When we want to go active/active mode. We duplicate preaching in both datacenters and randomly split requests between the datacenters. My suggestion is that in varnish and/or LVS, we make a setting that (for example) if URL of ores request matches "enwiki", "wikidatawiki" or "nlwiki".[1] The request should go to eqiad otherwise, send it to codfw. Like enwiki, wikidatawiki, and nlwiki belong to s1 shard and the others to s2 shard

In that case, we don't need duplication of precaching and we can have ChangeProp only does precaching of the proper wikis in the proper datacenter. That means real double capacity.

In ideal world, we might be able to load only proper wikis in ores nodes which would take less memory and we can increase the number of workers (i.e. quadruple capacity) but due to failover reasons, I think we should not let ores handle this (so once one datacenter goes down, simply the other one can handle all types of requests)

[1]: The exact wikis should be determined