Thanks for your comments. Optimizing the render part in insertion and deletion is also difficult. Some nodes might have user-level views, so some of them won't

Thanks for your comments. Optimizing the render part in insertion and deletion is also difficult. Some nodes might have user-level views, so some of them won't be visible to some other users. So you can't just label the nodes like this is "left" or "right" of that record. My experience with a similar problem: Instead of putting the rendering part to the insertion and deletion (nested set model), I was using a simple nginx file based cache. This is a trade off and I chose the application layer to be "consistent" and "reliable", over "being fast". Keeping the data integrity is difficult when you have a nested set model and it's also expensive to doing that calculation during the insertion. Sometimes you need a message broker to handle such an expensive query, because it's not safe to do it within a request-response lifecycle, even if you do have a transactional database, there's always a barrier of "request time out".