Task 4: BaseModel Controller PVC Handling

Overview

Update the BaseModel controller to handle the complete PVC storage flow: validation, job creation, and status management. This replaces the original design where model agent was involved.

Scope

• Add PVC validation logic to controller reconciliation
• Create metadata extraction jobs for PVC models
• Monitor job status and update BaseModel accordingly
• Handle all PVC-related errors and edge cases

Files to Modify

• pkg/controller/v1beta1/basemodel/controller.go - Main implementation
• config/rbac/role.yaml - Add PVC and Job permissions
• pkg/controller/v1beta1/basemodel/controller_test.go - Tests

Implementation Details

Key Functions to Add

1. PVC Detection in Reconcile Loop

   • Check if BaseModel uses PVC storage URI
   • Route to PVC-specific reconciliation logic
   • Maintain existing behavior for non-PVC models

2. PVC Validation

   • Parse PVC URI to extract components
   • Verify PVC exists in the same namespace
   • Check PVC binding status
   • Update BaseModel status with validation results

3. Metadata Extraction Job Management

   • Create Kubernetes Job for metadata extraction
   • Job should mount PVC read-only
   • Monitor job status (running, succeeded, failed)
   • Handle job lifecycle (creation, monitoring, cleanup)

4. Status Updates

   • Update BaseModel status based on PVC and job states
   • Store PVC information in status annotations
   • Clear node-related status fields (not applicable for PVC)
   • Handle error messages and failure reasons

Job Specification Requirements

The metadata extraction job should:

• Use the ome-agent image with model-metadata command
• Mount the PVC at a specific path with subpath support
• Run with appropriate resource limits
• Have a reasonable timeout (e.g., 5 minutes)
• Include TTL for automatic cleanup
• Use dedicated ServiceAccount with minimal permissions

Controller Watch Configuration

Update the controller to:

• Watch owned Jobs for status updates
• Filter ConfigMap watches to exclude PVC models
• Add appropriate event handlers

Test Cases

1. Valid PVC Flow:

   • PVC exists and is bound
   • Job created successfully
   • Metadata extracted and updated
   • Status becomes Ready

2. PVC Validation:

   • PVC not found
   • PVC not bound
   • Invalid PVC URI

3. Job Management:

   • Job creation
   • Job success handling
   • Job failure handling
   • Job already exists

4. Edge Cases:

   • Metadata already populated
   • Job timeout
   • Controller restart during job

Acceptance Criteria

• Controller validates PVC existence and binding
• Creates metadata extraction job with correct spec
• Monitors job status and updates BaseModel accordingly
• Handles all error cases gracefully
• Status reflects PVC storage accurately
• No interference with non-PVC models
• Comprehensive test coverage

Dependencies

• Task 1: Storage URI Parsing
• Task 3: Model Metadata Agent

Estimated Effort

5-6 hours (increased due to complete flow ownership)