Expected Behavior

When vector_enabled: true is configured and the pgvector extension is installed in the database (in public schema), Feast should be able to create tables with vector columns. The database's configured search_path should be respected.

Current Behavior

Feast fails with psycopg.errors.UndefinedObject: type "vector" does not exist because it hardcodes the connection search_path to only the db_schema value, excluding public where extensions are typically installed:

 # feast/infra/utils/postgres/connection_utils.py#L76
  "options": "-c search_path={}".format(config.db_schema or config.user),

This overrides any database or user-level search_path settings.

Steps to Reproduce

1. Install pgvector extension in PostgreSQL:
   CREATE EXTENSION vector; -- installs to public by default
2. Configure feature_store.yaml with a custom schema:
   online_store:
   type: postgres
   db_schema: feast_online
   vector_enabled: true
3. Run feast apply
4. Error occurs:
   psycopg.errors.UndefinedObject: type "vector" does not exist
   LINE 8: vector_value vector NULL,

Specifications

• Version: 0.58.0
• Platform: macOS / Linux
• Subsystem: PostgreSQL