Conversation
|
Hi, I’m Jit, a friendly security platform designed to help developers build secure applications from day zero with an MVS (Minimal viable security) mindset. In case there are security findings, they will be communicated to you as a comment inside the PR. Hope you’ll enjoy using Jit. Questions? Comments? Want to learn more? Get in touch with us. |
|
I realise now that on a fresh instance, unstable features needs to be enabled to use case():
So this change would be breaking until that feature becomes stable |
|
Hey @tomjohnhall, thanks for jumping on that! You're right that these changes are breaking until the If there isn't already one, could you raise an issue about that (along with a way to reproduce it, if you have one)? Once we have that, hopefully someone from our team, you, or someone else in the community can identify what's going on there and resolve it properly 🤞 |
|
Thanks @vishal-bala I've narrowed it down to an issue of scale, not number of vectors / missing values. Raised issue here |
2 participants
When running a multi vector query, I noticed redis would sometimes return:
Could not find the value for a parameter name, consider using EXISTS if applicable for distance_2This is running
redis-stack-server/7.4.0-v8from brew.I confirmed all documents contained the referenced embedding fields, but somehow the third distance was not always calculated and the query would error out. Maybe a match in other distances causes redis to skip the final distance calculation? Without debugging the specifics of query execution, this PR adds a case() in the query to set a default score of 0 for missing distances.
I also noticed that the vector radius is always hard coded to 2.0 - looks like there's a separate PR just opened to add custom ranges so I've left those changes out here. This fallback may solve issues downstream of that, and allow for safer handling of OR multi vectors of different ranges, rather than forcing AND.