ores in beta was like this:

• A web node (deployment-ores-web)
  • role::labs::ores::web
• A worker node sitting with other service (deployment-sca01)
  • role::labs::ores::worker
• A redis node (deployment-ores-redis)
  • role::labs::ores::redis

but that couldn't work for several reasons and ores-beta couldn't return scores for edits

So I made deployment-sca03 (since it was taking too much of RAM) and enabled ores production roles there. So now it's like this:

• A web and worker node (deployment-sca03)
  • ores::web
  • ores::worker
• A redis node (deployment-ores-redis)
  • role::labs::ores::redis

Now it works just fine, and every release in production must be done here beforehand.

(A side note: @thcipriani had to do ssh-keyscan -H deployment-sca03.deployment-prep.eqiad.wmflabs >> /etc/ssh/ssh_known_hosts as root in deployment-tin in order to let us deploy from deployment-tin)