database/postgres
database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-05-28 05:21:27 +03:00
Code

injection_points: Tweak more permutation in isolation test "basic"

Browse Source 
The CI has reported that using a marker to force the output of the
detach step to happen after the wait step was not enough, as
isolationtester has managed to report the detach step as waiting before
the wait step finishes in some runs.

src/test/isolation/README tells that there is a more drastic method to
enforce the ordering of the output: an empty step positioned just after
the wait step can force the wait step to complete before the detach step
begins.  This method has been able to pass 10 runs in the CI here, while
HEAD seems to fail 15~20% of the time in the CF bot.

Discussion: https://postgr.es/m/Z6WO8FbqK_FHmrzC@paquier.xyz
This commit is contained in:
Michael Paquier 2025-02-07 13:58:22 +09:00
parent 428fadb7e9
commit 926c7fce03
2 changed files with 6 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/test/modules/injection_points/expected/basic.out
View File

@ -1,6 +1,6 @@
Parsed test spec with 2 sessions
starting permutation: wait1 wakeup2 detach2
starting permutation: wait1 wakeup2 noop1 detach2
injection_points_attach
-----------------------
@ -19,6 +19,7 @@ injection_points_run
(1 row)
step noop1:
step detach2: SELECT injection_points_detach('injection-points-wait');
injection_points_detach
-----------------------

7
src/test/modules/injection_points/specs/basic.spec
View File

@ -20,6 +20,7 @@ setup {
SELECT injection_points_attach('injection-points-wait', 'wait');
}
step wait1 { SELECT injection_points_run('injection-points-wait'); }
step noop1 { }
session s2
step wakeup2 { SELECT injection_points_wakeup('injection-points-wait'); }
@ -27,9 +28,9 @@ step detach2 { SELECT injection_points_detach('injection-points-wait'); }
# Detach after wait and wakeup. Note that the detach may finish before
# the SQL function doing the wait returns its result. In order to avoid
# any ordering issues, the detach step reports its result only once the
# wait has completed. This is enforced with a parenthesized marker.
permutation wait1 wakeup2 detach2(wait1)
# any ordering issues, a no-op step is added after the wait, so as the
# detach is not launched until the wait has completed.
permutation wait1 wakeup2 noop1 detach2
# Detach before wakeup. s1 waits until wakeup, ignores the detach.
permutation wait1 detach2 wakeup2