[DRE-maint] Bug#901379: gitlab: postinst script should not call psql if database is not managed by dbconfig-common

[Pirate Praveen]

On Fri, 11 Dec 2020 23:53:27 +0100 <devel at sumpfralle.de> wrote:
 > Hello,
 >
 > I encounter the same problem (psql directly used in postinst script) 
during
 > every package upgrade.
 >
 > The main problem (in my case) is the misdetection of an 
uninitialized database
 > (see "db_relations" in /usr/lib/gitlab/scripts/rake-tasks.sh).
 > The script checks via "psql" (locally), whether the database is 
already
 > populated. Since the database is empty, the initial database setup 
is started
 > (this time using rake for interacting with the configured external 
database).
 > This initialization fails, since the real database is not empty.
 > Thus postinst fails.
 >
 > My current workaround is to run the following loop in a separate 
shell:
 >
 >  while sleep 1; do sed -i 's/db_relations"/db_relationsXXX"/' \
 >    /usr/lib/gitlab/scripts/rake-tasks.sh; done
 >
 > This overrides the test for an empty database.
 > Another approach could be to simply populate the local (unused) 
database.
 >
 > The proper solution would be to use the *configured* database 
connection
 > instead of running "psql" directly. Maybe this could be done via 
rake?
 >
 > If you could give me any hints in this direction, then I could 
propose a patch
 > for fixing the database accesses in the postinst other related 
scripts.

We can read dbc_dbserver value from /etc/dbconfig-common/gitlab.conf 
and use psql -h $dbc_dbserver to check remotely.