We used it in production for a few months with around 4k QPS and a P99 of ~100ms. We specifically were using it because it has a "follow the workload" feature w

We used it in production for a few months with around 4k QPS and a P99 of ~100ms. We specifically were using it because it has a "follow the workload" feature where it moves ranges of data to the server/region where they're most used. We ran into a few panics (crashes) but support was great and they managed to fix the issues within a few days. Ultimately we stopped using it after multiple cascading failures that ended up bringing down the whole cluster. Granted we were running 4-core nodes with high CPU utilization so I'm sure it would've performed better with more cores but it wasn't in our budget.