Breaking: Remove courseassets module, inline _assetIds on content documents

[taylortom]

Summary

The courseassets collection maintains a materialized join table (courseId + contentId → assetId) that is rebuilt on every content insert/update/delete. This is the primary performance bottleneck during clone operations — profiling shows postInsertHook (courseassets) takes 5-6 seconds per 242 items on the server, and with 8 concurrent multilang clones, causes 37+ seconds of contention.

The proposal is to store _assetIds: [...] directly on each content document, eliminating the separate collection and the courseassets module entirely.

Current problem

• Courseassets hooks into content's postInsertHook and runs per-item: deleteMany + getSchema + extractAssetIds + per-asset findOne + insert
• For 242 items with assets: 300-500+ DB round-trips per clone
• 8 concurrent multilang clones = 2,400-4,000 queries competing for the same DB connection
• Clone of 242-item × 8-language course: ~47 seconds, of which ~37 seconds is courseassets

Proposed changes

Remove courseassets module entirely:

• The only remaining logic is the asset deletion guard (~10 lines), which moves into ContentModule.init() as a preDeleteHook tap on the assets module
• Remove the courseassets collection, courseasset schema, routes, and package
• Remove adapt-authoring-courseassets from all peerDependencies and workspace configs
• The RESOURCE_IN_USE error moves to the content module's error definitions
• POST /api/courseassets/query has no frontend consumers (confirmed — the UI does not reference courseassets anywhere). Only server-side consumers are AdaptFrameworkBuild and AdaptFrameworkImport, which switch to querying content's _assetIds field directly

Content module:

• Move extractAssetIds utility from courseassets into content's lib/utils/
• Compute and store _assetIds during insert() and via preUpdateHook tap
• Add MongoDB index on { _assetIds: 1 }
• Add asset deletion guard in init() (tap assets.preDeleteHook)
• Clone: zero changes needed — _assetIds copies automatically with the document

AdaptFrameworkBuild:

• Change loadAssetData to query content.find({ _courseId }, { projection: { _assetIds: 1 } }) instead of courseassets.find({ courseId })

AdaptFrameworkImport:

• Remove courseassets dependency; _assetIds computed automatically during content insert

Schema (adapt-schemas):

• Add _assetIds array field to content schema with editorOnly flag

Migration

One-time migration to populate _assetIds on existing content documents:

1. For each content document, resolve its schema and extract asset IDs
2. Bulk $set _assetIds on all documents
3. Drop the courseassets collection

Performance impact

Phase	Before	After
Clone (per language)	300-500 DB queries (5-6s)	0 extra queries
Clone (8 languages)	37-47s total	~1-2s total
Regular insert	1 + 2N queries (N = assets)	0 extra queries (computed inline)
Regular update	1 + 2N queries	0 extra queries (computed inline)
Asset delete check	1 query on courseassets	1 query on content._assetIds (indexed)

Rollout

Can be phased:

1. Content module adds _assetIds computation (backward-compatible, courseassets still works in parallel)
2. Run migration to populate existing content
3. Switch consumers (build, import) to read _assetIds from content
4. Deprecate and remove courseassets module (breaking major version)

Related

• Breaking: Improve performance #108 (content performance improvements)
• New: Add dedicated tree endpoint for lightweight content structure queries #109 (clone optimisation)