-
Notifications
You must be signed in to change notification settings - Fork 683
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
BM25 Index Corruption / BM25 not crash-resistant #4125
Comments
Known Mitigation techniquesMulti-TenancyIf MT, the bug typically only affects a single or very few tenants. Here, the easiest would be to reinsert the entire tenant (using the Cursor API): Export all objects for the tenant. Delete the tenant, recreate the tenant, reinsert the data. Single-TenancyEssentially, the same mitigation as for MT (whole collection instead of tenant), but it may not always be feasible due to the size. If possible, restore a backup from a point before the issue occurred. Known workaroundsPure Vector Search instead of HybridThe issue is isolated to BM25 indexing. If you are using Hybrid Search and running into this issue, one option could be to temporarily use pure vector search. |
Possibly related: #3548 |
We are optimistic that we can provide a fix soon. Some updates:
|
Checklist for RolloutBefore Release
Release
Can happen after initial release
|
Update: A chaos pipeline that reproduces the issue fully automatically (starting with an empty instance) is now available: weaviate/weaviate-chaos-engineering#172 |
Hello @etiennedi |
@amourao Could you tell me if this bug was fully fixed with your PR? |
This should not have any relation to the v4 client.
It can mitigate it in many scenarios. I wouldn't call it "fully fixed" because more crash-safety improvements are coming. But it can't hurt to try if it fixes all issues for you already. |
@etiennedi Thank you, I will test it out soon. Edit: did not see it happen again, thanks. |
Is there any update on this issue? This comment made is sound like a fix was in sight but there haven't been any recent updates. |
#4262 -> addresses |
How to reproduce this bug?
Not known yet - working on reproduction. Few independent sightings
What is the expected behavior?
BM25 index is crash-resistant
What is the actual behavior?
Users occasionally see panics like the following:
Supporting information
We are aware and investigating. If you have a reliable way to reproduce this (from scratch), please do let us know.
Server Version
Spotted on 1.22.x and 1.23.x
Code of Conduct
The text was updated successfully, but these errors were encountered: