It is shipping collation-sensitive aggregates between servers which have different collations.
commit 7012b132d07c2b4ea15b0b3cb1ea9f3278801d98
Date: Fri Oct 21 09:54:29 2016 -0400
postgres_fdw: Push down aggregates to remote servers.
I've attached a reproducing case. Before this commit, the two final queries give the same answer, and after they give different answers. Maybe this isn't considered a bug? Is it just one of the "surprising semantic anomalies may arise when types or collations do not match"? It should be able to know what collation the local definition of the foreign table has; couldn't it pass that collation over the foreign side?
I don't really care, I was just using min as a way to get an arbitrary value in the cases where there are more than one, but I found the change surprising.
Cheers,
Jeff