Make staged, downtime-free deployments possible

[schurzi]

Is your feature request related to a problem? Please describe.

When deplyoing a new version we want to update our environment without downtime and without showing partial new content.
Sometimes new version of our application depend on specific new content (or vice versa) and we cannot do a "just restart and wait".

Describe the solution you'd like

To achieve a downtime free deployment folloing steps are neccessary:

1. stop replication to all RLS Pods
2. update CS/MLS
3. import, publish content
4. now for each RLS/CAE pair/group
   • remove CAE from loadbalancing
   • stop CAE and RLS
   • update RLS
   • wait for replication to finish
   • update CAE
   • insert CAE into loadbalancing

Maybe other backing services (like Solr) are also affected by this.