Re: pgsql: Fix restore of not-null constraints with inheritance

On 2024-Apr-18, Alvaro Herrera wrote:

On 2024-Apr-18, Alvaro Herrera wrote:

Lastly, make two changes to pg_dump: 1) do not try to drop a not-null
constraint that's marked as inherited; this allows a dump to restore
with no errors if a table with a PK inherits from another which also has
a PK; 2) avoid giving inherited constraints throwaway names, for the
rare cases where such a constraint survives after the restore.

Hmm, this last bit broke pg_upgrade on crake.  I'll revert this part,
meanwhile I'll be installing 9.2 to see if it can be fixed in a better way.

Eh, so:

1. running "make check" on pg_upgrade using an oldinstall pointing to
9.2 fails, because PostgreSQL::Test::Cluster doesn't support that
version -- it only goes back to 9.2.  How difficult was it to port it
back to all these old versions?

It's not that hard to make it go back to 9.2. Here's a version that's a couple of years old, but it supports versions all the way back to 7.2 :-)

If there's interest I'll work on supporting our official "old" versions (i.e. 9.2 and up)

2. running "make check" with an oldinstall pointing to 10 fails, because
the "invalid database" checks fail:

not ok 7 - invalid database causes failure status (got 0 vs expected 1)

#   Failed test 'invalid database causes failure status (got 0 vs expected 1)'
#   at t/002_pg_upgrade.pl line 407.
not ok 8 - invalid database causes failure stdout /(?^:invalid)/

3. Lastly, even if I put back the code that causes the failures on crake
and restore from 10 (and ignore those two problems), I cannot reproduce
the issues it reported.  Is crake running some funky code that's not
what "make check" on pg_upgrade does, perchance?

It's running the buildfarm cross version upgrade module. See <https://github.com/PGBuildFarm/client-code/blob/main/PGBuild/Modules/TestUpgradeXversion.pm>

It's choking on the change in constraint names between the dump of the pre-upgrade database and the dump of the post-upgrade database, e.g.

 CREATE TABLE public.rule_and_refint_t2 (
-    id2a integer CONSTRAINT pgdump_throwaway_notnull_0 NOT NULL NO INHERIT,
-    id2c integer CONSTRAINT pgdump_throwaway_notnull_1 NOT NULL NO INHERIT
+    id2a integer CONSTRAINT rule_and_refint_t2_id2a_not_null NOT NULL NO INHERIT,
+    id2c integer CONSTRAINT rule_and_refint_t2_id2c_not_null NOT NULL NO INHERIT );

look at the dumpdiff-REL9_2_STABLE file for the full list.

I assume that means pg_dump is generating names that pg_upgrade throws away? That seems ... unfortunate.

There is a perl module at src/test/perl/PostgreSQL/Test/AdjustUpgrade.pm. This is used to adjust the dump files before we diff them. Maybe you can remedy the problem by adding some code in there.

cheers

andrew

--
Andrew Dunstan
EDB: https://www.enterprisedb.com