fix: PersistentIndex.need_rebuild() should trigger rebuild when index is empty

[yc111233]

Bug Description

PersistentIndex.need_rebuild() always returns False, which prevents the index maintenance scheduler from ever rebuilding the index, even when the index is empty but the store contains data.

Root Cause

When upsert_data() is called, it writes to both the store (LevelDB) and attempts to update the FAISS index via upsert_to_index(). However, for persistent indexes, the C++ FAISS engine can silently fail to write data, resulting in:

• Store growing with data (e.g., 8.4MB, 1002 entries)
• FAISS index remaining empty (0 vectors)
• vector/count API returning 0

The index maintenance scheduler runs every 30 seconds and calls _rebuild_indexes_if_needed(), which checks index.need_rebuild(). Since PersistentIndex.need_rebuild() always returns False, the scheduler never triggers a rebuild, and the index stays permanently empty.

Impact

• Memory search returns no results even after successful embedding/vectorization
• ov reindex --regenerate completes but vectors are not searchable
• The only workaround is manual source code patching

Fix

Change PersistentIndex.need_rebuild() to check if the index has zero data entries. When the index is empty, return True to trigger the maintenance scheduler to rebuild from the store.

# Before
def need_rebuild(self) -> bool:
    return False

# After
def need_rebuild(self) -> bool:
    try:
        if self.get_data_count() == 0:
            return True
    except Exception:
        pass
    return False

Testing

1. Import memories via ov reindex --regenerate
2. Verify store has data: curl http://127.0.0.1:1933/api/v1/stats/memories
3. Wait 30 seconds for maintenance scheduler
4. Verify index is populated: search API returns results (previously returned 0)

Environment

• OV Version: 0.3.3
• Backend: local (FAISS)
• OS: Linux

[github-actions]

Failed to generate code suggestions for PR