Updated on 2026-06-27 GMT+08:00

IAM

This section describes the IAM permission configurations for all ModelArts functions.

IAM Permissions

If you need to assign different permissions to employees in your enterprise to access your purchased ModelArts resources, Identity and Access Management (IAM) is a good choice for fine-grained permissions management. IAM provides identity authentication, fine-grained permissions management, and access control. IAM helps you secure access to your cloud resources. If your Huawei account can meet your requirements and you do not need an IAM account to manage user permissions, skip this chapter.

IAM is a free service. You only pay for the resources in your account.

With IAM, you can assign permissions to control users' access to specific resources. For example, if the software developers in your enterprise need to own permissions to use ModelArts, yet you do not want them to own high-risk operation permissions such as deleting ModelArts, you can grant permissions using IAM to limit their permission on ModelArts.

For details about IAM, see What is IAM?

Role/Policy-based Authorization

ModelArts supports role/policy-based authorization. By default, new IAM users do not have any permissions. You need to add a user to one or more groups, and assign permissions policies or roles to these groups. Users inherit permissions of the groups to which they are added. This process is called authorization. The users then inherit permissions from the groups and can perform specified operations on cloud services.

ModelArts is a project-level service deployed for specific regions. When you set Scope to Region-specific projects and select the specified projects (for example, ap-southeast-2) in the specified regions (for example, AP-Bangkok), the users only have permissions for ModelArts resources in the selected projects. If you set Scope to All resources, the users have permissions for ModelArts resources in all region-specific projects. When accessing ModelArts, the users need to switch to a region where they have been authorized to use cloud services.

Table 1 lists all system-defined policies supported by ModelArts. If preset ModelArts permissions cannot meet your requirements, create a custom policy by referring to Policy Fields in JSON Format.

Table 1 System-defined policies supported by ModelArts

Policy

Description

Type

ModelArts FullAccess

All permissions for ModelArts administrators

System-defined policy

ModelArts CommonOperations

All operation permissions for ModelArts common users, which do not include managing dedicated resource pools.

System-defined policy

ModelArts Dependency Access

Permissions on dependency services for ModelArts

System-defined policy

ModelArts depends on other cloud services. To check or view the cloud services, configure the corresponding permissions on the ModelArts console. The dependencies are as follows.

Table 2 ModelArts dependencies

Console Function

Dependency

Data preparation

MapReduce Service (MRS)

Object Storage Service (OBS)

Software Repository for Container (SWR)

Data management (dataset/data labeling/data processing)

Object Storage Service (OBS)

Data Lake Insight (DLI)

MapReduce Service (MRS)

GaussDB(DWS)

ModelArts

Development environment notebook/Image management/Lite Server

Object Storage Service (OBS)

Cloud Trace Service (CTS)

Elastic Cloud Server (ECS)

SoftWare Repository for Container (SWR)

Scalable File Service (SFS)

Application Operations Management (AOM)

Data Encryption Workshop (DEW)

Bare Metal Server (BMS)

Image Management Service (IMS)

ModelArts

Virtual Private Cloud (VPC)

Elastic Volume Service (EVS)

CloudMatrix

Algorithm management/Training management/Workflow

Object Storage Service (OBS)

Simple Message Notification (SMN)

Cloud Trace Service (CTS)

Enterprise Project Management Service (EPS)

SFS Turbo

Software Repository for Container (SWR)

Application Operations Management (AOM)

Key Management Service (KMS)

VPC

ModelArts

Model management/Real-time service (old-version)/Batch service/Edge service/Dedicated resource pool for edge deployment

Object Storage Service (OBS)

Cloud Eye

Simple Message Notification (SMN)

Enterprise Project Management Service (EPS)

Cloud Trace Service (CTS)

Log Tank Service (LTS)

Virtual Private Cloud (VPC)

Software Repository for Container (SWR)

ModelArts

Real-time service (new version)

Virtual Private Cloud (VPC)

VPC Endpoint (VPCEP)

Object Storage Service (OBS)

Software Repository for Container (SWR)

Scalable File Service Turbo (SFS Turbo)

Key Management System (KMS)

Elastic Load Balance (ELB)

Cloud Eye

Simple Message Notification (SMN)

Enterprise Project Management Service (EPS)

Cloud Trace Service (CTS)

Log Tank Service (LTS)

Application Operations Management (AOM)

ModelArts

AI Gallery

Object Storage Service (OBS)

Cloud Trace Service (CTS)

Software Repository for Container (SWR)

ModelArts

Dedicated resource pool and Lite Cluster

Cloud Trace Service (CTS)

Cloud Container Engine (CCE)

Bare Metal Server (BMS)

Image Management Service (IMS)

Data Encryption Workshop (DEW)

Virtual Private Cloud (VPC)

Elastic Cloud Server (ECS)

Scalable File Service (SFS)

Object Storage Service (OBS)

Application Operations Management (AOM)

Tag Management Service (TMS)

ModelArts

Billing Center

Elastic Volume Service (EVS)

Software Repository for Container (SWR)

If system-defined policies cannot meet your requirements, you can create a custom policy. For details about the actions supported by custom policies, see ModelArts Resource Permissions.

To create a custom policy, choose either visual editor or JSON.

  • Visual editor: Select cloud services, actions, resources, and request conditions without the need to know policy syntax.
  • JSON: Create a JSON policy or edit an existing one.

For details, see Creating a Custom Policy. This section provides examples of common custom ModelArts policies.

  • Example 1: Grant permission to manage images.
    {
        "Version": "1.1",
        "Statement": [
            {
                "Effect": "Allow",
                "Action": [
                    "modelarts:image:register",
                    "modelarts:image:listGroup"
                ]
            }
        ]
    }
  • Example 2: Grant permission to deny creating, updating, and deleting a dedicated resource pool.

    A policy with only "Deny" permissions must be used together with other policies. If the permissions granted to an IAM user contain both "Allow" and "Deny", the "Deny" permissions take precedence over the "Allow" permissions.

    {
        "Version": "1.1",
        "Statement": [
            {
                "Action": [
                    "modelarts:*:*"
                ],
                "Effect": "Allow"
            },
            {
                "Action": [
                    "swr:*:*"
                ],
                "Effect": "Allow"
            },
            {
                "Action": [
                    "smn:*:*"
                ],
                "Effect": "Allow"
            },
            {
                "Action": [
                    "modelarts:pool:create",
                    "modelarts:pool:update",
                    "modelarts:pool:delete"
                ],
                "Effect": "Deny"
            }
        ]
    }
  • Example 3: Create a custom policy containing multiple actions.

    A custom policy can contain actions of multiple services that are of the global or project-level type. The following is an example policy containing actions of multiple services:

    {
        "Version": "1.1",
        "Statement": [
            {
                "Effect": "Allow",
                "Action": [
                    "modelarts:service:*"
                ]
            },
            {
                "Effect": "Allow",
                "Action": [
                    "lts:logs:list"
                ]
            }
        ]
    }

Policy Fields in JSON Format

Policy Structure

A policy consists of a version and one or more statements (indicating different actions).

Figure 1 Policy structure

Policy Parameters

The following describes policy parameters. You can create custom policies by specifying the parameters. For details, see Custom Policy Use Cases.

Table 3 Policy parameters

Parameter

Description

Value

Version

Policy version

1.1: indicates policy-based access control.

Statement: authorization statement of a policy

Effect

Whether to allow or deny the operations defined in the action

  • Allow: indicates the operation is allowed.
  • Deny: indicates the operation is not allowed.
    NOTE:

    If the policy used to grant user permissions contains both Allow and Deny for the same action, Deny takes precedence.

Action

Operation to be performed on the service

Format: "Service name:Resource type:Action". Wildcard characters (*) are supported, indicating all options.

Example:

modelarts:notebook:list indicates the permission to view a notebook instance list. modelarts indicates the service name, notebook indicates the resource type, and list indicates the operation.

View all actions of a service in its API Reference.

Condition

Condition for a policy to take effect, including condition keys and operators

Format: "Condition operator:{Condition key:[Value 1,Value 2]}"

If you set multiple conditions, the policy takes effect only when all the conditions are met.

Example:

StringEndWithIfExists":{"g:UserName":["specialCharacter"]}: The statement is valid for users whose names end with specialCharacter.

Resource

Resources on which a policy takes effect

Format: Service name:<Region>:<Account ID>:Resource type:Resource path. Asterisks (*) are supported for resource type, indicating all resources.

NOTE:

ModelArts authorization does not allow you to specify a resource path.

ModelArts Resource Types

Administrators can specify the scope based on ModelArts resource types. The following table lists the resource types supported by ModelArts:

Table 4 Resource types supported by ModelArts role/policy-based authorization

Resource Type

Description

notebook

Notebook instances in DevEnviron

exemlProject

ExeML projects

exemlProjectInf

ExeML-powered real-time inference service

exemlProjectTrain

ExeML-powered training jobs

exemlProjectVersion

ExeML project version

workflow

Workflow

pool

Dedicated resource pool

network

Networking of a dedicated resource pool

trainJob

Training job

trainJobLog

Runtime logs of a training job

trainJobInnerModel

Preset model

model

Models

service

Real-time service

nodeservice

Edge service

workspace

Workspace

dataset

Dataset

dataAnnotation

Dataset labels

aiAlgorithm

Algorithm for training jobs

image

Image

devserver

Elastic BMS

ModelArts Resource Permissions

For details, see "Permissions Policies and Supported Actions" in ModelArts API Reference.