Re: Eliminating SPI from RI triggers - take 2

On Sun, Oct 2, 2022 at 10:24 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
> On 2022-10-01 18:21:15 -0700, Andres Freund wrote:
> > On 2022-09-29 18:18:10 +0900, Amit Langote wrote:
> > > So, here's a final revision for today. Sorry for the noise.
> >
> > This appears to fail on 32bit systems. Seems the new test is indeed
> > worthwhile...
> >
> > https://cirrus-ci.com/task/6581521615159296?logs=test_world_32#L406
> >
> > [19:12:24.452] Summary of Failures:
> > [19:12:24.452]
> > [19:12:24.452] 2/243 postgresql:main / main/regress FAIL 45.08s (exit status 1)
> > [19:12:24.452] 4/243 postgresql:pg_upgrade / pg_upgrade/002_pg_upgrade ERROR 71.96s
> > [19:12:24.452] 32/243 postgresql:recovery / recovery/027_stream_regress ERROR 45.84s
> >
> > Unfortunately ccf36ea2580f66abbc37f27d8c296861ffaad9bf seems to not have
> > suceeded in capture the test files of the 32bit build (and perhaps broke it
> > for 64bit builds as well?), so I can't see the regression.diffs contents.
>
> Oh, that appears to have been an issue on the CI side (*), while uploading the
> logs. The previous run did catch the error:
>
> diff -U3 /tmp/cirrus-ci-build/src/test/regress/expected/alter_table.out /tmp/cirrus-ci-build/build-32/testrun/main/regress/results/alter_table.out
> --- /tmp/cirrus-ci-build/src/test/regress/expected/alter_table.out 2022-09-30 15:05:49.930613669 +0000
> +++ /tmp/cirrus-ci-build/build-32/testrun/main/regress/results/alter_table.out 2022-09-30 15:11:21.050383258 +0000
> @@ -672,6 +672,8 @@
> ALTER TABLE FKTABLE ADD FOREIGN KEY(ftest1) references pktable;
> -- Check it actually works
> INSERT INTO FKTABLE VALUES(42); -- should succeed
> +ERROR: insert or update on table "fktable" violates foreign key constraint "fktable_ftest1_fkey"
> +DETAIL: Key (ftest1)=(42) is not present in table "pktable".
> INSERT INTO FKTABLE VALUES(43); -- should fail
> ERROR: insert or update on table "fktable" violates foreign key constraint "fktable_ftest1_fkey"
> DETAIL: Key (ftest1)=(43) is not present in table "pktable".

Thanks for the heads up. Hmm, this I am not sure how to reproduce on
my own, so I am currently left with second-guessing what may be going
wrong on 32 bit machines with whichever of the 4 patches.

For now, I'll just post 0001, which I am claiming has no semantic
changes (proof pending), to rule out that that one's responsible.

--
Thanks, Amit Langote
EDB: http://www.enterprisedb.com