[GH-ISSUE #3423] Discussion: Collection handling in vector database #28790

Closed
opened 2026-04-25 03:20:56 -05:00 by GiteaMirror · 0 comments
Owner

Originally created by @jonathan-rohde on GitHub (Jun 25, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/3423

I want to raise the following discussion.

Status quo:

  • One document results in one collection in the vector database
  • The hash of the file content is the collection name
  • The meta data contains the filename
  • Deletion of documents keep data still in vector database

Questions:

  • What do you think of storing all documents in a single collection?

    • My opinion: A vector database is more efficient finding the k nearest vectors from one single collection than a python application fetching k nearest vectors from x collections, sorting them and taking first k vectors from result.
      Filtering based on meta data can easily be implemented to limit a search based on the file hash.
      Please correct me, if my assumption is incorrect.
  • Shouldn't the vectors get deleted in the vector store, once the document is deleted?

    • I can see the benefit of having it still in the store as if the file is scanned/uploaded again, it an just be reused.
      But I see a problem, when for example we want to add or update meta data, like the file name.
Originally created by @jonathan-rohde on GitHub (Jun 25, 2024). Original GitHub issue: https://github.com/open-webui/open-webui/issues/3423 I want to raise the following discussion. Status quo: - One document results in one collection in the vector database - The hash of the file content is the collection name - The meta data contains the filename - Deletion of documents keep data still in vector database Questions: - What do you think of storing all documents in a single collection? - My opinion: A vector database is more efficient finding the k nearest vectors from one single collection than a python application fetching k nearest vectors from x collections, sorting them and taking first k vectors from result. Filtering based on meta data can easily be implemented to limit a search based on the file hash. Please correct me, if my assumption is incorrect. - Shouldn't the vectors get deleted in the vector store, once the document is deleted? - I can see the benefit of having it still in the store as if the file is scanned/uploaded again, it an just be reused. But I see a problem, when for example we want to add or update meta data, like the file name.
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/open-webui#28790