Qdrant Vector Store no Response

[r0kk]

Hi, when using Qdrant vector store and if user doesn't query for long period of time (hours), then retrieving from vector database doesn't work. Application basically hangs on the following line (docs) in pipelines.py:

        # rerank
        s_time = time.time()
        print(f"retrieval_kwargs: {retrieval_kwargs.keys()}")
        docs = self.vector_retrieval(text=text, top_k=self.top_k, **retrieval_kwargs)
        print("retrieval step took", time.time() - s_time)

I tried to keep Qdrant warm by sending queries every couple of minutes, but it didn't work, so my assumption is that it has to do something with how following behaves in pipelines.py:

    @Node.auto(depends_on=["embedding", "VS", "DS"])
    def vector_retrieval(self) -> VectorRetrieval:
        return VectorRetrieval(
            embedding=self.embedding,
            vector_store=self.VS,
            doc_store=self.DS,
            retrieval_mode=self.retrieval_mode,  # type: ignore
            rerankers=self.rerankers,
        )

My best guess would be that it has something to do with how Node is caching Qdrant client, which connection gets somehow stale. I would be vey happy if you have any idea what could this be. vector_retrieval starts working again if app is restarted or after about 3-4 chat attempts.

Thank you.

[taprosoft]

This might be related to internal caching mechanism from Kotaemon. Will investigate this issue further when I have time.