For simple apps with not much logic, you can get away with different front-ends/offline-syncs adding their own rows, and then only respecting the latest row in the main db, usually based on timestamp. But if you have something more complex where multiple workflows get triggered as side effects, then you may run into trouble when the latest row is in fact not the truth of the reality that we wanted, so there is still some risk ending up with states or workflows based on a false reality.
This can become a nightmare if you a have few scheduled events/tasks/crons that does things periodically, thus the only way to mitigate that is to fully embrace eventual consistency and idempotency, and the "easiest" way to get there is to embrace the actor model paradigm (see erlang, akka/akka.net, Orleans framework, F# mailboxes, go-routines, etc).
Point being, comparing two versions of applications against each other is not enough - you may need to version your data too and use timestamps to make a final decision on which version is truth. You may also need to set or build a tolerance system to say it will only sync "old" data if within x amount of days, lets say less than 7 days old. And so on.
For simple apps with not much logic, you can get away with different front-ends/offline-syncs adding their own rows, and then only respecting the latest row in the main db, usually based on timestamp. But if you have something more complex where multiple workflows get triggered as side effects, then you may run into trouble when the latest row is in fact not the truth of the reality that we wanted, so there is still some risk ending up with states or workflows based on a false reality.
This can become a nightmare if you a have few scheduled events/tasks/crons that does things periodically, thus the only way to mitigate that is to fully embrace eventual consistency and idempotency, and the "easiest" way to get there is to embrace the actor model paradigm (see erlang, akka/akka.net, Orleans framework, F# mailboxes, go-routines, etc).
Point being, comparing two versions of applications against each other is not enough - you may need to version your data too and use timestamps to make a final decision on which version is truth. You may also need to set or build a tolerance system to say it will only sync "old" data if within x amount of days, lets say less than 7 days old. And so on.