mirror of
https://github.com/open-webui/open-webui.git
synced 2026-05-08 04:16:03 -05:00
[GH-ISSUE #1599] Large amount of RAG data possibly causing crash consistently #28091
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: github-starred/open-webui#28091
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @yesilzeytin on GitHub (Apr 18, 2024).
Original GitHub issue: https://github.com/open-webui/open-webui/issues/1599
Bug Report
Description
Bug Summary:
I have been working with Open WebUI for a few days now using several different GGUF models on a weak hardware. With deepseek-coder:1B, it has been going fine and smoothly with and without RAG. My documents until now were consisting of specifications and books converted from pdf to txt. However, when I tried to include 400 different code files with small sizes and tried to use their tag to provide as RAG data, the model just shuts down and gives the error: "Uh-oh! There was an issue connecting to Ollama." After that, I lose the Open WebUI itself. Therefore, there may be a problem with RAG implementation for large file counts. In Docker logs, I observe an error message related to "too many open files", so the problem most likely is due to that.
Steps to Reproduce:
Expected Behavior:
Open WebUI shouldn't have suddenly crash with many RAG data files, especially when it proves to be capable of working with larger data sizes.
Actual Behavior:
Open WebUI crashes and needs to be restarted to be able to work again.
Environment
Reproduction Details
Confirmation:
Logs and Screenshots
Browser Console Logs:
localhost.log
Docker Container Logs:
docker.log
Screenshots (if applicable):
Installation Method
Docker
Additional Information
Firstly, using this framework has been a blast so far and thanks a lot for creating and sharing such a wonderful thing. I am aware of that RAG feature is in Alpha stage and may have errors, but wanted to report it anyway thinking it could potentially help you further improve this already-great tool.
Note
If the bug report is incomplete or does not follow the provided instructions, it may not be addressed. Please ensure that you have followed the steps outlined in the README.md and troubleshooting.md documents, and provide all necessary information for us to reproduce and address the issue. Thank you!
@justinh-rahb commented on GitHub (Apr 18, 2024):
"Too many open files" would be an operating system imposed limit. There's kernel parameters that can be tuned to increase it, running
ulimit -n 2048in the container shell would temporarily double the default limit so you could test it.@yesilzeytin commented on GitHub (Apr 18, 2024):
I wasn't expecting such a quick response. Thanks a lot. Unfortunately, the method you proposed still cause the same error no matter what file limit I set. What you say sounds quite logical but I wonder if there could be something else considering this setup also fails:
@justinh-rahb commented on GitHub (Apr 18, 2024):
Will investigate further @yesilzeytin. It's also possible that the host machine itself needs this tweak applied as well.
@jarekmor commented on GitHub (Apr 19, 2024):
I have got the same problem.
I am using OI with the speakleash/Bielik-7B-Instruct-v0.1 model with 4K context window. The model is based on Mistral architecture and was fine-tuned on Polish data set. The problem i got is that after every 6-7 passages of conversation I got error mentioned above ("Uh-oh! "). My guess is that the cause is the short context window which can not handle properly the conversation history but I can be wrong. I am using it with my documents embedded into the vectorstore.
unlimit = unlimited on both docker container and my host machine.
@justinh-rahb commented on GitHub (Apr 19, 2024):
@jarekmor I've seen that you've posted this on our Discord. I don't believe it to be related to the issue in this thread, it's not the same error. If you'd like you should open a new issue for that matter if you aren't getting a response on Discord.
@tjbck commented on GitHub (May 7, 2024):
Closing in favour of #2044, Let's continue our discussion there!