Note the unix_socket_directory parameter, which is indeed being applied because we can see it again in the ps output: > pgsql 26636 0.0 1.4 66960 14512 - Is 4:08AM 0:00.06 > /var/tmp/pgbin.SPOsRj4D/bin/postgres -D /usr/local/pgsql/data91 -p 50432 -b > -c listen_addresses= -c unix_socket_permissions=0700 -c > unix_socket_directory=/usr/local/pgsql
However, your psql is looking for the socket in /tmp: > $ psql -p 50432 -d template1 > psql: could not connect to server: No such file or directory > Is the server running locally and accepting > connections on Unix domain socket "/tmp/.s.PGSQL.50432"?
You could successfully connect to that server with "psql -p 50432 -h /usr/local/pgsql ...", I expect.
The question is why pg_upgrade issued that option and then failed to cope with the consequences. I suspect it has something to do with one installation being configured with different default socket directory than the other, but I don't have enough facts.
regards, tom lane
Yes, it does connect using Unix domain socket as you suggested. PG 9.5 is the stock install as present on FreeBSD. I will have to check the script that installs PG 9.1 in an alternate location for any changes from the default.