Enable Dynamic Specification of Custom Schema Names in PostgreSQL PGVector vector store

[paulonasc]

Checked

• I searched existing ideas and did not find a similar one
• I added a very descriptive title
• I've clearly described the feature request and motivation for it

Feature request

As a user of LangChain, I want to be able to specify a custom database schema at runtime when initializing the PGVector class.

This feature proposal aims to enhance the LangChain framework by enabling dynamic specification of database schema names when initializing the PGVector class. This functionality will allow users to set custom schemas for their database tables directly through the class constructor, thereby facilitating decoupling.

Relevant Discussion: #18883

Motivation

In current implementations of the LangChain framework, the database table name is hardcoded without a schema name being specified, therefore having it default to public. This makes the PGVector connector lack the flexibility needed for dynamic configuration during runtime. This limitation poses challenges such as multi-tenant architectures where different clients or projects may require separate schemas.

It is frustrating to have to manually adjust the selected database schema within the codebase.

Relevant Discussion: #18883

Proposal (If applicable)

1. Add a Schema Parameter to PGVector Constructor:

• Add a schema_name parameter to the PGVector constructor.
• Store the schema_name parameter in an instance variable for use in the _get_embedding_collection_store function.

2. Modify _get_embedding_collection_store Function:

• Modify the _get_embedding_collection_store function to accept schema_name as a parameter.
• Update the model classes within the function to use the passed schema name for table definitions.

[paulonasc]

@hwchase17 👋 how can I turn this idea into a feature? I'd like to add the code changes and open a PR. Do I have to follow a process? Thanks!