"couldn't find the type 'vector' in the types registry"

I'm trying to run the PgVector benchmark with the VectorDBBench tool against Azure PostgreSQL, but the benchmark consistently fails with the following error:

KeyError: "couldn't find the type 'vector' in the types registry"

This error occurs during the data insertion phase. The benchmark successfully connects to the database and registers the vector type in the main process, but fails when trying to insert data.

Azure PostgreSQL Configuration
I've set up Azure PostgreSQL specifically for this benchmark:

• Created an Azure Database for PostgreSQL server
• Added the "vector" extension to the allowlist in Azure Portal:
  Went to Azure Portal → PostgreSQL server → Settings → Server parameters
  Added "vector" to the "azure.extensions" parameter
• Added my client IP address to the firewall rules in Azure Portal:
  Went to Azure Portal → PostgreSQL server → Settings → Connection security
  Added my client IP to the allowed IP addresses

Changes Made to Support Azure PostgreSQL:

Modified the connection parameters to use SSL (required by Azure):
conn = psycopg.connect(**kwargs, sslmode="require")

Added explicit creation of the pgvector extension before registering the vector type:
conn.autocommit = True cursor = conn.cursor() cursor.execute("CREATE EXTENSION IF NOT EXISTS vector") conn.commit() register_vector(conn)

Error Logs:
2025-04-25 11:34:19,888 | INFO: Successfully registered vector type (pgvector.py:94) (8951) 2025-04-25 11:34:20,292 | INFO: (SpawnProcess-1:1) Start inserting embeddings in batch 100 (serial_runner.py:43) (8951) 2025-04-25 11:34:20,293 | INFO: Get iterator for shuffle_train.parquet (dataset.py:268) (8951) 2025-04-25 11:34:20,754 | WARNING: Failed to insert data into pgvector table (pg_vector_collection), error: "couldn't find the type 'vector' in the types registry" (pgvector.py:472) (8951) 2025-04-25 11:34:20,768 | WARNING: VectorDB load dataset error: "couldn't find the type 'vector' in the types registry" (serial_runner.py:141) (7992)

Read pgvector.py around vector registration and insertion, then serial_runner.py around the spawned insertion worker; compare what the main process registers with what the worker sees. Reproduce the failure using the reported Azure PostgreSQL SSL and extension setup, and verify that batch insertion completes without the type-registry error.