The main problem is "persistent storage". In k8s, a PVC/PV is probably close to the right answer, but in general that gives you a file-system view of the storage, not a block-device view, which for some databases might matter).
Also, in general, that persistent storage is (generally) not locally attached, which means that (some) file system interactions may have unexpected semantics. Now, with a DB, you're probably only going to have the PV attached on a single worker node, so most of that will actually just work.
Also, in general, that persistent storage is (generally) not locally attached, which means that (some) file system interactions may have unexpected semantics. Now, with a DB, you're probably only going to have the PV attached on a single worker node, so most of that will actually just work.