I've migrated a production app from mysql to pgsql although not with django but a different orm and language. The app also had some custom sql. The reason was for proper json support as we have a 2tb table with unstructured data (1tb at the time) that was slowing mysql down severely (most used table by far). A later version of mysql with json support was not available on aws so wasn't an option. This was about two to three years in so the app was actually almost finished and in maintenance mode.

It worked. It took many months and copying the data and doing the migration was a pita. The orm helped in that I had to modify fewer queries by hand, but modifying the queries we did have was trivial also. It's one of only two times off the top of my head that the orm was a net benefit rather than a liability in the apps that use it. Would I use an orm in the future in case this comes up? Hell no. I'd pick a proper db first, aka pgsql (that decision was before my time). But even if I didn't, I'd still prefer to convert the sql one by one from mysql to pgsql rather than incur the horrific penalties and extremely slow run time and development time the orm imposed on us. Developing with the orm was many times slower and the runtime was ten times slower than the actual query time due to hydration. Rewriting a few thousand queries is nothing compared to losing months to the slow development speed and the slow runtime speed which required a completely separate rewrite of most of our functionality alongside the orm just so we could serve api requests in a manner somewhat reminiscent of not completely slow.

So yeah, even when switching from mysql to pgsql, the orm wasn't worth it. It never is or has been.