Software Dependencies

Ascend Docker Runtime

Docker in the current environment must be 18.09 or later.
The driver and firmware have been installed on the host. For details, see "Installing the NPU Driver and Firmware" in CANN Software Installation Guide (Commercial Edition) or "Installing the NPU Driver and Firmware" in CANN Software Installation Guide (Community Edition).
To install Ascend Docker Runtime on an Atlas 500 A2 edge station, you need to modify the Docker configuration. Run the vi /etc/sysconfig/docker command to delete --config-file="" and run the systemctl restart docker command to make the configuration take effect.
The MEF service preset on the Atlas 500 A2 edge station performs security hardening on Docker. Ascend Docker Runtime cannot be used in the Docker environment after security hardening. If you need to use Ascend Docker Runtime, manually uninstall the MEF service. For details, see "Uninstalling MEF Edge" in MindEdge Framework User Guide.

If the command output contains /docker_entrypoint.sh after the systemctl status docker command is executed, Docker is securely hardened by the MEF service.

Other Cluster Scheduling Components

The dependencies of ARM are different from those of x86_64. Select dependencies based on the system architecture. The cluster scheduling components support both IPv4 and IPv6. By default, IPv4 is used.

**Table 1** Software environment
Software Name	Supported Version	Installation Position	Description
Kubernetes	1.17.x to 1.34.x (Version 1.19.x or later is recommended.) NOTE: Select the latest bugfix version. If Volcano is required, install Kubernetes 1.19.x or later. For details about the Kubernetes version, refer to Kubernetes compatibility on the Volcano official website.	All nodes	For details about how to use Kubernetes, see Kubernetes Documentation.
(Optional) Docker	18.09.x to 28.5.1	All nodes	Obtain it from the Docker community or official website. The versions of Docker and Kubernetes must match. For details about the version mapping, see CHANGELOG of Kubernetes or obtain it from the Kubernetes community. Select the latest bugfix version.
(Optional) containerd	1.4.x to 2.1.4 (Version 1.6.x is recommended.)	All nodes	You can obtain it from official website or community of containerd. The latest bugfix version is recommended. Pay attention to CRI version support used with Kubernetes.
Driver and firmware of the Ascend AI processor	Select the driver and firmware packages that match MindCluster according to your specific hardware model.	Compute nodes	For details, see Driver and Firmware Installation and Upgrade Guides of hardware products to obtain the guide of the corresponding version. NOTE: To ensure that NPU Exporter can be installed as a non-root user (for example, hwMindX) when its binary package is deployed, use the --install-for-all parameter during driver installation. Example: ./Ascend-hdk-<chip_type>-npu-driver_<version>_linux-<arch>.run --full --install-for-all
(Optional) CANN	If only the cluster scheduling components are installed, you do not need to install CANN. Alternatively, you can install a corresponding CANN package by referring to Version Mapping based on your installation requirements.	Compute nodes or training/inference containers	Install the CANN package on the host. For details, see CANN Software Installation Guide (commercial edition) or CANN Software Installation Guide (community edition).
Python	3.8 to 3.12	Training or inference container	The Python version is subject to the AI framework.

Install Docker or containerd based on the actual service scenario.
For details about how to install an OS on an Atlas server, see Installation Guides (ARM) and Installation Guides (x86_64). The installation guides do not cover all the preceding OSs, and they are for reference only.
For the Atlas A2 training product, OS requirements vary depending on the VM scenario.

Parent topic: Environment Dependencies