Distribution updates during replication should be atomic #7333
Description
Is your feature request related to a problem? Please describe.
When a replication is triggered, each individual replica is not coordinated in terms of when the distributions switch over. Also, the repositories within the replica are not processed in a consistent order, nor is it atomic. Therefore clients looking at the repositories on a replica can run into dependency conflicts (e.g. metalink files in a file repository depend on the associated rpm repositories being synchronised at the same time) inside a certain window while replication is in progress, because not all the repositories on the replica are reflecting the new target state, the distributions are updated whenever that individual repository has synced, they are not updated in one go at the end of replication.
Describe the solution you'd like
All distributions within a replica during a replication operation should switch over atomically, or at least in a coordinated fashion once all repositories have synced, generated publications etc.
* There is also a potential use case for the final distribution update to optimally be separate and manually triggered - as this would allow the external management application to trigger replication on a number of replicas and once all replicas report that they are ready to update their distributions, that external application can trigger that distribution update on all replicas simultaneously.
* Stretch goal: All replicas should, if possible, also be coordinated in their switching, or provide the ability to coordinate.
Additional context
Add any other context about the feature request here.