In general, "ordinary" software has these advantages for the simple reason that it's ephemeral. The software starts with a clean universe each time you kill it and relaunch. That makes lots of things much, much easier.

Databases don't get to do that. The data is persistant between instances, between sessions and between copies. It never goes away and the database must always consider all other considerations secondary. If you want to change the object model of a Ruby program, you change the source and relaunch. If you want to change the schema of a database, you will need to provide the database instructions on what you want to do with all the existing data that is now inconsistent.

If you look at the hoops people jump through in always-on software, as opposed to "restart with #e33cf4" software, it's similarly gnarly.

Relational databases are kissing cousins with heavy-duty type systems. They make you do homework and eat your vegetables to forestall classes of error lots of people have never encountered and so discount to zero. Then it's decided that they aren't agile and fwoosh, out they go.