Yes but that would have involved changing hundreds of CGI scripts. Dropping
the private tmp was easier.
On 11/10/14 07:01, Jim Nasby wrote:
FWIW, you could also use an IP connection to Postgres instead of the local
socket.
On 10/8/14, 6:34 PM, Stephen Davies wrote:
This is not the same issue.
However, I had already disabled SELinux for other reasons.
The actual cause of my issue was the "new" private tmp facility in systemd
startup of httpd. This makes the PostgreSQL socket invisible to CGI scripts.
We have survived for many years without this before migrating to CentOS 7 so
I simply disabled this too and all came good.
Cheers and thanks,
Stephen
On 08/10/14 23:49, Adrian Klaver wrote:
On 10/07/2014 09:10 PM, Stephen Davies wrote:
The permissions on the socket are 777 owner/group postgres.
I installed the 9.3 onto the Centos 7 server using the repo at
postgresql.org.
(http://yum.postgresql.org/9.3/redhat/rhel-$releasever-$basearch)
There is no /var/run/postgresql and find cannot find another socket
anywhere else.
Sounds similar to this:
Long version:
http://serverfault.com/questions/609947/database-connection-to-postgresql-refused-for-flask-app-under-mod-wsgi-when-start
Short version:
Disable SELinux
Cheers and thanks,
Stephen
On 08/10/14 14:32, Tom Lane wrote:
Stephen Davies <sdavies@xxxxxxxxxx> writes:
I am in the process of migrating a bunch of databases and associated CGI
scripts from 9.1.4 to 9.3 (and from 32-bit to 64-bit).
The database migration has been successful but I have an issue with psql
connections from CGI scripts.
I can connect to the 9.3 server locally with psql from the command
line, with
psql from other boxes on the LAN via TCP, via JDBC from programs and
servlets
but cannot connect locally via CGI.
If I run any of the CGI scripts from the command line they work but when
invoked by Apache, they fail with the usual question as to whether
anything is
listening on socket /tmp/.s.PGSQL.5432.
Some Linux variants think it improves security to run daemons like apache
in a context where what the daemon sees as /tmp has been mapped somewhere
else.
If you're running one of these platforms, the Postgres server and libpq
distributed by the vendor will have been hacked to cope, typically by
agreeing that the socket location is something like /var/run/postgresql/
rather than /tmp. I'm guessing your 9.3 installation was self-built
and hasn't been configured that way.
regards, tom lane
--
=============================================================================
Stephen Davies Consulting P/L Phone: 08-8177 1595
Adelaide, South Australia. Mobile:040 304 0583
Records & Collections Management.
--
Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general
