你当前正在访问 Microsoft Azure Global Edition 技术文档网站。 如果需要访问由世纪互联运营的 Microsoft Azure 中国技术文档网站,请访问 https://docs.azure.cn

在 Batch 池中预配 Linux 计算节点

可以使用 Azure Batch 在 Linux 和 Windows 虚拟机上运行并行计算工作负荷。 本文详细介绍如何使用 Batch PythonBatch .NET 客户端库在 Batch 服务中创建 Linux 计算节点池。

虚拟机配置

在 Batch 中创建计算节点池时,可以使用两个选项来选择节点大小和操作系统:“云服务配置”和“虚拟机配置”。 虚拟机配置池由 Azure VM 组成,这些 VM 可以从 Linux 或 Windows 映像中创建。 使用虚拟机配置创建池时,请指定可用的计算节点大小、要安装在节点上的虚拟机映像引用以及 Batch 节点代理 SKU(在每个节点上运行并提供节点和 Batch 服务之间接口的程序)。

虚拟机映像引用

Batch 服务使用虚拟机规模集提供虚拟机配置中的计算节点。 可指定 Azure 市场中的一个映像,或使用 Azure Compute Gallery 准备自定义映像

在创建虚拟机映像引用时,必须指定以下属性:

映像引用属性 示例
发布者 Canonical
产品/服务 UbuntuServer
SKU 20.04-LTS
版本 最新

提示

可以在使用 Azure CLI 在 Azure 市场中查找 Linux VM 映像中详细了解这些属性以及如何指定市场映像。 请注意,某些市场映像当前不与 Batch 兼容。

虚拟机映像列表

并非所有市场映像都与当前可用的 Batch 节点代理兼容。 若要列出 Batch 服务及其相应节点代理 SKU 支持的所有市场虚拟机映像,请使用 list_supported_images (Python)、ListSupportedImages (Batch .NET) 或其他语言 SDK 的相应 API。

节点代理 SKU

Batch 节点代理是一个程序,它在池中的每个节点上运行,并在节点与 Batch 服务之间提供命令和控制接口。 节点代理对于不同操作系统有不同的实现(称为 SKU)。 从根本上讲,在创建虚拟机配置时,需要先指定虚拟机映像引用,然后指定要在其上安装映像的代理节点。 通常,每个节点代理 SKU 与多个虚拟机映像兼容。 若要查看支持的节点代理 SKU 和虚拟机映像兼容性,可以使用 Azure Batch CLI 命令

az batch pool supported-images list

有关详细信息,请参阅帐户 - 列出支持的映像 - REST API(Azure Batch 服务)| Microsoft Docs

创建 Linux 池:Batch Python

下面的代码片段示范如何使用用于 Python 的 Microsoft Azure Batch 客户端库创建 Ubuntu Server 计算节点池。 有关 Batch Python 模块的更多详细信息,请查看参考文档

此代码片段显式创建 ImageReference,并指定它的每个属性(publisher、offer、SKU、version)。 但是,我们建议在生产代码中使用 list_node_agent_skus 方法在运行时从可用映像和节点代理 SKU 组合中做出选择。

# Import the required modules from the
# Azure Batch Client Library for Python
import azure.batch.batch_service_client as batch
import azure.batch.batch_auth as batchauth
import azure.batch.models as batchmodels

# Specify Batch account credentials
account = "<batch-account-name>"
key = "<batch-account-key>"
batch_url = "<batch-account-url>"

# Pool settings
pool_id = "LinuxNodesSamplePoolPython"
vm_size = "STANDARD_D2_V3"
node_count = 1

# Initialize the Batch client
creds = batchauth.SharedKeyCredentials(account, key)
config = batch.BatchServiceClientConfiguration(creds, batch_url)
client = batch.BatchServiceClient(creds, batch_url)

# Create the unbound pool
new_pool = batchmodels.PoolAddParameter(id=pool_id, vm_size=vm_size)
new_pool.target_dedicated = node_count

# Configure the start task for the pool
start_task = batchmodels.StartTask()
start_task.run_elevated = True
start_task.command_line = "printenv AZ_BATCH_NODE_STARTUP_DIR"
new_pool.start_task = start_task

# Create an ImageReference which specifies the Marketplace
# virtual machine image to install on the nodes
ir = batchmodels.ImageReference(
    publisher="Canonical",
    offer="UbuntuServer",
    sku="20.04-LTS",
    version="latest")

# Create the VirtualMachineConfiguration, specifying
# the VM image reference and the Batch node agent
# to install on the node
vmc = batchmodels.VirtualMachineConfiguration(
    image_reference=ir,
    node_agent_sku_id="batch.node.ubuntu 20.04")

# Assign the virtual machine configuration to the pool
new_pool.virtual_machine_configuration = vmc

# Create pool in the Batch service
client.pool.add(new_pool)

如上所述,建议使用 list_supported_images 方法从当前支持的节点代理/市场映像组合中进行动态选择(而不是显式创建 ImageReference)。 以下 Python 代码片段演示如何使用此方法。

# Get the list of supported images from the Batch service
images = client.account.list_supported_images()

# Obtain the desired image reference
image = None
for img in images:
  if (img.image_reference.publisher.lower() == "canonical" and
        img.image_reference.offer.lower() == "ubuntuserver" and
        img.image_reference.sku.lower() == "20.04-lts"):
    image = img
    break

if image is None:
  raise RuntimeError('invalid image reference for desired configuration')

# Create the VirtualMachineConfiguration, specifying the VM image
# reference and the Batch node agent to be installed on the node
vmc = batchmodels.VirtualMachineConfiguration(
    image_reference=image.image_reference,
    node_agent_sku_id=image.node_agent_sku_id)

创建 Linux 池:批处理 .NET

以下代码片段示范如何使用 Batch .NET 客户端库创建 Ubuntu Server 计算节点池。 有关 Batch .NET 的更多详细信息,请查看参考文档

以下代码片段使用 PoolOperations.ListSupportedImages 方法从当前支持的市场映像和节点代理 SKU 组合列表中进行选择。 建议使用这种方法,因为受支持组合的列表有时会发生变化。 通常情况下,添加支持的组合。

// Pool settings
const string poolId = "LinuxNodesSamplePoolDotNet";
const string vmSize = "STANDARD_D2_V3";
const int nodeCount = 1;

// Obtain a collection of all available node agent SKUs.
// This allows us to select from a list of supported
// VM image/node agent combinations.
List<ImageInformation> images =
    batchClient.PoolOperations.ListSupportedImages().ToList();

// Find the appropriate image information
ImageInformation image = null;
foreach (var img in images)
{
    if (img.ImageReference.Publisher == "Canonical" &&
        img.ImageReference.Offer == "UbuntuServer" &&
        img.ImageReference.Sku == "20.04-LTS")
    {
        image = img;
        break;
    }
}

// Create the VirtualMachineConfiguration for use when actually
// creating the pool
VirtualMachineConfiguration virtualMachineConfiguration =
    new VirtualMachineConfiguration(image.ImageReference, image.NodeAgentSkuId);

// Create the unbound pool object using the VirtualMachineConfiguration
// created above
CloudPool pool = batchClient.PoolOperations.CreatePool(
    poolId: poolId,
    virtualMachineSize: vmSize,
    virtualMachineConfiguration: virtualMachineConfiguration,
    targetDedicatedComputeNodes: nodeCount);

// Commit the pool to the Batch service
await pool.CommitAsync();

尽管上述代码片段是使用 PoolOperations.istSupportedImages 方法来动态列出支持的映像和节点代理 SKU 组合并从中做出选择(建议的做法),但也可以显式配置 ImageReference

ImageReference imageReference = new ImageReference(
    publisher: "Canonical",
    offer: "UbuntuServer",
    sku: "20.04-LTS",
    version: "latest");

使用 SSH 连接到 Linux 节点

在开发期间或进行故障排除时,可能会发现需要登录到池中的节点。 与 Windows 计算节点不同,你无法使用远程桌面协议 (RDP) 来连接到 Linux 节点。 相反,Batch 服务在每个节点上启用 SSH 访问以建立远程连接。

以下 Python 代码片段会在池中的每个节点上创建一个用户(远程连接时需要)。 然后列显每个节点的安全外壳 (SSH) 连接信息。

import datetime
import getpass
import azure.batch.batch_service_client as batch
import azure.batch.batch_auth as batchauth
import azure.batch.models as batchmodels

# Specify your own account credentials
batch_account_name = ''
batch_account_key = ''
batch_account_url = ''

# Specify the ID of an existing pool containing Linux nodes
# currently in the 'idle' state
pool_id = ''

# Specify the username and prompt for a password
username = 'linuxuser'
password = getpass.getpass()

# Create a BatchClient
credentials = batchauth.SharedKeyCredentials(
    batch_account_name,
    batch_account_key
)
batch_client = batch.BatchServiceClient(
    credentials,
    base_url=batch_account_url
)

# Create the user that will be added to each node in the pool
user = batchmodels.ComputeNodeUser(username)
user.password = password
user.is_admin = True
user.expiry_time = \
    (datetime.datetime.today() + datetime.timedelta(days=30)).isoformat()

# Get the list of nodes in the pool
nodes = batch_client.compute_node.list(pool_id)

# Add the user to each node in the pool and print
# the connection information for the node
for node in nodes:
    # Add the user to the node
    batch_client.compute_node.add_user(pool_id, node.id, user)

    # Obtain SSH login information for the node
    login = batch_client.compute_node.get_remote_login_settings(pool_id,
                                                                node.id)

    # Print the connection info for the node
    print("{0} | {1} | {2} | {3}".format(node.id,
                                         node.state,
                                         login.remote_login_ip_address,
                                         login.remote_login_port))

此代码将会有类似于以下示例的输出。 在本例中,该池包含四个 Linux 节点。

Password:
tvm-1219235766_1-20160414t192511z | ComputeNodeState.idle | 13.91.7.57 | 50000
tvm-1219235766_2-20160414t192511z | ComputeNodeState.idle | 13.91.7.57 | 50003
tvm-1219235766_3-20160414t192511z | ComputeNodeState.idle | 13.91.7.57 | 50002
tvm-1219235766_4-20160414t192511z | ComputeNodeState.idle | 13.91.7.57 | 50001

在节点上创建用户时不需要指定密码,而可以指定 SSH 公钥。

在 Python SDK 中,请在 ComputeNodeUser 上使用 ssh_public_key 参数。

在 .NET 中,请使用 ComputeNodeUser.SshPublicKey 属性。

定价

Azure Batch 构建在 Azure 云服务和 Azure 虚拟机技术基础之上。 Batch 服务本身是免费提供的,这意味着,只需支付 Batch 解决方案使用的计算资源费用(以及产生的相关费用)。 如果选择“虚拟机配置”,系统会根据虚拟机定价结构收费。

如果使用应用程序包将应用程序部署到 Batch 节点,系统还会对应用程序包使用的 Azure 存储资源收费。

后续步骤