psql's 001_basic.pl test could fail on very slow machines
Greetings, everyone!
I've gotten a very-very rare failure of psql's 001_basic.pl:
# +++ tap check in src/bin/psql +++
t/001_basic.pl ................. 79/?
# Failed test '\watch, 2 minimum rows: matches'
# at t/001_basic.pl line 380.
# ''
# doesn't match '(?^l:^123$)'
# Looks like you failed 1 test of 110.
t/001_basic.pl ................. Dubious, test returned 1 (wstat 256,
0x100)
Failed 1/110 subtests
The failed query itself:
psql_like(
$node,
sprintf(
q{with x as (
select now()-backend_start AS howlong
from pg_stat_activity
where pid = pg_backend_pid()
) select 123 from x where howlong < '2 seconds' \watch i=%g m=2},
0.5),
qr/^123$/,
'\watch, 2 minimum rows');
It seems that on very slow machines time between backend's start and
actual query execution can be more than two seconds
You can reproduce something similar with first attached patch,
manual_reproduction, using post_auth_delay
This can be reproduced on 17+ since this is when m[in_rows] was
introduced
Since this test checks that "\watch" clause "m[in_rows]" stops query
execution if amount of rows is less than the value of m,
we could simplify the test by replacing pg_stat_activity access
with plain select (see second attached patch, rewrite_test)
