[Bug]: Local vectordb backend fails with RocksDB LOCK conflict - vector search returns NoneType

[happy5318]

Bug Description

When using the local vectordb backend, the OpenViking server fails to initialize the vector database properly, resulting in:

1. HTTP server (port 1933) never binds despite process running at 100% CPU
2. Vector search fails: 'NoneType' object has no attribute 'search_similar_memories'
3. Memory extraction fails: 'NoneType' object has no attribute 'enqueue_embedding_msg'
4. QueueManager stops after processing pending messages

Error Logs

2026-03-27 08:03:12,164 - vikingdb - ERROR - Failed to open data db: IO error: lock /path/to/vectordb/context/store/LOCK: Resource temporarily unavailable
2026-03-27 08:03:12,164 - openviking.storage.viking_vector_index_backend - ERROR - Error creating collection context: IO error: lock .../LOCK: Resource temporarily unavailable
2026-03-27 08:03:16,607 - openviking.session.memory_deduplicator - WARNING - Vector search failed: 'NoneType' object has no attribute 'search_similar_memories'
2026-03-27 08:03:19,432 - openviking.storage.queuefs.queue_manager - WARNING - [QueueManager] Worker thread Semantic did not exit in time
2026-03-27 08:03:19,438 - openviking.storage.queuefs.queue_manager - INFO - [QueueManager] Stopped

Steps to Reproduce

1. Fresh install OpenViking v0.2.13
2. Configure with local vectordb backend:

   {
     "storage": {
       "workspace": "/path/to/data",
       "vectordb": { "backend": "local" }
     },
     "embedding": {
       "dense": { "provider": "openai", "model": "text-embedding-3-large", "dimension": 3072 }
     }
   }

3. Import memories (8,000+ records)
4. Start server: python -m openviking.server.bootstrap --host 127.0.0.1 --port 1933 --config ov.conf
5. Observe: Process runs at 100% CPU, port never binds, vector operations fail

Environment

• OpenViking version: 0.2.13
• Python: 3.10
• OS: Linux (Ubuntu 24.04)
• Deployment: Bare metal (not Docker)

Additional Context

• This appears related to issue RocksDB LOCK conflict: vectordb local backend fails with 'already held by process' #925 (RocksDB LOCK conflict)
• The LOCK file exists but the error suggests the vector database object is not properly initialized
• After deleting the LOCK file and restarting, the same error persists
• Console UI (port 8020) works fine, only the backend server is affected

Expected Behavior

• Vector database should initialize successfully
• HTTP server should bind to port 1933
• Vector search and memory extraction should work

Actual Behavior

• Vector database initialization fails
• HTTP server never binds
• All vector operations fail with NoneType errors

[zhoujh01]

Are there multiple openviking processes running simultaneously that are reading from the same workspace? Only one process can be reading at a time.