Auth issues present as "ValueError: Id column, langchain_id, does not exist."

[glasnt]

Environment details

• OS type and version: Colab Enterprise
• Python version: py-310 (from reasoning-engine image)
• pip version: pip --version Unsure
• langchain-google-alloydb-pg version: pip show langchain-google-alloydb-pg
  • both langchain-google-alloydb-pg-0.8.0 and langchain-google-alloydb-pg-0.7.0

Steps to reproduce

1. Try running Deploying a RAG Application with AlloyDB to LangChain on Vertex AI
2. Use suggested IAM-based authentication in similarity_search
3. Get ValueError

Code example

Directly from notebook

def similarity_search(query: str) -> list[Document]:

    engine = AlloyDBEngine.from_instance(
        PROJECT_ID,
        REGION,
        CLUSTER,
        INSTANCE,
        DATABASE,
        # Uncomment to use built-in authentication instead of IAM authentication
        # user="postgres",
        # password=PASSWORD,
    )
    )

    vector_store = AlloyDBVectorStore.create_sync(
        engine,
        table_name=TABLE_NAME,
        embedding_service=VertexAIEmbeddings(
            model_name="textembedding-gecko@latest", project=PROJECT_ID
        ),
    )
    retriever = vector_store.as_retriever()
    return retriever.invoke(query)


vertexai.init(project=PROJECT_ID, location="us-central1", staging_bucket=STAGING_BUCKET)

remote_app = reasoning_engines.ReasoningEngine.create(
    reasoning_engines.LangchainAgent(
        model="gemini-pro",
        tools=[similarity_search],
        model_kwargs={
            "temperature": 0.1,
        },
    ),
    requirements=[
        "google-cloud-aiplatform[reasoningengine,langchain]",
        "langchain-google-alloydb-pg",
        "langchain-google-vertexai",
    ],
    display_name="PrebuiltAgent",
)

response = remote_app.query(input="Find movies about engineers")
print(response["output"])

Stack trace

...
File "/usr/local/lib/python3.10/concurrent/futures/_base.py", line 403, in __get_result
raise self._exception
File "/usr/local/lib/python3.10/site-packages/langchain_google_alloydb_pg/async_vectorstore.py", line 168, in create
raise ValueError(f"Id column, {id_column}, does not exist.")
ValueError: Id column, langchain_id, does not exist.

Debugging

Example schema from movies.csv uses id, wasn't sure if I should have been renaming this (it wasn't it).

But after finding the code in this package, I saw that it could be a case of the columns var in async_vectorstore.py's create() (via AlloyDBVectorStore.create_sync) is empty, hence the error.

If I changed the authentication from IAM built-in to the username password, the functionality worked.

Possible fix: before the # Check columns maybe check if there are any columns, and return an authentication error. This would at least present the actual issue rather than a schema issue.

b/356000377

[averikitsch]

Thank you so much @glasnt for the detailed bug!