Managing Remote Users

When using OIDC or SAML as your authentication protocol you can configure HPE Machine Learning Development Environment to remote provision and manage your users.

Remote user provisioning lets you include and synchronize any information about the user stored in your IdP such as their username, display name, and groups. Once configured, you can manage remote users without having to manually modify the users or update them via SCIM. Each time the remote user accesses HPE Machine Learning Development Environment, their information is synchronized.

Attention

This feature applies only to Determined Enterprise Edition.

Enable Remote User Management

Set the Auto Provision Option

The first step is to configure the master configuration file to enable user auto-provisioning and the remote management of any information attached to the users.

OIDCSAML

• Enable OIDC.

• Set oidc.auto_provision_users option to true in your master configuration file.

Note

If scim_enabled is true, then oidc.auto_provision_users must be false.

For example:

oidc:
    enabled: true
    provider: "Okta"
    idp_recipient_url: "https://determined.example.com/oidc/sso"
    idp_sso_url: "https://dev-00000000.okta.com"
    client_id: "xx0xx0"
    client_secret: "xx0xx0"
    auto_provision_users: true
    always_redirect: true
    posix_user_claim: "preferred_username"
    display_name_attribute_name: "XYZ"
    agent_uid_attribute_name: "user_id_key"
    agent_gid_attribute_name: "group_id_key"
    agent_user_name_attribute_name: "agent_user_key"
    agent_group_name_attribute_name: "agent_group_key"

HPE Machine Learning Development Environment sets the username of the user to the IdP email address. You cannot set the username independently.

To automatically link custom claims, you can configure the attribute name with the relevant field. For example:

agent_uid_attribute_name: "custom_uid_attribute_name"

Note

Once this feature is enabled and an agent UID/GID is assigned to a user, it can only be reset manually. To revert the UID or GID back to the default value specified by security.default_task.uid(gid), an admin must manually remove the agent UID/GID settings for that user.

Set the Groups Claim Name Option

HPE Machine Learning Development Environment receives OIDC and SAML claims via a JSON Web Token (JWT) that you send from your IdP. If there is a group that does not already exist in HPE Machine Learning Development Environment, then the system creates the group.

To enable group membership synchronization:

OIDCSAML

• Set the groups_attribute_name option to match the claim name for group memberships from your authenticator (i.e., either a custom user attribute such as groups_memberships or usergroup_memberships, or the authenticator’s groups claim, passed in as groups, etc.).

For example, in the following claim, let’s say the user-group information is passed through group_memberships in your IdP.

{
  email: dee.ray@example.com
  group_memberships: ['A', 'B'],
  displayName: "Dee Ray"
 }

Then, HPE Machine Learning Development Environment creates the following user:

{
 username: dee.ray@example.com
 groups: ['A', 'B'],
 displayName: "Dee Ray"
}

Each time the authenticated user accesses HPE Machine Learning Development Environment, their information is passed to HPE Machine Learning Development Environment, and group memberships are updated. For example, when a user is assigned to a new group via your IdP, that information is updated in HPE Machine Learning Development Environment.

Complete the Auto Provision Process

Once auto provisioning is configured, the user simply signs in with their username.

For example, to sign in to HPE Machine Learning Development Environment via Okta, the user performs the following steps:

• Visit the HPE Machine Learning Development Environment URL, e.g., https://determined.example.com.

• Under Sign in with Okta, enter the SSO-enabled email address.

If the sign in is successful, HPE Machine Learning Development Environment provisions the user, adds the user to the user table, and authenticates the user to HPE Machine Learning Development Environment.

Manage Remote Users and Groups via the WebUI

To manage users and groups in the WebUI:

• Sign in to the HPE Machine Learning Development Environment cluster via the HPE Machine Learning Development Environment URL, e.g., https://determined.example.com, as a cluster admin.

• View Admin Settings by selecting your profile in the upper left corner and then choosing Admin.

Actions you can take include adding new users, switching existing users to remote users, performing bulk actions, and more. For example, using the filters, you can select to view only active users. You can also manage user groups.

To find out how to manage remote users via the WebUI, including adding a new remote user, visit managing RBAC via the WebUI.

Handling SSO Session Expiry

SSO users are redirected to the SSO provider’s authentication URIs when their session token has expired, rather than displaying the HPE Machine Learning Development Environment sign-in page. This ensures a continuous user experience without manual intervention when sessions expire.

Note

The always_redirect option enables this behavior. If a user does not have a session token, this option generally bypasses the HPE Machine Learning Development Environment sign-in page, except when the user signs out through the WebUI. If multiple SSO providers have always_redirect enabled, the system will attempt to use each provider. This differs from the session expiry handling described above.