Hi Adrian, thanks for your reply. Here is the clarification.
1) It is indeed a pg_upgrade from 9.2 to 10.4. Depending on the test machine, it runs in between 15 and 20 minutes for just over 100GB. I can negotiate this time with our customer. The vacuum process took another 5 to 7 minutes. This this what I was referring to with the 30 minutes (point 3 in your questions)
2) After pg_upgrade, I published the tables on the database (in the sense "CREATE DATABASE") and subscribed to this publication on the second server (logical replication). The data copy processed started immediately and took around 1 hour. I then loaded the indexes, what took another 2h20m. At that point the active-passive cluster was ready to go. Note that the active and the passive databases are on different machines.
4) By "database" I mean the result of "CREATE DATABASE" and we have 1 per server (or "cluster" in your terminology - I tend to use this word for a group of machines). We are currently using a streaming replication between the 9.2 servers, so it could be a fall-back option after the upgrade (I wanted to remove part of the indexes on the master to lower the load, reason to use the logical replication... if the execution time is not too excessive).
Hope it clarifies the question
Best regards
Olivier
