Horton

Re: Extracting cross-version-upgrade knowledge from buildfarm client

Alvaro Herrera <alvherre@alvh.no-ip.org>

View thread View on postgr.es
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-hackers@lists.postgresql.org, Andrew Dunstan <andrew@dunslane.net>
Date: 2023-07-19T11:05:04Z
Lists: pgsql-hackers 
I just hit a snag testing this.  It turns out that the
PostgreSQL::Version comparison stuff believes that 16beta2 < 16, which
sounds reasonable.  However, because of that, the AdjustUpgrade.pm
stanza that tries to drop tables public.gtest_normal_child{2} in
versions earlier than 16 fails, because by 16 these tables are dropped
in the test itself rather than left to linger, as was the case in
versions 15 and earlier.

So, if you try to run the pg_upgrade test with a dump created by
16beta2, it will fail to drop these tables (because they don't exist)
and the whole test fails.  Why hasn't the buildfarm detected this
problem?  I see that Drongo is happy, but I don't understand why.
Apparently, the AdjustUpgrade.pm stuff leaves no trace.

I can fix this either by using DROP IF EXISTS in that stanza, or by
making AdjustUpgrade use 'version <= 15'.  Any opinions on which to
prefer?


(Well, except that the tests added by c66a7d75e65 a few days ago fail
for some different reason -- the tests want pg_upgrade to fail, but it
doesn't fail for me.)

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
<Schwern> It does it in a really, really complicated way
<crab> why does it need to be complicated?
<Schwern> Because it's MakeMaker.

Commits

  1. Compare only major versions in AdjustUpgrade.pm

  2. Get rid of the "new" and "old" entries in a view's rangetable.

  3. Create common infrastructure for cross-version upgrade testing.