DocsICP FilingConsole
官方文档
首页

Create a DSW instance

更新时间：
复制 MD 格式
产品详情
我的收藏

Data Science Workshop (DSW) provides a cloud-based Integrated Development Environment (IDE) for AI development. Developers who are familiar with Notebook or VSCode can start model development quickly. This topic describes how to create a DSW instance and provides solutions to common issues that may occur during instance startup and release.

Quickly create a basic DSW instance

  1. Log on to the PAI console, select the target Region, click Workspaces in the left navigation pane, and then select the target workspace.

  2. In the left navigation pane, click Interactive Modeling (DSW) > Create Instance. Configure the following key parameters and leave the other parameters at their default values. For a complete list of console parameters, see List of All Console Parameters.

    Parameter

    Description

    Instance Name

    Example: dsw_test.

    Resource Type

    Select Public Resources, which uses the pay-as-you-go billing method.

    Instance Type

    Select an instance type, such as ecs.gn7i-c8g1.2xlarge (1 A10 GPU, 8 vCPUs, 30 GiB memory).

    If this instance type is out of stock, try another instance type from the list or switch to another region.
    Warning

    • If you use a DSW free trial resource plan, make sure that the selected instance type is covered by the plan (ecs.g6.xlarge, ecs.gn7i-c8g1.2xlarge, or ecs.gn6v-c8g1.2xlarge). If you select an unsupported instance type, its fees cannot be deducted. For more information, see Obtain, use, and release a free trial.

    • After your free quota is used up or expires, if the DSW instance is still in the "Running" state, the system automatically switches to the pay-as-you-go billing method and deducts fees from your account balance. Make sure to release the instance promptly.

    Image config

    Select Alibaba Cloud Image, search for and select modelscope:1.31.0-pytorch2.8.0-gpu-py311-cu124-ubuntu22.04 (Python version: 3.11, CUDA version: 12.4).

    We recommend that you use the ModelScope image because it offers good compatibility and includes a comprehensive set of third-party libraries.

    Click OK to create the instance. A status of Running indicates that the instance has been created successfully.

    If the instance fails to start, refer to DSW instance startup.

  3. On the DSW instance list page, click the Open button in the Actions column to open the DSW instance and start model development.

    For more information about the DSW instance interface features and how to stop, release, and change the configuration of the DSW instance, see Console access and management of DSW.

    Important

    • Hourly billing for a DSW instance created with public resources begins as soon as its status changes to Running, even if you do not open the WebIDE or run code.

    • Closing the browser or logging out does not stop the instance or the billing.

    • If you use a free trial resource plan, the billing method automatically switches to pay-as-you-go after the quota is used up. The instance is not automatically stopped.

  4. Stop the instance. When you have finished your development tasks and no longer need the DSW instance, return to the DSW instances page:

    • If you want to temporarily stop the instance, click the Stop button to the right of the instance. Note that if the system disk has been scaled out, billing for the system disk continues after the instance is stopped.

      Important

      By default, data is stored on a free cloud disk. If the instance is stopped for more than 15 days, the data on the cloud disk is deleted and cannot be recovered. This restriction does not apply if you have expanded the system disk.

    • If you no longer need the instance, click More > Delete to the right of the instance to delete it and permanently stop all billing. Before you delete the instance, back up any important data. Data from a deleted instance cannot be recovered.

Configuration for typical application scenarios

The basic configuration of a DSW instance may not meet your AI development needs. The following table summarizes configurations for typical application scenarios.

Scenario

Requirement/Pain Point

Key configurations

References

Persistently store code and data

The system disk of a DSW instance is temporary storage. Data is cleared when the instance is deleted or stopped for an extended period.

You need to save important files for a long time or share data among multiple instances.

You can use Dataset Mounting or Mount storage to mount cloud storage, such as Object Storage Service (OSS), to a specified directory within an instance.

Mount a dataset, or an OSS, NAS, or CPFS file system

Improve public network download speed

By default, DSW instances use a shared gateway. Because of bandwidth limits, the network speed may be insufficient when you download large files.

In the Network Information section, configure a VPC to use a Private Gateway. You also need to create a NAT Gateway and an Elastic IP Address (EIP) for the VPC.

Accelerate Internet access with a dedicated gateway

Develop remotely using SSH

You are accustomed to using local tools such as VSCode or PyCharm for development and debugging and do not want to be limited to a web IDE.

In the access configuration, select Enable SSH, enter your SSH Public Key, select the Access over Internet method, and associate an existing NAT Gateway and Elastic IP Address (EIP).

Connecting remotely via direct SSH

Access web services within the instance

Publish a web application that is running in the instance to the Internet so that it can be directly accessed or shared through a URL.

In the access configuration, add a Custom Services, configure the service port, and enable public network access. Then, add an inbound rule in the security group to allow traffic on this port.

Accessing services on an instance over the Internet

Complete list of console parameters

Basic Information

Parameter

Description

Instance Name

Configure the instance name based on the prompts on the interface.

Tag

Add tags to instances as needed. This helps you search for, locate, manage, and allocate costs for resources from multiple dimensions.

Resource Information

Parameter

Description

Resource Type

  • Public Resources: The billing method is pay-as-you-go, and they cannot be converted to a subscription.

    Note

    GPU card limit: When using public resources, each Alibaba Cloud account is limited to two GPU cards per region. An error may occur if your resource usage exceeds this limit. To increase the limit, submit a ticket.

    • Instance Type: You can select GPU, CPU, or free trial resources. For more information, see Instance Family.

      Warning

      • If you use a DSW free trial resource plan, make sure that the selected instance type is covered by the plan (ecs.g6.xlarge, ecs.gn7i-c8g1.2xlarge, or ecs.gn6v-c8g1.2xlarge). If you select an unsupported instance type, its fees cannot be deducted. For more information, see Obtain, use, and release a free trial.

      • After your free quota is used up or expires, if the DSW instance is still in the "Running" state, the system automatically switches to the pay-as-you-go billing method and deducts fees from your account balance. Make sure to release the instance promptly.

    • Bidding Purchase: You can use spot instances to reduce running costs. If the No preemptible instances in stock message is displayed, you can try to select other specifications.

      This parameter is currently supported only in the China (Hangzhou), China (Shanghai), China (Beijing), China (Ulanqab), China (Shenzhen), China (Guangzhou), Japan (Tokyo), and Singapore regions.

    • Driver Settings: You can set the driver version for GPU-accelerated instances that use public resources. Different GPU models support different major driver versions.

  • Resource Quota: The billing method is subscription.

    • Resource Quota: You can select General computing resources or Lingjun resources. If no resources are available, click Associate Resource Quota to configure a quota.

    • Instance Type: Set the GPU, CPU, and memory based on your actual requirements.

    • Priority: The priority range is from 1 to 9. A larger value indicates a higher priority.

    • Node-Specific Scheduling: You can select a specific node to create a DSW instance.

    • CPU Affinity: Allows you to bind processes in a container or pod to specific CPU cores. This prevents issues such as CPU cache misses and context switching, increases CPU utilization, and improves application performance. You can enable this feature for applications that are performance-sensitive or have high real-time requirements. This feature is available only in the China (Beijing) and China (Shenzhen) regions.

    • Driver Settings: When you use a GPU with a Lingjun resource quota, you can set the instance driver version. Different GPU models support different major driver versions.

Environment Information

Parameter

Description

Image Configuration

The following image types are supported:

  • Alibaba Cloud Image: PAI provides official runtime images for popular open source frameworks and Python versions. For example, pytorch:2.4.1-gpu-py312-cu124-ubuntu22.04 indicates a PyTorch version of 2.4.1, support for GPU-accelerated instances, a Python version of 3.12, and a CUDA version of 12.4.

    If you need a specific version dependency, you can search for keywords in the search box. For example, search for cu124 to find images with CUDA version 12.4.

  • Custom Image: You can use custom images that are added to PAI. The image repository must be public, or the image must be stored in Container Registry (ACR). For more information, see Custom images.

  • Image Address: You can specify an address for a custom image or an Alibaba Cloud Image that is accessible from the public network.

    • To improve image pulling speed, see Image acceleration.

    • If it is a private image address, click Enter Username and Password and configure the image repository username and password, or try temporarily setting the image repository to allow public pulling.

System Disk

This is used to store files during the development process. When Resource Type is set to Public Resources, or when Resource Quota is set to a subscription general computing resource (with at least 2 CPU cores and 4 GB of memory, or a GPU), each instance includes a 100 GiB free cloud disk as a system disk. The cloud disk can be expanded. For specific pricing, refer to the console.

Warning

  • If you only use the free quota cloud disk, its contents will be cleared if the instance is stopped for more than 15 days.

  • After expansion, the entire cloud disk (free + paid) is no longer subject to the 15-day stop-and-release restriction. However, the expanded portion will continue to incur fees until the instance is deleted.

  • Disks cannot be scaled in after being expanded. Expand the disk as needed.

  • When an instance is deleted, its cloud disk is released simultaneously. Make sure to back up necessary data before deleting the instance.

If you need persistent storage, you can configure Dataset Mounting or Mount storage.

Dataset Mounting

Can be used to store datasets that need to be read or to persistently store files during development. The following two dataset types are supported:

  • Custom Dataset: You can create a custom dataset to store data files required for training, with support for read-only settings and version selection.

  • Public Dataset: PAI provides built-in public datasets that support only the read-only mounting mode.

Mount Path: The path where the dataset is mounted in DSW, for example, /mnt/data. In your code, you can retrieve the dataset from this path.

Note

  • The mount paths for multiple datasets cannot be the same.

  • If you configure a CPFS dataset, you need to set the network configuration, and the selected VPC must be the same as the CPFS. Otherwise, the DSW instance may fail to be created.

  • When the resource group is a dedicated resource group, the first dataset must be a NAS dataset. It will be mounted to both your specified path and the default DSW working directory /mnt/workspace/.

For more information about mounting, see Mount a dataset, OSS, NAS, or CPFS.

Mount storage

You can also use storage mounting to store datasets that need to be read or to persistently store files during development.

For more information about mounting, see Mount a dataset, OSS, NAS, or CPFS.

Working Directory

This parameter can be configured when you select public resources. The working directory is the startup path for Notebook and the Web IDE. The default is /mnt/workspace.

Expand for more configurations

Parameter

Description

Custom Startup Script

Used to customize the environment or perform initialization tasks during instance startup. The custom script is executed after the image and resources are ready, but before development applications such as JupyterLab and Code Server start.

Note

  • 3-minute timeout: The custom script increases the instance startup time. The script has a timeout of 3 minutes. Do not perform long-running tasks such as image downloads in the custom script.

  • View script logs: After the instance starts, you can find the logs generated by the custom script in the /var/log/user-command/ path.

Environment Variable

Used for main container startup, system processes, and user processes. You can add custom environment variables or overwrite default system environment variables as needed.

Note: We recommend that you do not modify the following environment variables:

# Modifications will not take effect
USER_NAME # Will be overwritten by logic in the service

# System variables that are not recommended to be modified, as changes may affect normal use
JUPYTER_NAME: Constructed from instance information by default, can be used to modify the JupyterLab URL access path
JUPYTER_COMMAND: Jupyter startup command, default is 'lab' to start JupyterLab
JUPYTER_SERVER_ADDR: JupyterLab service listening address, default is 0.0.0.0
JUPYTER_SERVER_PORT: JupyterLab service listening port, default is 8088
JUPYTER_SERVER_AUTH: JupyterLab access password, default is empty
JUPYTER_SERVER_ROOT: Jupyter working directory, has lower priority than WORKSPACE_DIR
CODE_SERVER_ADDR: code-server service listening address, default is 0.0.0.0
CODE_SERVER_PORT: code-server service listening port, default is 8082
CODE_SERVER_AUTH: code-server access password, default is empty
WORKSPACE_DIR: The system sets this environment variable based on the working directory parameter set when the instance was created. It can change the startup directory for Jupyter and code-server. An error may occur if the path does not exist.

Advanced Configurations

Allows users to adjust certain secure kernel parameters required by their services. Currently, only Lingjun resource group instances support these settings. For parameter details, see the table below.

Advanced Configuration Parameter

Description

Example Value

VmMaxMapCount

Sets the maximum number of memory mapping (mmap) areas a process can have. The default value is 65530. Values less than 65530 are ineffective. Excessively high values may lead to wasted memory resources.

1024000

EnableNvidiaIBGDA

Enables the IBGDA feature when loading the GPU driver. This must be enabled to use the DeepEP feature.

true

EnableNvidiaGDRCopy

Installs the GDRCopy kernel module, currently version 2.4.4. This must be enabled to use the DeepEP feature.

true

VpmuFeature

When enabled, supports capabilities such as troubleshooting CPU hot spots with asys. Optional values are 0, 1, and 2. The default is 0 (disabled). 1 means partially enabled. 2 means fully enabled. Enabling vpmu may cause a performance loss, so enable it as needed.

2

EnableVcpuTier

Enables the rund core binding optimization. Default is false. This option allows RunD vCPUs to be promptly scheduled to runnable CPUs by the kernel. However, it may cause scheduling contention and interference if two vCPUs are scheduled on the same Hyper-Threading pair. The following conditions must also be met:

  1. Must be a full-card instance (no other instances or tasks on the node).

  2. Number of CPUs ≥ 4 × Number of GPUs.

  3. The number of CPUs must be even.

true

VcpuTierHighCPU

Used to manually specify Highcpu after enabling EnableVcpuTier. If not specified, the default value (number of GPUs per worker, such as 8 or 16) is used. The value must be in the range of [Number of GPUs, Number of CPUs / 2).

4

Network Information

Parameter

Description

VPC Settings

You can configure this parameter only when Resource Type is set to Public Resources.

When you want to use a DSW instance in a Virtual Private Cloud (VPC), you can create a VPC in the same region as the DSW instance and configure this parameter. You also need to configure a vSwitch and a Security Group. For details about configuration policies for different scenarios, see Network Configuration.

vSwitch

This parameter can be configured when a VPC is configured. A vSwitch is a subnet within a VPC. Your DSW instance and other cloud resources are connected to the vSwitch.

Security Group

This parameter must be configured when a VPC is configured. A security group is a virtual firewall for your DSW instance that controls all inbound and outbound network traffic.

Internet Access Gateway

The following configuration methods are supported:

  • Public Gateway: A public gateway has limited network bandwidth, and the network speed may be insufficient for high-concurrency access or large file downloads.

  • Private Gateway: To overcome the bandwidth limitations of the public gateway, you can create an Internet NAT gateway, attach elastic IP addresses (EIPs), and configure SNAT entries in the virtual private cloud (VPC) of DSW. For more information, see Increase the public network access speed using a dedicated gateway.

You can configure the following parameters only when you select a CPFS-type dataset for Mount Configuration:

  • Enable all options: This is disabled by default. The system will disable VPCs that are not connected to the CPFS dataset.

Note

If you select a CPFS dataset for the mounting configuration, you must configure a VPC, and the selected VPC must be the same as the CPFS.

Extended CIDR Block

You can configure this parameter after a vSwitch is configured. If the number of available IP addresses in the VPC is insufficient for business expansion or if the initial network planning was inadequate, you can use a secondary CIDR block to expand the VPC address space. For more information, see Expand a CIDR block using a secondary CIDR block.

Access Configuration

Parameter

Description

Enable SSH

This option enables you to remotely connect to the instance and can be configured after you select a virtual private cloud (VPC). When you enable this option, a Custom Services named SSH appears. If you configure a custom image, you must ensure that sshd is installed on the custom image.

SSH Public Key

After you enable the SSH Configuration switch, you can configure this parameter.

Note

To support both VPC logon and public network logon, you must add the public keys of multiple clients. Add the public keys one by one, each on a new line. You can add up to 10 public keys.

Service Access and Port Configuration

You can use this to configure SSH remote access or access services in an instance over the Internet.

  • Listener Port: The port on which the service in the DSW instance listens.

  • Service Access Method:

    • Access over VPC: This access method is enabled by default. You can access DSW services from other resources, such as ECS instances, within the same VPC.

    • Access over Internet: Enables public network access. You must also configure a NAT Gateway and an EIP.

  • Internet Access Port: The port that enables access from the public network.

Create Private Zone in VPC

Creates a built-in authoritative domain name (PrivateZone). You can use this domain name within the VPC to access services on the instance, such as SSH, without having to use the instance's IP address, which may change. Note that creating a built-in authoritative domain name incurs fees. For more information, see Alibaba Cloud DNS Billing.

Public Network Access

NLB:

  • NLB Instance Resource Group: Select the resource group that contains the NLB instance.

  • NLB Instance: Select an NLB instance in the same VPC as the DSW instance.

DNAT + EIP:

  • NAT Gateway: A NAT Gateway maps public network requests (EIP:port) to private DSW instances (private IP:port), enabling public network access to services within an instance.

  • EIP: A public IP address that enables access to services in an instance over the public network.

Roles and Permissions

Parameter

Description

Visibility

You can select Visible to the Instance Owner or Visible to Current Workspace.

Instance Owner

Only a workspace administrator can change the instance owner.

Show More

Parameter

Description

Instance RAM Role

When accessing other cloud resources from within a DSW instance, you can associate a RAM role with the instance. This method uses temporary STS credentials to access other cloud resources, eliminating the need to configure long-term AccessKeys and effectively reducing the risk of key leakage.

The instance RAM role can be configured as:

  • Default Roles of PAI: Grants permissions to access PAI services, MaxCompute, and OSS. If you use the temporary credentials issued by the PAI default role, you are granted the same permissions as the DSW instance owner when you access PAI services and MaxCompute tables. When you access OSS, you can access only the bucket that is configured as the default storage path for the current workspace.

  • Custom Roles: When you want customized permissions or more fine-grained permission management, you can configure a custom role.

  • Does Not Associate Role: If you want to directly access other cloud products using an AccessKey, you can choose not to associate a role.

For more information about configuring an instance RAM role, see Configure a DSW instance RAM role.

FAQ

DSW instance startup

Click to expand

Q: DSW instance fails to start

Troubleshooting: Click the DSW instance name and check the error message on the Events tab.

image

  • Your requested resource type [ecs.******] is not enough currently, please try other regions or other resource types

    • Cause: The selected instance type has insufficient inventory in the current region.

    • Solution: Try creating the instance again later, or select a different instance type or region.

  • Your resource usage has exceeded the default limitation. Please contact us via ticket system to raise the limitation.

    • Cause: Each Alibaba Cloud account is limited to 2 GPUs per region by default. This error occurs when the selected instance type exceeds this quota.

    • Solution: To request a quota increase, submit a ticket.

  • the available zone with vSwitch is out of stock or InternalError-ResourceAllocateFailed

    • Cause: When a VPC is configured, the associated vSwitch restricts resource allocation to its availability zone. This causes failures if that zone has insufficient capacity.

    • Solution:

      1. Try creating a vSwitch and a DSW instance in a different availability zone.

      2. Try selecting a different instance type for the DSW instance.

  • Sales of this resource are temporarily suspended in the specified zone. We recommend that you use the multi-zone creation function to avoid the risk of insufficient resource.

    Try one of the following:

    • Switch to a different region.

    • Change the instance type.

    • Start the instance during off-peak hours.

  • CommodityInstanceNotAvailableError: The commodity instance has been released due to an overdue balance. Please create a new instance.

    • Cause: The system automatically released the instance due to prolonged overdue payments.

    • Solution: Create a new instance.

  • The charge of current ECI instance has been stopped, but the related resources are still being cleaned.

    • Cause: Trial resources are shared and may experience high demand during peak hours, causing startup to exceed 30 minutes. If resources are not acquired within one hour, startup fails.

    • Solution: Try the following options:

      • Switch to a different region.

      • Change the instance type. You cannot change the type of a pending instance — stop it first.

      • Try again during off-peak hours.

      • If these methods do not resolve the issue, contact your account manager for assistance.

  • The cluster resources are fully utilized. Please try later or other regions.

    • Cause: The compute resources are fully occupied.

    • Solution: Try the following options:

      • Switch to a different region.

      • Change the instance type. Note that you cannot change the instance type of a pending instance. You must stop the instance before changing its type.

      • Try again during off-peak hours.

      • If these methods do not resolve the issue, contact your account manager for assistance.

  • Create ECI failed because the specified instance is out of stock. It is recommended to use the multi-zone creation function to avoid the risk of stockout.

    • Cause: The specified compute resource is out of stock.

    • Solution: Try the following options:

      • Switch to a different region.

      • Change the instance type. You cannot change the type of a pending instance — stop it first.

      • Try again during off-peak hours.

      • If these methods do not resolve the issue, contact your account manager for assistance.

  • back-off 10s restarting failed container=dsw-notebook pod

    • Cause: The system disk is full.

      Check the system disk usage:

      image

      image

    • Solution: Expand the system disk using Change Settings.

      image

      Important

      An expanded system disk incurs charges continuously, regardless of whether the instance is running. To stop all billing, delete the instance. Back up all necessary data before deletion.

  • Startup failed, with the message "Workspace member not found"

    This error indicates that the account you are logged in with is not a member of the target workspace. Contact your workspace administrator to add your account to the workspace.

  • failed to create containerd container: failed to prepare layer from archive: failed to validate archive quota ...

    • Cause: The image used to create the instance is too large for the system disk, resulting in insufficient space.

    • Solution: Open the instance details page and expand the system disk. Note that expanding the system disk incurs additional charges based on its capacity.

      image

  • InternalError-Failed to perform action, error: OperationDenied.NoStock: The resource is out of stock in the specified zone. Please try other types, or choose other regions and zones.

    The resource is temporarily out of stock in the specified availability zone. Try selecting a different instance type, or switch to another region and availability zone.

  • RISK.RISK_CONTROL_REJECTION

    Your account has been restricted by risk control. You must resolve this issue before you can create an instance.

  • Creation failure due to overdue payments

    If your account has an overdue balance, you cannot create a DSW instance. Vouchers cannot be used to offset overdue payments. To check your account balance, log in to the Billing Management console.

Q: Instance type is out of stock or quota is insufficient

Common error messages:

  • "Your requested resource type [ecs.******] is not enough currently" (insufficient resource inventory).

  • "Your resource usage has exceeded the default limitation" (exceeds the 2-GPU quota per region).

  • "The cluster resources are fully utilized" (compute resources are fully occupied).

Cause Analysis:

  • Insufficient public resource inventory

    • Public resources are shared among multiple users, so inventory may be tight during peak hours.

    • Specific GPU instance types, such as high-end GPUs, are more likely to be sold out.

    • Each account has a quota of 2 GPUs per region.

  • Insufficient dedicated resource quota

    • The purchased dedicated resource quota has been exhausted.

    • Ineffective quota allocation has resulted in an insufficient quota for a specific workspace.

Solutions:

  • Change the instance type: If the selected GPU type is out of stock, try a different GPU type.

  • Switch the region: In the upper-left corner of the PAI console, switch to another region and try to create an instance.

  • Increase the GPU quota: If you need to use public resources with more than 2 GPUs, submit a ticket.

  • Purchase a dedicated resource quota: For guaranteed resource availability, purchase a dedicated resource quota. Purchase general-purpose compute resources and Manage resource quotas.

Q: Execute a Python file on DSW startup

Yes, you can configure a Custom Startup Script when you create a DSW instance or by modifying an existing instance's configuration.

image

The custom script runs after the image and resources are prepared but before development applications (JupyterLab, Code Server) launch.

Note

  • 3-minute timeout: A custom startup script increases instance startup time and will time out after 3 minutes. Avoid including long-running tasks, such as downloading large images, in the script.

  • Log access: After the instance starts, you can find logs generated by the script in the /var/log/user-command/ directory.

Q: Cannot find my DSW instance

The overview page shows instances you created. Switch between instance types and regions to locate your instance.

image

Q: DSW page is blank or unresponsive

Blank pages, continuously loading Notebooks, or unresponsive Terminals are typically caused by your local environment. Try the following:

  1. Clear your browser's cache and try again.

  2. Access the page using your browser's incognito or private mode.

  3. Switch your network environment, for example, from a corporate network to a mobile hotspot, to rule out firewall restrictions.

  4. Try using a different browser, such as Chrome or Firefox.

Q: Data persistence on cloud disk instances

DSW instances that use a Disk for the system disk include instances created in public resource groups and general-purpose resource instances. Data persistence on the system disk for these instances is as follows:

  • Stopping the instance: Data might be lost. If the cloud disk has not been expanded and the instance remains stopped for more than 15 days, the data is erased and cannot be recovered. If the cloud disk has been expanded or the instance has been stopped for 15 days or less, the data is preserved.

  • Restarting the instance: Data is not lost. When an instance is stopped or restarted, all packages installed via pip, code files, and other data stored on the system disk are retained.

  • Changing the instance type: Data is not lost. Adjusting the instance type (such as CPU, memory, or GPU configuration) does not affect data on the system disk.

  • Changing the instance image: Some data may be lost. Mounted datasets and OSS data are unaffected, but system disk content may be reset. Back up your data before changing the image. Mount a dataset, OSS bucket, NAS file system, or CPFS file system.

For general-purpose resource instances that use Temporary Storage as the system disk, all data is lost when the instance is stopped, restarted, or reconfigured (by changing its instance type or image). This occurs even if the resource group is configured with a prepaid cloud disk.

Q: Recover an instance released due to inactivity

No, the data cannot be recovered. If a DSW instance created from public resources has a standard (non-expanded) system disk and remains stopped for more than 15 consecutive days, the system disk is automatically erased.

Q: DSW instance startup becomes slow

A large saved image can cause a gradual increase in startup time.

Q: When creating a DSW instance, I get an "out of stock" or "InternalError-ResourceAllocateFailed" error. Alternatively, the instance fails to start after I configure an SSH or public IP but succeeds after I remove the configuration. What is the reason?

This error usually occurs because of insufficient resources in the zone where the VPC vSwitch is located. When you specify a VPC and vSwitch while you create a DSW instance, the system limits resource scheduling to the zone where the vSwitch is located. If the resources in that zone are unavailable, the instance creation fails. The root cause is insufficient resources in that specific zone, not a network configuration conflict.

Common error messages:

  • out of stock

  • InternalError-ResourceAllocateFailed

  • Startup fails after you configure SSH access or a public IP, but succeeds after these configurations are removed.

Solution: Switch to a vSwitch that corresponds to another zone that has available resources. We recommend that you create vSwitches in multiple zones under the VPC to increase the probability of successful resource scheduling.

Q: My DSW instance cannot be restarted after being stopped. How can I check the restock time for public resources?

The inventory of public resources is updated dynamically every day, and you cannot query the specific restock time. When a DSW instance is manually stopped, the underlying compute resources are released. Restarting the instance is equivalent to reapplying for resources. If the resources for the original specification have been allocated to other users at the time of restart, the startup may fail.

Solution:

  • Switch to another region.

  • Switch to another available specification.

  • Retry during off-peak hours.

Q: When creating a DSW instance, I get a "Private zone service status is not OPENED" error. What should I do?

This error indicates that the Alibaba Cloud DNS PrivateZone service has not been activated. DSW relies on PrivateZone to complete VPC network resolution initialization.

Solution:

  1. Log on to the Alibaba Cloud Management Console, then search for and activate the Alibaba Cloud DNS PrivateZone service.

  2. Confirm that the VPC where the DSW instance is located is associated with this service.

  3. Recreate or restart the DSW instance to trigger network initialization.

Q: When creating a DSW instance, I get a "Workspace member not found" or "Workspace not exists" error. What should I do?

This error usually occurs for the following reasons:

  • The currently logged-in account is not a member of the target workspace.

  • The wrong workspace was selected.

Solution:

  • Switch to the correct workspace.

  • Contact the workspace administrator to add the current account as a workspace member.

Stopping or releasing a DSW instance

Click to expand

Q: How do I release a DSW instance?

On the DSW instance page, click Stop or Delete.

image

Important: If you expanded the system disk when you created the DSW instance, the system disk continues to incur charges regardless of whether the instance is running. To completely stop billing for the DSW instance, you must delete it.

Q: Why can't I find my DSW instance?

If you cannot find the instance, try switching to a different region or workspace.

image

Q: How do I release a free trial resource package?

You do not need to stop or release free trial resource packages.

Q: How do I stop all charges for a DSW instance? What is the difference between Stop and Delete?

  • Stopping an instance: Releases the instance's compute resources (CPU/GPU) and pauses compute charges. Note: An expanded system disk continues to incur charges.

  • Deleting an instance: Permanently deletes the instance and all its resources, including the system disk. All associated billing stops completely.

How to choose:

  • Stop: If you don't need the instance for a while but want to retain its data and environment to restart it later.

  • Delete: If you no longer need the instance and want to stop all billing. Be sure to back up your data first.

Q: Why is my DSW instance stuck in the Stopping or Deleting state for a long time?

The system needs to safely terminate tasks, save state, and reclaim resources. Common causes of extended wait times:

  • Some processes in the instance did not exit cleanly.

  • High memory usage prevents the instance from responding to the shutdown command.

If this occurs, wait a few minutes and then refresh the page. The instance should stop normally.

Q: Will I lose my data and code after I stop or delete a DSW instance?

Data retention depends on the operation you perform and your instance's system disk type.

  • Stopping an instance:

    When an instance is stopped, data retention depends on its system disk type.

    • Instances with a cloud disk system disk (most pay-as-you-go specifications and general-purpose resource instances that use a Disk as the system disk): If the cloud disk was not expanded and the instance is stopped for more than 15 days, the data is cleared and cannot be recovered. If the cloud disk was expanded or the instance has been stopped for 15 days or less, the data is not lost.

    • Instances that use Temporary Storage as the system disk: When you stop the instance, its data is deleted and cannot be recovered.

  • Deleting an instance:

    All data on its system disk is permanently erased and cannot be recovered. Therefore, make sure to back up all important data before you delete the instance.

Q: Why does my running DSW instance stop automatically?

The instance is configured with an idle auto-shutdown policy, enabled by default for free trial instances.

  • Trigger condition: The CPU and GPU utilization stays below a set threshold for 3 consecutive hours.

  • Recommended actions:

    • Manual stop: To ensure resource savings, manually stop the instance when it is not in use. The auto-shutdown policy is not guaranteed to trigger every time.

    • Modify the policy: If you need to run long-running tasks, modify or disable this policy.

      Modifying the DSW auto-shutdown policy

      1. Go to the workspace details page and click Configure Workspace > Auto-stop Settings.

        image

      2. In the DSW configuration section, you can modify the shutdown and exclusion policies. For example, if you do not want an instance to shut down automatically, you can add its name to the exclusion policy.

        image

Q: Why do I still see a "Running" status or receive bills after stopping or deleting all my DSW instances?

Check the following common causes:

  • You may be confusing a resource package with an instance. The "Running" status can refer to a resource package (such as "250 compute hours per month"), not an instance. A resource package remains active until it expires, and its status is independent of any instance.

  • Stopping an instance only pauses compute charges. An expanded system disk continues to incur storage charges.

  • Billing delay. Billing is not real-time, and bills may be generated several hours after you use a resource. For example, charges incurred in the morning might not appear on your bill until the afternoon.

DSW instance free trial

For more information about how to obtain, use, and release a DSW free trial, see Obtain, use, and release a free trial.

Appendix: Create an instance using the Python SDK

  1. The Alibaba Cloud SDK uses the Credentials tool to obtain credential information. Before you call the API, you must install and configure this tool. The requirements are as follows:

    • Python version >= 3.7.

    • Use the V2.0 generation of the Alibaba Cloud SDK.

    The installation command is as follows:

    pip install alibabacloud_credentials

  2. Obtain an AccessKey. This example uses AccessKey information to configure access credentials. To prevent this information from being exposed, we recommend that you configure the AccessKey as environment variables. The environment variable for the AccessKey ID is named ALIBABA_CLOUD_ACCESS_KEY_ID and the variable for the AccessKey Secret is named ALIBABA_CLOUD_ACCESS_KEY_SECRET.

  3. Install the PAI-related Python SDK.

    # Install the workspace SDK.
pip install alibabacloud-aiworkspace20210204 -U -q
# Install the DSW SDK.
pip install alibabacloud_pai_dsw20220101 -U -q
# OpenAPI dependency.
pip install alibabacloud_tea_openapi -U -q
# Install the SDK for querying subscription resource groups.
pip install https://sdk-portal-us-prod.oss-accelerate.aliyuncs.com/downloads/u-b8602de7-c468-436c-8a02-2eca4a30d376-python-paistudio.zip -U -q

  4. Create a DSW instance.

    Code example for creating an instance

    import os

from alibabacloud_aiworkspace20210204.client import Client as AIWorkspaceClient
from alibabacloud_aiworkspace20210204.models import (ListWorkspacesRequest,
                                                     ListImagesRequest)
from alibabacloud_credentials.client import Client
from alibabacloud_credentials.models import Config
from alibabacloud_pai_dsw20220101.client import Client as DSWClient
from alibabacloud_pai_dsw20220101.models import (GetInstanceRequest,
                                                 ListEcsSpecsRequest,
                                                 CreateInstanceRequest)
from alibabacloud_tea_openapi.client import TeaException
from alibabacloud_tea_openapi.models import Config as AliyunConfig

# Configure access credentials.
# An Alibaba Cloud account AccessKey has full permissions on all APIs. We recommend that you use a RAM user for API access or daily O&M.
# We strongly recommend that you do not save your AccessKey ID and AccessKey secret in your project code. This may lead to an AccessKey leak and threaten the security of all resources under your account.
# This example reads the AccessKey from environment variables by default through the Credentials SDK for identity verification.
region_id = 'cn-beijing'  # The region, such as cn-hangzhou, cn-shanghai, or cn-shenzhen.
config = Config(
    type='access_key',
    access_key_id=os.environ.get('ALIBABA_CLOUD_ACCESS_KEY_ID'),
    access_key_secret=os.environ.get('ALIBABA_CLOUD_ACCESS_KEY_SECRET'),
)
cred = Client(config)
# client config.
workspace_client = AIWorkspaceClient(
    config=AliyunConfig(
        credential=cred,
        region_id=region_id,
        endpoint="aiworkspace.{}.aliyuncs.com".format(region_id),
    )
)
dsw_client = DSWClient(
    config=AliyunConfig(
        credential=cred,
        region_id=region_id,
        endpoint='pai-dsw.{}.aliyuncs.com'.format(region_id),
    )
)


# Define a helper function to display DSW instance information.
def show_instance(instance_id):
    instance = dsw_client.get_instance(instance_id=instance_id, request=GetInstanceRequest()).body
    print(instance.status, instance.instance_name, instance.ecs_spec, instance.accumulated_running_time_in_ms)


# Query the properties and ID of an existing workspace.
workspace_name = 'Your existing AI workspace name'
# Get the list of workspaces.
workspaces = workspace_client.list_workspaces(ListWorkspacesRequest(
    page_number=1,
    page_size=10,
    workspace_name=workspace_name,  # Fuzzy match. If no name is specified, all workspaces are returned.
))
if len(workspaces.body.workspaces) == 0:
    raise RuntimeError('Specify the correct workspace_name')
for workspace in workspaces.body.workspaces:
    print(workspace.workspace_name,
          workspace.workspace_id,
          workspace.status, workspace.creator)
# Use the first result as the workspace for subsequent operations. You can switch as needed or directly specify a string ID.
workspace_id = workspaces.body.workspaces[0].workspace_id
# Get the list of images. You can use labels for filtering.
images = workspace_client.list_images(ListImagesRequest(
    page_size=100,
    # workspace_id=workspace_id, # Not specifying workspace_id means querying all built-in images on the PAI platform.
    labels=','.join(['system.supported.dsw=true',
                     # 'system.framework=tensorflow', # Specify pytorch or tensorflow.
                     'system.pythonVersion=3.6',
                     ]),
    verbose=True  # verbose=True lists more detailed information, including labels.
))
# You can view all available images.
for image in images.body.images:
    print(image.image_id, image.image_uri)
# Get the image used to submit the task. The first one is taken as an example here.
image_uri = images.body.images[0].image_uri
print('image_uri', image_uri)
# Get the list of DSW node specifications.
try:
    resp = dsw_client.list_ecs_specs(ListEcsSpecsRequest(accelerator_type='CPU',  # CPU or GPU
                                                         )).body
except TeaException as t:
    print("List ECS Specs failed:", t.message)
else:
    for spec in resp.ecs_specs:
        print(spec.instance_type + ", CPU: " + str(spec.cpu) + ", Memory: " + str(spec.memory))
    # Get the node specification for submitting the task.
    ecs_spec = resp.ecs_specs[0].instance_type
    print('Selected ecs_spec:', ecs_spec)

# Create a DSW instance.
request = CreateInstanceRequest(instance_name="Test_From_SDK_1",
                                ecs_spec=ecs_spec,
                                workspace_id=workspace_id,
                                # image_id='', # You can specify the ID of an image in the workspace, but only one of image_id and image_url can be specified.
                                image_url=image_uri)
try:
    ins_resp = dsw_client.create_instance(request)
except TeaException as t:
    print('Failed to create the instance. Error message: ' + t.message)
else:
    instance_id = ins_resp.body.instance_id
    print("Created Instance ID:", instance_id)
    show_instance(instance_id)

For more information, see API overview.

Previous：Create and manage DSW instancesNext：Access and manage DSW instances