delete and delete_collection on CrateDBVectorSearch don't delete the actual embeddings

[andnig]

System Info

Langchain 0.0.315

Who can help?

@amotl

Information

• [ ] The official example notebooks/scripts
• [ ] My own modified scripts

Related Components

• [ ] LLMs/Chat Models
• [ ] Embedding Models
• [ ] Prompts / Prompt Templates / Prompt Selectors
• [ ] Output Parsers
• [ ] Document Loaders
• [x] Vector Stores / Retrievers
• [ ] Memory
• [ ] Agents / Agent Executors
• [ ] Tools / Toolkits
• [ ] Chains
• [ ] Callbacks/Tracing
• [ ] Async

Reproduction

from langchain.schema import Document

doc = Document(page_content="this is such a nice text")
vector_store = CrateDBVectorSearch.from_documents([doc], OpenAIEmbeddings(), collection_name="wow_such_nice", connection_string="crate://localhost:4200?schema=langchain")
vector_store.delete_collection()

Expected behavior

The collection should be deleted as well as all embeddings which are part of this collection. While the collection gets deleted from the collection table, the embeddings of this collection in the embedding table are still there.

[amotl]

Hi @andnig,

thanks for your report. @ckurze recently reported GH-5, which sounds a bit similar to me, and I think we made a few improvements after that.

You are reporting that you observed the problem with LangChain 0.0.315. The current cratedb branch is on LangChain version 0.0.338. Can I ask you to try again, using the most recent version here?

With kind regards, Andreas.

[amotl]

Hi @andnig,

thanks for your report.

While the collection gets deleted from the collection table, the embeddings of this collection in the embedding table are still there.

The cascading delete does not work because CrateDB doesn't know anything about foreign key relationships. The corresponding operation will probably need to be emulated. I will look into how this could be implemented.

With kind regards, Andreas.

[hlcianfagna]

Relates with https://github.com/crate/crate/issues/1376

[ckurze]

Thanks for reporting - as soon as we slightly change the logic how we handle tables and embeddings (https://github.com/crate-workbench/langchain/issues/12), this operation will translate into a DELETE FROM <collection_name> or DROP TABLE <collection_name>, respectively. We should prioritize https://github.com/crate-workbench/langchain/issues/12 over this issue.

[amotl]

Dear @andnig,

GH-14 fixed this problem. Thanks again for the report.

With kind regards, Andreas.