Amazon Web Services offer a number of continuous deployment and management tools such as Elastic Beanstalk, OpsWorks, Cloud Formation and Code Deploy depending on your needs. The basic idea being to facilitate code deployment and upgrade with zero downtime. They also help manage best architectural practice using AWS resources.
For simplicity lets assuming a basic architecture where you have a 2 tear structure; a collection of application servers behind a load balancer and then a persistence layer using a multi-zone RDS DB.
The actual code upgrade across a fleet of instances (app servers) is easy to understand. For a very simplistic overview the AWS service upgrades each node in turn handing connections off so the instance in question is not being used.
However, I can't understand how DB upgrades are managed. Assume that we are going from version 1.0.0 to 2.0.0 of an application and that there is a requirement to change the DB structure. Normally you would use a script or a library like Flyway to perform the upgrade. However, if there is a fleet of servers to upgrade there is a point where both 1.0.0 and 2.0.0 applications exist across the fleet each requiring a different DB structure.
I need to understand how this is actually achieved (high level) to know what the best way/time of performing the DB migration is. I guess there are a couple of ways they could be achieving this but I am struggling to see how they can do it and allow both 1.0.0 and 2.0.0 to persist data without loss.
If they migrate the DB structure with the first app node upgrade and at the same time create a cached version of the 1.0.0. Users connected to the 1.0.0 app persist using the cached version of the DB and users connected to the 2.0.0 app persist to the new migrated DB. Once all the app nodes are migrated, the cached data is merged into the DB.
It seems unlikely they can do this as the merge would be pretty complex but I can't see another way. Any pointers/help would be appreciated.