What portion of writes typically have to go to the primary server? If users are local then their data maybe is often local as well. In other words not all writes need to first go the central server from where they are distributed back to all edge-databases.
I can see that the organization running the application needs a global view of all data. But regional users perhaps don't. Often they just need to know their own data.
Write first to edge then copy to central database rather than write first to central database then trickle down to the site from where the write originated in. Just wondering what portion of applications could use this alternative design.
It’s been done and it becomes quite complicated very quickly. Then it becomes necessary to manage that complexity and you end up with something like Spanner.
This could work in very small projects. Still very risky.
Say you flag a write as "local". Later, some other place in your app starts relying on this write in another locality. If you don't update your write spec from "local" to "primary", you don't have a consistent database anymore, but you will make decisions thinking that you do.
I can see that the organization running the application needs a global view of all data. But regional users perhaps don't. Often they just need to know their own data.
Write first to edge then copy to central database rather than write first to central database then trickle down to the site from where the write originated in. Just wondering what portion of applications could use this alternative design.