Hello Nazir and Michael!
01.07.2025 10:57, Nazir Bilal Yavuz wrote:
I agree with you. So, the current logic is:
If primary is not alive: Do not report anything.
If only primary is alive: Report the entire diff file.
If both primary and standby are alive: Report entire diff file and add
head+tail of diff to the failure message.
Done like above in v2.
The new check has failed on mamba [1], apparently because this animal is
too slow for pg_isready:
regress_log_027_stream_regress:
...
# Running: pg_isready --host /home/buildfarm/bf-data/tmp/ep8AOH4m7l --port 24781
/home/buildfarm/bf-data/tmp/ep8AOH4m7l:24781 - no response
[08:01:50.899](1505.313s) ok 2 - regression tests pass
[08:01:50.902](0.003s) ok 3 - primary alive after regression test run
[08:01:50.905](0.003s) not ok 4 - standby alive after regression test run
[08:01:50.908](0.003s)
[08:01:50.908](0.000s) # Failed test 'standby alive after regression test run'
# at t/027_stream_regress.pl line 104.
[08:01:50.909](0.001s) # got: '0'
# expected: '1'
027_stream_regress_standby_1.log:
2025-07-28 07:37:27.237 EDT [22920:1] LOG: starting PostgreSQL 19devel on powerpc-unknown-netbsd10.1, compiled by cc
(nb3 20231008) 10.5.0, 32-bit
2025-07-28 07:37:27.239 EDT [22920:2] LOG: listening on Unix socket
"/home/buildfarm/bf-data/tmp/ep8AOH4m7l/.s.PGSQL.24781"
2025-07-28 07:37:27.281 EDT [25395:1] LOG: database system was interrupted;
last known up at 2025-07-28 07:36:48 EDT
2025-07-28 07:37:27.282 EDT [25395:2] LOG: starting backup recovery with redo LSN 0/02000028, checkpoint LSN
0/02000080, on timeline ID 1
2025-07-28 07:37:27.283 EDT [25395:3] LOG: entering standby mode
2025-07-28 07:37:27.287 EDT [25395:4] LOG: redo starts at 0/02000028
...
2025-07-28 08:01:47.884 EDT [6985:1] [unknown] LOG: connection received:
host=[local]
2025-07-28 08:01:48.261 EDT [6985:2] [unknown] LOG: connection authenticated: user="buildfarm" method=trust
(/home/buildfarm/bf-data/HEAD/pgsql.build/src/test/recovery/tmp_check/t_027_stream_regress_standby_1_data/pgdata/pg_hba.conf:117)
2025-07-28 08:01:48.261 EDT [6985:3] [unknown] LOG: connection authorized: user=buildfarm database=postgres
application_name=027_stream_regress.pl
2025-07-28 08:01:51.552 EDT [6985:4] 027_stream_regress.pl LOG: could not send
data to client: Broken pipe
### 3 seconds is the pg_isready's default timeout
2025-07-28 08:01:51.552 EDT [6985:5] 027_stream_regress.pl FATAL: connection to
client lost
2025-07-28 08:01:51.552 EDT [6985:6] 027_stream_regress.pl LOG: disconnection: session time: 0:00:03.670 user=buildfarm
database=postgres host=[local]
...
What do you think of increasing the timeout (e.g. , to 10 seconds)?
[1]
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=mamba&dt=2025-07-28%2007%3A46%3A26
Best regards,
Alexander
