2222"
    ]
}

tf_operator的工作就是创建对应的5个Pod, 并且将环境变量TF_CONFIG传入到每个Pod中,TF_CONFIG包含三部分的内容,当前集群ClusterSpec, 该节点的角色类型,以及id。比如该Pod为worker0,它所收到的环境变量TF_CONFIG为:

{  
   "cluster":{  
      "master":[  
         "distributed-mnist-master-0:2222"
      ],
      "ps":[  
         "distributed-mnist-ps-0:2222"
      ],
      "worker":[  
         "distributed-mnist-worker-0:2222",
         "distributed-mnist-worker-1:2222"
      ]
   },
   "task":{  
      "type":"worker",
      "index":0
   },
   "environment":"cloud"
}

在这里,tf_operator负责将集群拓扑的发现和配置工作完成,免除了使用者的麻烦。对于使用者来说,他只需要在这里代码中使用通过获取环境变量TF_CONFIG中的上下文。

这意味着,用户需要根据和TFJob的规约修改分布式训练代码:

# 从环境变量TF_CONFIG中读取json格式的数据
tf_config_json = os.environ.get("TF_CONFIG", "{}")

# 反序列化成python对象
tf_config = json.loads(tf_config_json)

# 获取Cluster Spec
cluster_spec = tf_config.get("cluster", {})
cluster_spec_object = tf.train.ClusterSpec(cluster_spec)

# 获取角色类型和id, 比如这里的job_name 是 "worker" and task_id 是 0
task = tf_config.get("task", {})
job_name = task["type"]
task_id = task["index"]

# 创建TensorFlow Training Server对象
server_def = tf.train.ServerDef(
    cluster=cluster_spec_object.as_cluster_def(),
    protocol="grpc",
    job_name=job_name,
    task_index=task_id)
server = tf.train.Server(server_def)

# 如果job_name为ps,则调用server.join()
if job_name == 'ps':
    server.join()

# 检查当前进程是否是master, 如果是master,就需要负责创建session和保存summary。
is_chief = (job_name == 'master')


# 通常分布式训练的例子只有ps和worker两个角色,而在TFJob里增加了master这个角色,实际在分布式TensorFlow的编程模型并没有这个设计。而这需要使用TFJob的分布式代码里进行处理,不过这个处理并不复杂,只需要将master也看做worker_device的类型
with tf.device(tf.train.replica_device_setter(
    worker_device="/job:{0}/task:{1}".format(job_name,task_id),
    cluster=cluster_spec)):

具体代码可以参考示例代码

2. 在本例子中,将演示如何使用TFJob运行分布式训练,并且将训练结果和日志保存到NAS存储上,最后通过Tensorboard读取训练日志。

2.1 创建NAS数据卷,并且设置与当前Kubernetes集群的同一个具体vpc的挂载点。操作详见文档

2.2 在NAS上创建 /training的数据文件夹, 下载mnist训练所需要的数据

mkdir -p /nfs
mount -t nfs -o vers=4.0 xxxxxxx.cn-hangzhou.nas.aliyuncs.com:/ /nfs
mkdir -p /nfs/training
umount /nfs

2.3 创建NAS的PV, 以下为示例nas-dist-pv.yaml

apiVersion: v1
kind: PersistentVolume
metadata:
  name: kubeflow-dist-nas-mnist
  labels:
    tfjob: kubeflow-dist-nas-mnist
spec:
  capacity:
    storage: 10Gi
  accessModes:
    - ReadWriteMany
  storageClassName: nas
  flexVolume:
    driver: "alicloud/nas"
    options:
      mode: "755"
      path: /training
      server: xxxxxxx.cn-hangzhou.nas.aliyuncs.com
      vers: "4.0"

将该模板保存到nas-dist-pv.yaml, 并且创建pv:

# kubectl create -f nas-dist-pv.yaml
persistentvolume "kubeflow-dist-nas-mnist" created

2.4 利用nas-dist-pvc.yaml创建PVC

kind: PersistentVolumeClaim
apiVersion: v1
metadata:
  name: kubeflow-dist-nas-mnist
spec:
  storageClassName: nas
  accessModes:
    - ReadWriteMany
  resources:
    requests:
      storage: 5Gi
  selector:
    matchLabels:
      tfjob: kubeflow-dist-nas-mnist

具体命令:

# kubectl create -f nas-dist-pvc.yaml
persistentvolumeclaim "kubeflow-dist-nas-mnist" created

2.5 创建TFJob

apiVersion: kubeflow.org/v1alpha1
kind: TFJob
metadata:
  name: mnist-simple-gpu-dist
spec:
  replicaSpecs:
    - replicas: 1 # 1 Master
      tfReplicaType: MASTER
      template:
        spec:
          containers:
            - image: registry.aliyuncs.com/tensorflow-samples/tf-mnist-distributed:gpu
              name: tensorflow
              env:
              - name: TEST_TMPDIR
                value: /training
              command: ["python", "/app/main.py"]
              resources:
                limits:
                  nvidia.com/gpu: 1
              volumeMounts:
              - name: kubeflow-dist-nas-mnist
                mountPath: "/training"
          volumes:
            - name: kubeflow-dist-nas-mnist
              persistentVolumeClaim:
                claimName: kubeflow-dist-nas-mnist
          restartPolicy: OnFailure
    - replicas: 1 # 1 or 2 Workers depends on how many gpus you have
      tfReplicaType: WORKER
      template:
        spec:
          containers:
          - image: registry.aliyuncs.com/tensorflow-samples/tf-mnist-distributed:gpu                        
            name: tensorflow
            env:
            - name: TEST_TMPDIR
              value: /training
            command: ["python", "/app/main.py"]
            imagePullPolicy: Always
            resources:
              limits:
                nvidia.com/gpu: 1
            volumeMounts:
              - name: kubeflow-dist-nas-mnist
                mountPath: "/training"
          volumes:
            - name: kubeflow-dist-nas-mnist
              persistentVolumeClaim:
                claimName: kubeflow-dist-nas-mnist
          restartPolicy: OnFailure
    - replicas: 1  # 1 Parameter server
      tfReplicaType: PS
      template:
        spec:
          containers:
          - image: registry.aliyuncs.com/tensorflow-samples/tf-mnist-distributed:cpu                      
            name: tensorflow
            command: ["python", "/app/main.py"]
            env:
            - name: TEST_TMPDIR
              value: /training
            imagePullPolicy: Always
            volumeMounts:
              - name: kubeflow-dist-nas-mnist
                mountPath: "/training"
          volumes:
            - name: kubeflow-dist-nas-mnist
              persistentVolumeClaim:
                claimName: kubeflow-dist-nas-mnist
          restartPolicy: OnFailure

将该模板保存到mnist-simple-gpu-dist.yaml, 并且创建分布式训练的TFJob:

# kubectl create -f mnist-simple-gpu-dist.yaml
tfjob "mnist-simple-gpu-dist" created

检查所有运行的Pod

# RUNTIMEID=$(kubectl get tfjob mnist-simple-gpu-dist -o=jsonpath='{.spec.RuntimeId}')
# kubectl get po -lruntime_id=$RUNTIMEID
NAME                                        READY     STATUS    RESTARTS   AGE
mnist-simple-gpu-dist-master-z5z4-0-ipy0s   1/1       Running   0          31s
mnist-simple-gpu-dist-ps-z5z4-0-3nzpa       1/1       Running   0          31s
mnist-simple-gpu-dist-worker-z5z4-0-zm0zm   1/1       Running   0          31s

查看master的日志,可以看到ClusterSpec已经成功的构建出来了

# kubectl logs -l runtime_id=$RUNTIMEID,job_type=MASTER

2018-06-10 09:31:55.342689: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1105] Found device 0 with properties:
name: Tesla P100-PCIE-16GB major: 6 minor: 0 memoryClockRate(GHz): 1.3285
pciBusID: 0000:00:08.0
totalMemory: 15.89GiB freeMemory: 15.60GiB
2018-06-10 09:31:55.342724: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1195] Creating TensorFlow device (/device:GPU:0) -> (device: 0, name: Tesla P100-PCIE-16GB, pci bus id: 0000:00:08.0, compute capability: 6.0)
2018-06-10 09:31:55.805747: I tensorflow/core/distributed_runtime/rpc/grpc_channel.cc:215] Initialize GrpcChannelCache for job master -> {0 -> localhost:2222}
2018-06-10 09:31:55.805786: I tensorflow/core/distributed_runtime/rpc/grpc_channel.cc:215] Initialize GrpcChannelCache for job ps -> {0 -> mnist-simple-gpu-dist-ps-m5yi-0:2222}
2018-06-10 09:31:55.805794: I tensorflow/core/distributed_runtime/rpc/grpc_channel.cc:215] Initialize GrpcChannelCache for job worker -> {0 -> mnist-simple-gpu-dist-worker-m5yi-0:2222}
2018-06-10 09:31:55.807119: I tensorflow/core/distributed_runtime/rpc/grpc_server_lib.cc:324] Started server with target: grpc://localhost:2222
...

Accuracy at step 900: 0.9709
Accuracy at step 910: 0.971
Accuracy at step 920: 0.9735
Accuracy at step 930: 0.9716
Accuracy at step 940: 0.972
Accuracy at step 950: 0.9697
Accuracy at step 960: 0.9718
Accuracy at step 970: 0.9738
Accuracy at step 980: 0.9725
Accuracy at step 990: 0.9724
Adding run metadata for 999

2.6 部署TensorBoard,并且查看训练效果

为了更方便 TensorFlow 程序的理解、调试与优化,可以用 TensorBoard 来观察 TensorFlow 训练效果,理解训练框架和优化算法, 而TensorBoard通过读取TensorFlow的事件日志获取运行时的信息。

在之前的分布式训练样例中已经记录了事件日志,并且保存到文件events.out.tfevents*

# tree
.
└── tensorflow
    ├── input_data
    │   ├── t10k-images-idx3-ubyte.gz
    │   ├── t10k-labels-idx1-ubyte.gz
    │   ├── train-images-idx3-ubyte.gz
    │   └── train-labels-idx1-ubyte.gz
    └── logs
        ├── checkpoint
        ├── events.out.tfevents.1528760350.mnist-simple-gpu-dist-master-fziz-0-74je9
        ├── graph.pbtxt
        ├── model.ckpt-0.data-00000-of-00001
        ├── model.ckpt-0.index
        ├── model.ckpt-0.meta
        ├── test
        │   ├── events.out.tfevents.1528760351.mnist-simple-gpu-dist-master-fziz-0-74je9
        │   └── events.out.tfevents.1528760356.mnist-simple-gpu-dist-worker-fziz-0-9mvsd
        └── train
            ├── events.out.tfevents.1528760350.mnist-simple-gpu-dist-master-fziz-0-74je9
            └── events.out.tfevents.1528760355.mnist-simple-gpu-dist-worker-fziz-0-9mvsd

5 directories, 14 files

在Kubernetes部署TensorBoard, 并且指定之前训练的NAS存储

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  labels:
    app: tensorboard
  name: tensorboard
spec:
  replicas: 1
  selector:
    matchLabels:
      app: tensorboard
  template:
    metadata:
      labels:
        app: tensorboard
    spec:
      volumes:
      - name: kubeflow-dist-nas-mnist
        persistentVolumeClaim:
            claimName: kubeflow-dist-nas-mnist
      containers:
      - name: tensorboard
        image: tensorflow/tensorflow:1.7.0
        imagePullPolicy: Always
        command:
         - /usr/local/bin/tensorboard
        args:
        - --logdir
        - /training/tensorflow/logs
        volumeMounts:
        - name: kubeflow-dist-nas-mnist
          mountPath: "/training"
        ports:
        - containerPort: 6006
          protocol: TCP
      dnsPolicy: ClusterFirst
      restartPolicy: Always

将该模板保存到tensorboard.yaml, 并且创建tensorboard:

# kubectl create -f tensorboard.yaml
deployment "tensorboard" created

TensorBoard创建成功后,通过kubectl port-forward命令进行访问

PODNAME=$(kubectl get pod -l app=tensorboard -o jsonpath='{.items[0].metadata.name}')
kubectl port-forward ${PODNAME} 6006:6006

通过http://127.0.0.1:6006登录TensorBoard,查看分布式训练的模型和效果:

tensorboard-0.jpg

tensorboard-1.jpg

总结

利用tf-operator可以解决分布式训练的问题,简化数据科学家进行分布式训练工作。同时使用Tensorboard查看训练效果, 再利用NAS或者OSS来存放数据和模型,这样一方面有效的重用训练数据和保存实验结果,另外一方面也是为模型预测的发布做准备。如何把模型训练,验证,预测串联起来构成机器学习的工作流(workflow), 也是Kubeflow的核心价值,我们在后面的文章中也会进行介绍。

用云栖社区APP,舒服~

【云栖快讯】Apache旗下顶级开源盛会 HBasecon Asia 2018将于8月17日在京举行,现场仅600席,免费赠票领取入口  详情请点击

网友评论