discovery(celery): cleanup_orphan_documents, cleanup_generated_files, and 4 memory tasks also unregistered prior to MVA-341 fix #7766

Discovery

While fixing MVA-341 (KeyError on tasks.prune_sync_queue_done), the root cause was found to affect all tasks in knowledge_tasks.py and memory_tasks.py — not just prune_sync_queue_done.

Unregistered tasks before fix:

• tasks.cleanup_orphan_documents (beat-scheduled)
• tasks.cleanup_generated_files (beat-scheduled)
• tasks.refresh_system_knowledge
• tasks.reindex_knowledge_base
• tasks.scan_man_page_changes
• tasks.full_man_page_index
• memory.write_verbatim (queued from chat pipeline)
• memory.extract_facts (queued from chat pipeline)
• memory.update_graph (queued from chat pipeline)
• memory.compact_snapshot (queued from chat pipeline)

The beat-scheduled knowledge tasks likely also fired with KeyError on every run. The memory tasks (memory.* queue) likely silently dropped or errored.

Fix Applied

MVA-341 (branch issue-MVA-341) adds all missing imports to tasks/__init__.py. This issue tracks:

1. Auditing Celery worker logs to confirm how long cleanup_orphan_documents and cleanup_generated_files have been failing silently
2. Checking if orphan documents and expired generated files have accumulated due to missed prune runs
3. Auditing whether memory tasks have been reliably processed or whether chat memory is missing data

Severity

The memory.* tasks are on the critical write path for chat memory (ChromaDB verbatim store). If they have been failing, user conversation memory may have gaps.

Filed By

BackendEngineer during MVA-341 implementation (2026-05-16)