Engine profiles
An engine profile is an administrator-defined template that sets resource constraints, timeout limits, and configuration defaults for AI engines in H2O AI Cloud. Every engine (Driverless AI, H2O-3, or Notebook) requires a profile before you can create it.
How profiles relate to engines
Profiles and engines serve different roles:
| Concept | Created by | Purpose |
|---|---|---|
| Engine profile | Platform administrator | Defines the allowed ranges for CPU, GPU, memory, storage, and timeouts |
| Engine | End user | A running instance of Driverless AI, H2O-3, or a Notebook with specific resource values |
When you create an engine, you select a profile first. The profile determines the minimum, maximum, and default values for each resource. You can customize your engine's resources within those bounds.
Profile (admin-defined) Engine (user-created)
┌────────────────────────────┐ ┌─────────────────────┐
│ CPU: 1–16 (default 4) │ ──► │ CPU: 8 │
│ GPU: 0–4 (default 0) │ │ GPU: 1 │
│ Mem: 4–64 GiB (default 8) │ │ Mem: 32 GiB │
│ Max idle: 1–8 hours │ │ Max idle: 2 hours │
└────────────────────────────┘ └─────────────────────┘
Profile types
Each engine type has its own profile type with fields tailored to that engine:
- Driverless AI Engine profiles: Include configuration editability controls, TOML-based configuration layering, Triton Inference Server support, and storage constraints.
- H2O Engine profiles: Include node count constraints for distributed processing across multiple H2O-3 nodes.
- Notebook Engine profiles: Include storage constraints and Git repository sync settings.
All three profile types share the following fields:
| Field | Required | Type | Description |
|---|---|---|---|
display_name | Optional | String | Human-readable name for the profile |
priority | Optional | Integer | Display order priority. A lower value means higher priority |
enabled | Optional | Boolean | Whether the profile is active and available to users |
assigned_oidc_roles_enabled | Optional | Boolean | Whether to restrict access by OpenID Connect (OIDC) role. OIDC is an identity protocol your organization uses to manage user roles |
assigned_oidc_roles | Optional | List of strings | OIDC roles that grant access to this profile |
max_running_engines | Optional | Integer | Maximum number of engines of this type a single user can run simultaneously, counting all active engines in the workspace regardless of profile. When unset, there is no limit. This constraint is currently enforced for Driverless AI and Notebook engines only |
cpu_constraint | Required | Numeric constraint | Allowed range for CPU units |
gpu_constraint | Required | Numeric constraint | Allowed range for GPU units |
memory_bytes_constraint | Required | Numeric constraint | Allowed range for memory |
max_idle_duration_constraint | Required | Duration constraint | Allowed range for max idle time |
max_running_duration_constraint | Required | Duration constraint | Allowed range for max running time |
Driverless AI and Notebook profiles also include a storage_bytes_constraint field for persistent storage. See the Driverless AI and Notebook profile pages for details.
Resource constraints
Every profile defines constraints using a minimum/maximum/default model:
| Field | Description |
|---|---|
minimum | The lowest value a user can set |
maximum | The highest value a user can set (unset means unlimited) |
default | The value used when the user does not specify one |
cumulative_maximum | The maximum total value a single user can consume across all engines of this type in the workspace. For storage constraints, the cumulative total includes all engines regardless of state because storage persists throughout the engine lifecycle. Unset means unlimited |
Numeric constraints (CPU, GPU, memory, storage) support all four fields. Duration constraints (max idle time, max running time) support minimum, maximum, and default only. They do not have a cumulative limit.
When you create an engine, the resource sliders and input fields reflect the minimum and maximum values from your selected profile. If you cannot set a value higher than expected, the profile's constraints are the reason. Each resource field has an info icon (ⓘ) that displays the exact bounds from the selected profile.
Access control with OIDC roles
Administrators can restrict which users see a profile by assigning OpenID Connect (OIDC) roles:
- When
assigned_oidc_roles_enabledis turned on, only users with at least one matching OIDC role can see and use the profile. - When turned off, the profile is visible to all users in the workspace.
This lets administrators create different profiles for different teams. For example, a high-resource profile for the ML engineering team and a lighter profile for analysts.
If assigned_oidc_roles_enabled is turned on but assigned_oidc_roles is empty, no user can see or use the profile. Always assign at least one OIDC role when turning on role-based access.
Permissions
Creating, updating, and deleting profiles requires administrator permissions on the workspace. All workspace users can list their assigned profiles.
If a profile does not appear in the Profile dropdown during engine creation, verify that the profile is enabled, the user has a matching OIDC role (when role-based access is turned on), and the profile belongs to the correct workspace.
Manage profiles in the UI
Administrators can manage engine profiles directly in the H2O AI Cloud interface:
- In the left sidebar, navigate to Admin > AI Engines.
- Under the engine type (DAI Engine, H2O Engine, or Notebook Engine), click Profiles.
- The profile list displays each profile's display name, ID, priority, and enabled status.
From the profile list, you can:
- Click Add [Engine] Engine Profile to create a new profile.
- Click View to open a profile's configuration details.
- Use the dropdown menu (▾) next to View to edit or delete a profile.
Profile lifecycle
- Profiles are workspace-scoped: Each profile belongs to a workspace. In most deployments, profiles are created in
workspaces/global, which makes them available across the platform. Administrators can copy profiles across workspaces. - Profiles are snapshots at engine startup: When an engine starts or resumes from a paused state, the platform captures a fresh snapshot of the profile's settings. Changes to the profile do not affect engines that are actively running, but paused engines pick up the updated profile on their next resume.
- Profiles can be disabled: Administrators can disable a profile without deleting it. Disabled profiles do not appear in the Profile dropdown.
- Deleting a profile does not affect running engines: Running engines retain their snapshot. However, paused engines cannot be resumed with a deleted profile — an administrator must assign a new valid profile before the engine can resume. Users cannot create new engines with a deleted profile.
For Python client examples, see the engine-specific profile pages linked in Profile types.
- Submit and view feedback for this page
- Send feedback about AI Engine Manager to cloud-feedback@h2o.ai