Skip to main content

Migration from Steam

AI Engine Manager (AIEM) manages AI Engines (Driverless AI, H2O-3, and Notebook) in H2O AI Cloud and introduces a new API with a corresponding Python client.

Breaking Changes

AIEM replaces Steam with breaking API changes that affect all Python scripts, Wave apps, Jupyter Notebooks, and other Python projects using the Steam Python API. You must update your code to use the AIEM client.

Migrate your Steam Python API code to the AIEM client to continue using AI Engines in H2O AI Cloud. Contact us with questions and feature requests during your migration.

Documentation​

Refer to the following guides to complete your migration:

Review these examples to understand the AIEM API before migrating your code.

Workspaces​

Workspaces are the new way of organizing resources in H2O AI Cloud.

In AIEM, almost everything is a resource (AI Engines, models, app instances, etc.), and all resources belong to a workspace. A user can be a member of multiple workspaces, enabling collaboration and resource sharing across teams.

Key concepts:

  • Use workspace_id="default" to access your personal workspace
  • Resources are scoped to workspaces for access control
  • Workspace membership determines which engines and resources you can access

DAI Client Management​

Steam managed the H2O Driverless AI (DAI) Python client centrally, downloading it from the Steam server before connecting to DAI engines.

Key Difference

AIEM does not manage the DAI client centrally. You must have the driverlessai Python package installed in your environment to connect to DAI engines.

The AIEM Python client includes the DAI client as a dependency (driverlessai>=1.10.0). When you run pip install h2o-engine-manager, pip automatically installs the latest compatible version of the DAI client.

You do not need to take additional action if you include h2o-engine-manager in your requirements.txt file.

H2O Client Management​

The h2o Python client requires version matching: the client version must exactly match the H2O-3 engine version you use in AIEM.

Version Matching Required

You must ensure your h2o Python client version matches your H2O-3 engine version. The client cannot connect to engines with different versions.

This version matching requirement exists in both Steam and AIEM. You must install the specific h2o client version that matches your engine version.

FAQ​

The engine has been starting for an extended period. What is the issue?​

Timeout Behavior Change

AIEM removes the timeout constraint when creating engines. Engines continue attempting to start instead of failing after a timeout period.

When the engine is in STATE_STARTING: AIEM waits for the Kubernetes infrastructure to provision and allocate resources (CPU, memory, GPU). This typically takes 2-5 minutes depending on cluster load. Wait for the infrastructure to allocate resources—you cannot accelerate this process.

When the engine is in STATE_CONNECTING for an extended period (>10 minutes): The engine is unhealthy due to a startup issue, hang, or connection problem. The engine may recover on its own, but you should restart it by pausing and resuming, or by deleting and recreating the engine.

Feedback