spring.datasource.initialization-order
was introduced in M2 to allow DataSource initialization to be deferred until after any EntityManagerFactory
had been created. This meant that the general initialization had become somewhat aware of a JPA specific concept. That's not too bad for JDBC, but when we support initialization with R2DBC as well it'll become a bigger problem.
We should replace spring.datasource.initialization-order
with a JPA-specific property.