Appendix

Reference Information

Download Reference

Supported OSs and Software Packages Available for Download

When using MindCluster Ascend Deployer for download, see Download Precautions. For the value range of OS, see Supported OSs and Architectures. For the value range of software packages, see Table 2.

Download Notes

• Because a large number of software packages need to be downloaded, MindCluster Ascend Deployer enables the parallel download function by default. You can modify the parallel_download field in the ascend_deployer/downloader/config.ini file to determine whether to enable the parallel download function. 1 indicates enabled, and 0 indicates disabled. The default number of parallel download threads is 16. You can modify the parallel number through the ASCEND_DEPLOYER_DOWNLOAD_MAX_SIZE environment variable.
• MindCluster Ascend Deployer has provided a repository configuration file, which uses the Huawei repository by default. You can replace it as needed. For details, see Configuring the Repository.
• When downloading OS dependencies, to compile and install the hccl-test tool, the software packages of MPI 3.2.1 and Open MPI 4.1.5 are automatically downloaded. During installation, the corresponding dependency is used for compilation based on the configured IP type of the environment to be installed. For IPv4 format, MPI 3.2.1 is installed; for IPv6 format or Atlas A3 Training Series Products, Open MPI 4.1.5 is installed.

Supported OSs and Architectures

Table 1 Supported OSs and architectures for MindCluster Ascend Deployer

Supported Downloadable Software Packages

Table 2 Software packages supported for download by MindCluster Ascend Deployer

Installation and Upgrade Reference

install Parameter Description

Select the corresponding parameters based on actual needs to complete the installation.

Command format:

• Method 1: Download the ZIP package, extract it, and use bash install.sh [options].
• Method 2: Install MindCluster Ascend Deployer via the pip command and use ascend-deployer [options].
• Method 3: Download the ZIP package, extract it, and use bash large_scale_install.sh [options].
• Method 4: Install MindCluster Ascend Deployer using the pip command and use large-scale-deployer [options].

For option, see Table 1. The range of values for each option in the table can be viewed by running bash install.sh --help.

Table 1 Option description

Example of Check Result File

When the --check_mode option is used, a check_res_output.json check report will be generated in the ~/.ascend_deployer/deploy_info/ directory, accurately indicating error information. An example of the result is shown below:

{
  "CheckList": [
    {
      "check_item": "check_card",
      "desc_en": "Check NPU compatibility",
      "desc_zh": "Check NPU compatibility",
      "tip_en": "",
      "tip_zh": ""
    },
   {
        "check_item": "check_k8s_version",
        "desc_en": "Judgment: 1. kubelet, kubectl, and kubeadm all exist"
                   "2. kubelet --version == kubeadm version == kubectl version "
                   "3. kubelet version < 1.29 "
                   "4. kubelet version >=1.19.16.",
        "desc_zh": "Judgment: 1. kubelet, kubectl, and kubeadm all exist"
                   "2. kubelet --version == kubeadm version == kubectl version "
                   "3. kubelet version < 1.29 "
                   "4. kubelet version >=1.19.16。",
        "tip_en": "Execute the version query command to confirm whether the component has been installed, "
                  "whether the version number is the same, and whether the version is within the supported range.",
        "tip_zh": "Run the version query command to confirm whether the components are installed, whether the version numbers are the same, and whether the versions are within the supported range.",
        "help_url": ""
    }
  ],
  "HostCheckResList": {
    "xx.xx.xx.x1": [
      {
        "check_item": "check_card",
        "status": "failed",
        "error_msg": "Check card failed: [ASCEND] A300i-pro has no support for MTOS_22.03LTS-SP4_aarch64 on this device"
      }
    ]
  }
}

Table 2 Parameter description of the check result file

Supported Installation and Upgrade Scenarios

MindCluster Ascend Deployer supports the following basic installation and upgrade scenarios.

Installation Notes

• During installation, the runtime environment time must be calibrated to the correct UTC time using the date -s command.
• MindCluster Ascend Deployer can only install the most basic libraries to ensure that TensorFlow and PyTorch can run. If you need to run more complex inference services or model training, the model code may contain libraries related to specific services, which must be installed by the user.
• MindCluster Ascend Deployer requires many dependencies. It is recommended to perform installation or upgrade operations on user‑uploaded software packages only after the OS dependencies downloaded by the tool have been successfully installed or upgraded.
• Do not execute the --install-scene command repeatedly, as this will cause the installation to fail.

Optional Installation and Upgrade Scenarios

Table 1 Optional installation and upgrade scenarios

Supported Installation and Upgrade Packages

MindCluster Ascend Deployer supports specifying software packages listed in Table 1 for installation. You can run bash install.sh --help to view the supported parameters for --install=<package_name>.

Installation Notes

• During installation, the runtime environment time must be calibrated to the correct UTC time using the date -s command.

• When specifying software installation or upgrade, MindCluster Ascend Deployer only supports specifying the installation or upgrade of software versions released within the past year.

• MindCluster Ascend Deployer can only install the most basic libraries to ensure that TensorFlow and PyTorch can run. If you need to run more complex inference services or model training, the model code may contain libraries related to specific services, which must be installed by the user.

• If the GCC version is earlier than 7.3.0, MindCluster Ascend Deployer will automatically install it.

• MindCluster Ascend Deployer has many required dependencies. It is recommended that you perform operations such as installing or upgrading user-uploaded software packages only after the OS dependencies downloaded by the tool are installed.

Optional Installation and Upgrade Software Packages

Table 1 Optional installation and upgrade software packages

Typical Configurations for Prefill-Decode Disaggregation

Currently, only classic configuration parameters are supported. Other parameters are automatically populated based on the specified max_seq_len maximum sequence length parameter (corresponding to the maxSeqLen parameter in the following table) and the number of D instances.

Typical Configurations for Atlas 800I A2 Inference Server

Table 1 Key parameter description for Atlas 800I A2 inference server

Table 2 Typical configurations for AI computing nodes of the Atlas 800I A2 inference server

Typical Configurations for Atlas 800I A3 SuperPoD

Table 3 Key parameter description for Atlas 800I A3 SuperPoD server devices

Table 4 Typical configurations for AI computing nodes of Atlas 800I A3 SuperPoD

(Optional) Certificate Configuration

The Parameter Description file for certificates is stored in $HOME/ascend-deployer/ascend_deployer/group_vars/master/tls_config.yaml. The following is an example of the file content.

# group_vars/tls_config.yaml
tls_config:
  tls_enable: false
  kmc_ksf_master: "./security/master/tools/pmt/master/ksfa"
  kmc_ksf_standby: "./security/standby/tools/pmt/standby/ksfb"
  infer_tls_items:
    ca_cert: "./security/infer/security/certs/ca.pem"
    tls_cert: "./security/infer/security/certs/cert.pem"
    tls_key: "./security/infer/security/keys/cert.key.pem"
    tls_passwd: "./security/infer/security/pass/key_pwd.txt"
    tls_crl: "infer"
  management_tls_items:
    ca_cert: "./security/management/security/certs/ca.pem"
    tls_cert: "./security/management/security/certs/cert.pem"
    tls_key: "./security/management/security/keys/cert.key.pem"
    tls_passwd: "./security/management/security/pass/key_pwd.txt"
    tls_crl: "management"

  # The runtime scenario of Atlas 800I A2 Inference Server does not require configuring ccae_tls_enable and ccae_tls_items.
  ccae_tls_enable: false
  ccae_tls_items:
    ca_cert: "./security/ccae/security/certs/ca.pem"
    tls_cert: "./security/ccae/security/certs/cert.pem"
    tls_key: "./security/ccae/security/keys/cert.key.pem"
    tls_passwd: "./security/ccae/security/pass/key_pwd.txt"
    tls_crl: "ccae"
  cluster_tls_enable: false
  cluster_tls_items:
    ca_cert: "./security/clusterd/security/certs/ca.pem"
    tls_cert: "./security/clusterd/security/certs/cert.pem"
    tls_key: "./security/clusterd/security/keys/cert.key.pem"
    tls_passwd: "./security/clusterd/security/pass/key_pwd.txt"
    tls_crl: "clusterd"
  etcd_server_tls_enable: false
  etcd_server_tls_items:
    ca_cert: "./security/etcd_server/security/certs/ca.pem"
    tls_cert: "./security/etcd_server/security/certs/cert.pem"
    tls_key: "./security/etcd_server/security/keys/cert.key.pem"
    tls_passwd: "./security/etcd_server/security/pass/key_pwd.txt"
    kmc_ksf_master: "./security/etcd_server/tools/pmt/master/ksfa"
    kmc_ksf_standby: "./security/etcd_server/tools/pmt/standby/ksfb"
    tls_crl: ""

Other

User Information List

Update user passwords regularly to avoid the risks associated with using the same password for an extended period.

Users in the openEuler Base Image in the Dockerfile Example

Password Complexity Requirements

Passwords must meet at least the following requirements:

1. The password must be at least 8 characters long.

2. The password must contain a combination of at least two of the following character types:

   • One lowercase letter
   • One uppercase letter
   • One digit
   • A special character: `~!@#$%^&*()-_=+\|[{}];:'",<.>/? and space

3. The password cannot be the same as the account name.

CA Certificate

Table 1 CA certificate