Hi Anthony, On Sat, Sep 06, 2025 at 11:33:31PM -0400, Anthony Deschamps wrote: > Hello, > > First off, my apologies for the bug. This change was my contribution, > and I hope the regression didn't cause serious issues for anyone.
Don't be sorry, bugs happen. I can say it's also my fault for not having spotted it in the review. > Personally, I don't have any deployments that would be negatively > affected by simply applying the fix. OK that's what I wanted to know, thank you! > Of course, that may result in a burst > of cache misses for some users -- I can't speak for every system, but > my general inclination is that it's better to (if necessary) take one-time > mitigating actions such as adding capacity or deploying during a low > traffic period if it means that the ongoing maintenance/complexity will > be minimized. Yep. I tend to think that large cache deployments are not very frequent, and that most users either have not upgraded yet or didn't feel the rehash because the load was low. Also that's the type of environment where you generally don't want to take the risk to install the latest version but prefer the stability of the LTS version that works for you. Here the users who reported this issue were on 2.8 and were trying an upgrade to 3.0. 3.0 is only one year old, which is what makes me think that it's unlikely that many users have upgraded from an older version. And most new deployments would likely have read the doc and used the hash-key parameter. I'll privately check with a few large-scale users, to be sure. Thanks, Willy
