Add support to specify custom table schema

[Anupkumar Ghogare (anupkg1985)]

Current LangChain-Postgres framework, the database table name is hardcoded and defaults to the public schema due to the absence of a specified schema name. This limitation affects the flexibility of the PGVector connector, restricting its ability to dynamically configure during runtime. This poses significant challenges in multi-tenant architectures, where it's essential for different clients or projects to operate within separate schemas to ensure data isolation and management efficiency. Addressing this issue would involve enabling schema specification, thereby enhancing the connector's versatility and making it more suitable for diverse deployment scenarios.

Proposed solution -

• Introduced new argument/ parameter table_schema to `PGVector' class
• This new parameter is optional and defaulted to public schema

Discussions related to issue -
https://github.com/langchain-ai/langchain/discussions/20877
https://github.com/langchain-ai/langchain/discussions/18883

Example usage -

Default publicschema -

connection = "postgresql+psycopg://langchain:langchain@localhost:6024/langchain" 
collection_name = "my_docs"
embeddings = CohereEmbeddings()

vectorstore = PGVector(
    embeddings=embeddings,
    collection_name=collection_name,
    connection=connection,
    use_jsonb=True,
)

Specify custom table-schema -

connection = "postgresql+psycopg://langchain:langchain@localhost:6024/langchain" 
collection_name = "my_docs"
embeddings = CohereEmbeddings()

vectorstore = PGVector(
    embeddings=embeddings,
    collection_name=collection_name,
    connection=connection,
    use_jsonb=True,
    table_schema="custom"
)

[Marko (markobogoevski)]

Will this get merged? Im in need of setting up pg vector tables per environment which I am isolating by the schema

[Ben Bradford (bradfordben)]

I would also like this feature

[Thomas Brittain (Ladvien)]

Not having the ability to select a schema in a way that's idiomatic to langchain is problematic. Can this be merged?

[Ben Bradford (bradfordben)]

Do you need to link reviewers to get this reviewed and merged?

[Anupkumar Ghogare (anupkg1985)]

Do you need to link reviewers to get this reviewed and merged?

Yes, could you please provide the link?

[Ivan Krakhmaliuk (LifeDJIK)]

Table schema can already be specified using schema_translate_map from sqlalchemy:

engine_args={
    "execution_options": {
        "schema_translate_map": {
            None: "target_schema_name",
        },
    },
}

I think more useful will be to allow table names to be changed (like in #106).