We should treat all storage engines the same in so much as they should all defined both the data distribution (replicated vs. clustered vs. distributed) and the format (row vs. columnar), and make it easy to use multiple storage engines together... and provide a higher abstraction such that picking a storage engine is more like picking the characteristics required.

In the end, it should more/less no longer necessary to choose from InnoDB or Galera, or find the right combination of configuration options to make multiple storage engines work together.

It should be simple to create a table with replicated row data "engine=replicated", a table with clustered row data "engine=clustered", a table with distributed data "engine=distributed" and a table with columnar data "engine=columnar" all within the same database.

Small tables should use InnoDB. If they have customer generated data, maybe choose clustering for durability. If they have business generated data, maybe replication is fine. Then there may be very large tables or high throughput queries where distributed makes more sense. And finally, some tables may use columnar as transactional applications begin to incorporate more real-time analytics.