Hello hackers,
A recent desman failure [1] with the following diagnostics:
# parallel group (2 tests): subscription publication
not ok 157 + publication 2251 ms
ok 158 + subscription 415 ms
--- /home/fedora/17-desman/buildroot/REL_16_STABLE/pgsql.build/src/test/regress/expected/publication.out 2024-12-09
18:34:02.939762233 +0000
+++ /home/fedora/17-desman/buildroot/REL_16_STABLE/pgsql.build/src/test/regress/results/publication.out 2024-12-09
18:44:48.582958859 +0000
@@ -1193,23 +1193,29 @@
ERROR: permission denied for database regression
SET ROLE regress_publication_user;
GRANT CREATE ON DATABASE regression TO regress_publication_user2;
+ERROR: tuple concurrently updated
SET ROLE regress_publication_user2;
SET client_min_messages = 'ERROR';
CREATE PUBLICATION testpub2; -- ok
+ERROR: permission denied for database regression
and postmaster.log containing:
2024-12-09 18:44:46.753 UTC [1345157:903] pg_regress/publication STATEMENT:
CREATE PUBLICATION testpub2;
2024-12-09 18:44:46.753 UTC [1345158:287] pg_regress/subscription LOG: statement: REVOKE CREATE ON DATABASE REGRESSION
FROM regress_subscription_user3;
2024-12-09 18:44:46.754 UTC [1345157:904] pg_regress/publication LOG:
statement: SET ROLE regress_publication_user;
2024-12-09 18:44:46.754 UTC [1345157:905] pg_regress/publication LOG: statement: GRANT CREATE ON DATABASE regression TO
regress_publication_user2;
2024-12-09 18:44:46.754 UTC [1345157:906] pg_regress/publication ERROR: tuple
concurrently updated
2024-12-09 18:44:46.754 UTC [1345157:907] pg_regress/publication STATEMENT: GRANT CREATE ON DATABASE regression TO
regress_publication_user2;
shows that the subscription and publication tests are not concurrent-safe,
because modifying the same pg_database entry might fail with the "tuple
concurrently updated" error.
I've managed to reproduce the error with:
sed -E "s/(REVOKE CREATE ON DATABASE REGRESSION FROM
regress_subscription_user3;)/$(printf '\\1%.0s' {1..2000})/" -i.bak \
src/test/regress/sql/subscription.sql
src/test/regress/expected/subscription.out
sed -E "s/(GRANT CREATE ON DATABASE regression TO
regress_publication_user2;)/$(printf '\\1%.0s' {1..1000})/" -i.bak \
src/test/regress/sql/publication.sql src/test/regress/expected/publication.out
sed -E "s/(test: publication subscription$)/$(printf '\\1\\n%.0s' {1..10})/"
-i.bak src/test/regress/parallel_schedule
This makes `make check` fail like below:
# parallel group (2 tests): subscription publication
ok 170 + publication 202 ms
ok 171 + subscription 100 ms
# parallel group (2 tests): subscription publication
ok 172 + publication 198 ms
not ok 173 + subscription 107 ms
# parallel group (2 tests): subscription publication
ok 174 + publication 204 ms
ok 175 + subscription 100 ms
src/test/regress/regression.diffs contains:
+ERROR: tuple concurrently updated
This issue is reproduced starting from commit c3afe8cf5 (dated 2023-03-30),
which added "REVOKE CREATE ON DATABASE REGRESSION ..." into the subscription
test.
[1]
https://buildfarm.postgresql.org/cgi-bin/show_stage_log.pl?nm=desman&dt=2024-12-09%2018%3A33%3A49&stg=check
Best regards,
Alexander
