It makes sense when you have enough scale that DDLing all the sharded databases for every release from every team is too much overhead. (Storing unindexed colu

It makes sense when you have enough scale that DDLing all the sharded databases for every release from every team is too much overhead. (Storing unindexed columns (as JSON rather than binary though) is where I'm headed at my job, due to dynamic schema, the cost of coordinating DDL, and the excruciating cost of joins.)