Re: Add relcreated (timestamp) column to pg_class catalog to record the time an object was created

"I just cannot see that in the following:
pg_upgrade 9.4 --> 9.5"
The 9.5 database is the same as the 9.4 one."

You are speaking of the case where relcreatedat did not/does not exist in the previous database?
True, but the whole point of this request is to "start obtaining creation dates". Once we are at a point
where we have pg_class with relccreatedat, then all subsequent create dates will be correct. I cannot think
of a single case where having incorrect creation dates from previous/upgraded databases will cause any harm
or hinder operation of the PostgreSQL. I can however, cite instances where users create their own tables but do
not notify the dba as such, then cry when something happens because they are not replicated, as in the case of
using slony. In which case having relcreatedat would go a long ways to preventing that.

"Master dies
Standby gets promoted to new Master.
STONITH the original Master
Work continues on the new Master.
The old Master is resurrected as a new Standby.
To me it would be important to know when the objects actually appeared in the various databases as a way of figuring what the above timeline was"

That is the whole point of relcreatedat. A properly replicated database brings over ALL needed data from the master to the slave(s), including created objects. It works when promoting the slave, and restoring the master providing you follow correct procedure. Otherwise, your replication is useless. I know when working with slony this can cause a problem, but that is a weakness of slony, not of relcreatedat. Besides, are you more concerned with keeping the database on line, or tracking object creation dates when PosgreSQL crashes? You are quoting the corner case. That's like saying "I refuse to wear a seat belt because I may pass out and drive into a lake", even though you are in Kansas and driving I-70.

--