Re: 002_pg_upgrade is broken for custom install

On Fri, Apr 4, 2025 at 5:23 PM Álvaro Herrera <alvherre@alvh.no-ip.org> wrote:
>
> On 2025-Apr-04, Ashutosh Bapat wrote:
>
> > connection error: 'psql: error: connection to server on socket
> > "/tmp/LiPa_UJpSb/.s.PGSQL.13779" failed: FATAL: database "regression"
> > does not exist'
> > while running '/home/ashutosh/work/units/pghead/build/dev/bin/psql
> > --no-psqlrc --no-align --tuples-only --quiet --dbname port=13779
> > host=/tmp/LiPa_UJpSb dbname='regression' --file - --variable
> > ON_ERROR_STOP=1' at
> > /home/ashutosh/work/units/pghead/coderoot/pg/src/test/perl/PostgreSQL/Test/Cluster.pm
> > line 2256.
> > # Postmaster PID for node "old_node" is 779230
> > ### Stopping node "old_node" using mode immediate
>
> This is saying that Cluster->psql() (line 2256) tried to connect to
> database regression and failed unexpectedly.  Is your "olddump" file
> created with pg_dumpall from a cluster that contains such a database?
> If not, then the test isn't broken, you're just not operating it
> correctly :-)

My bad, didn't pay attention to that error (even while pasting it) :(.
You are right. I used pg_dump (without --create) to create olddump.
The comment about olddump just mentions dump output. I think we should
mention pg_dumpall there.

>
> Maybe the failure could be clearer: rather than blindly trying to
> connect to regression, first see if it exists, and die with a hard
> failure ("your old dump must contain database regression") if not.
> Alternatively, if the regression database doesn't exist, just skip that
> part.

Right. Something like attached?

--
Best Wishes,
Ashutosh Bapat