On-Prem Development, Training & Assets

This scenario guides users through creating an On-Prem development or training environment and preserving code, data, images, models, and outputs as reusable assets.

Applicable Roles

• End User creating development or training workloads
• Platform Operator preparing images, storage, compute, quota, and monitoring

Target Outcome

• Users can create development or training environments with an available image, flavor, and storage.
• Code and data reside on persistent storage instead of an ephemeral container filesystem.
• Customized environments can be saved as images and model outputs can be stored in controlled locations.
• Logs, monitoring, and usage remain visible, and idle resources can be released.

Before You Start

1. The operator has enabled the target region, images, storage, flavors, and tenant quota.
2. Estimate CPU, memory, accelerator, and runtime requirements.
3. Prepare the code repository, data paths, base image, and output directory.
4. Confirm data governance, image provenance, and external network boundaries.

Procedure

1. Prepare object or file storage and confirm that data and output paths are accessible. See Object Storage and File Storage.
2. Prepare or select a versioned runtime image that the target region can pull. See Image Service.
3. Open Development Environments, create an online IDE, select the image and resource specification, mount the persistent workspace, and confirm that the IDE opens normally.

4. Open Model Training, create a training or batch runtime instance, review code, data, output directory, and startup command, and confirm that the workload runs and emits logs.

5. Preserve images, models, and outputs in controlled storage so later workloads can reuse them.
6. Review Job Monitoring and Resource Usage, then stop idle instances.

Asset Guidance

Asset	Recommended Location	Do Not Keep Only In
Code and configuration	Version control and persistent workspace	Ephemeral container directory
Datasets	Object or file storage	Local temporary disk
Runtime environment	Versioned image	Manual installation notes
Model weights	Controlled object storage or model directory	One instance filesystem
Logs and outputs	Persistent output directory	Terminal scrollback

Completion Checklist

Purpose: These are the exit criteria for the current feature task. Use them to decide whether the result is observable and reviewable and whether you can continue to the next step in the scenario. They do not repeat the procedure; if any item fails, follow the troubleshooting section below.

Check	Pass Criteria
1	The development or training instance is healthy without repeated errors.
2	Persistent code and data remain available after restart or rebuild.
3	Images, models, and outputs have clear versions and owners.
4	Flavor, runtime, and usage records match expectations.
5	Unused instances and temporary resources are stopped or removed.

Troubleshooting

Symptom	Check First
No image or flavor is available	Region, authorization, image state, cluster association, and quota
IDE cannot be opened	Instance state, service port, network entry, and logs
Data is missing	Storage component, mount path, permissions, and region
Workload remains pending or fails	Capacity, image, command, accelerator, and quota
Artifacts disappear	Whether they were written to temporary storage and synchronized before completion