Xin Sun

> @ddupg curious if you have any workaround for this on live cluster? do you restart regionservers? @virajjasani Previously we had to restart regionservers, after this patch, the problem still...

@adamretter Thank you for your reply. I found this [PR](https://github.com/facebook/rocksdb/pull/8638) that supports customizing the `TablePropertiesCollectorFactory` with a string, and I saw an option called "table_properties_collectors" [here](https://github.com/facebook/rocksdb/blob/22fe23edc89e9842ed72b613de172cd80d3b00da/options/cf_options.cc#L782), but I'm not sure...

@adamretter Thank you for your detailed response. Given this situation, I will attempt alternative methods. Looking forward to the day when this feature is available in RocksJava.

We are currently exploring enhancing Lance's multimodal capabilities and came across this issue. Could @rok @wjones127 share why it was not continued?

> I don't like this, as this is very brittle, and we will have the same problem again soon. (The previous patch has the same problem) > > The real...

Hi @westonpace @jtuglu1 , is this issue still being worked on? I’ve come up with an approach: 1. We can request the the data be sorted by value using `BitmapIndexPlugin::new_training_request`...

> Hmm, this issue is mainly a problem at large scales. However, at large scales, the out-of-core sort will be expensive. > > I wonder if it would be better...

> Hmm, this issue is mainly a problem at large scales. However, at large scales, the out-of-core sort will be expensive. I implemented spilling in #5201 and performed performance tests....

The python example is currently blocked at #5144

Could @jackye1995 please review this again when you have time?