在本地获取 sagemaker 容器

Question

当我尝试在脚本模式下为 tensorflow 在本地运行 sagemaker 时。似乎我无法拉动 docker 容器。我已经从 sagemaker 笔记本实例运行了下面的代码，一切运行良好。但是在我的机器上运行它就不行了。

如何下​​载容器，以便在本地调试？

import os

import sagemaker
from sagemaker.tensorflow import TensorFlow


hyperparameters = {}
role = 'arn:aws:iam::xxxxxxxx:role/yyyyyyy'
estimator = TensorFlow(
    entry_point='train.py',
    source_dir='.',
    train_instance_type='local',
    train_instance_count=1,
    hyperparameters=hyperparameters,
    role=role,
    py_version='py3',
    framework_version='1.12.0',
    script_mode=True)

estimator.fit()

我得到这个输出

INFO:sagemaker:Creating training-job with name: sagemaker-tensorflow-
scriptmode-2019-01-28-18-51-57-787
WARNING! Using --password via the CLI is insecure. Use --password-stdin.
Error response from daemon: pull access denied for 520713654638.dkr.ecr.eu-west-2.amazonaws.com/sagemaker-tensorflow-scriptmode, repository does not exist or may require 'docker login'

subprocess.CalledProcessError: Command 'docker pull 520713654638.dkr.ecr.eu-west-2.amazonaws.com/sagemaker-tensorflow-scriptmode:1.12.0-cpu-py3' returned non-zero exit status 1.

警告看起来像您在使用 docker login stuff here 时得到的输出。如果我按照这些步骤注册到带有 tensorflow 容器的目录，它会显示登录成功

Invoke-Expression -Command (aws ecr get-login --no-include-email --registry-ids 520713654638 --region eu-west-2)
WARNING! Using --password via the CLI is insecure. Use --password-stdin.
Login Succeeded

然后我还是拉不出来

docker pull 520713654638.dkr.ecr.eu-west-2.amazonaws.com/sagemaker-tensorflow-scriptmode:1.11.0-cpu-py3
Error response from daemon: pull access denied for 520713654638.dkr.ecr.eu-west-2.amazonaws.com/sagemaker-tensorflow-scriptmode, repository does not exist or may require 'docker login'

Answer 1

相同的顺序在本地对我有效：'aws ecr get-login'、'docker login'、'docker pull'。

您的本地 IAM 用户是否有足够的凭证从 ECR 中提取？ 'AmazonEC2ContainerRegistryReadOnly' 策略应该足够了：https://docs.aws.amazon.com/AmazonECR/latest/userguide/ecr_managed_policies.html

或者，您可以从 Github 获取容器并构建它：https://github.com/aws/sagemaker-tensorflow-container