I'm aware of how I can solve it. I wanted to understand why after installing the pg 9.6 packages suddenly psql tries to access the socket on /var/run/postgresql. Does the libpq default unix socket is changed between those two versions ? (9.6,9.2)
בתאריך יום ד׳, 9 בינו׳ 2019 ב-16:55 מאת Tom Lane <tgl@sss.pgh.pa.us>:
Mariel Cherkassky <mariel.cherkassky@gmail.com> writes: > I'm trying to understand some issues that I'm having with the unix_socket > settings and pgsql. > I have 2 machines with pg v9.2.5 with the same next settings : > #listen_addresses = 'localhost' > #unix_socket_directory = ''
This will result in the server creating the socket in whatever it thinks is the default socket directory. Traditionally PG uses /tmp as the default socket directory, and your netstat result is consistent with that:
> psql: could not connect to server: No such file or directory > Is the server running locally and accepting > connections on Unix domain socket > "/var/run/postgresql/.s.PGSQL.5432"?
shows that your psql is using a libpq that thinks the default socket directory is /var/run/postgresql. That's a build-time option, and I recall that Red Hat builds their postgresql package that way. I'm not 100% sure which way the PGDG RPMs do it.
You could override libpq's default, for instance via "psql -h /tmp". But probably you'd be better off removing any packages that provide libpq versions that don't match your server.
Alternatively, you could configure the server to create socket files in both places.