On 11/27/18 1:32 AM, Andres Freund wrote:
The cross-version bug around pg_largeobject_metadata ought to be fixed
now. I'm hoping for a response by Andrew about the ordering bit, you're
welcome to opine on that too.
I see that prairiedog
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=prairiedog&dt=2018-11-27%2005%3A30%3A39
failed just it's last run, but I'm not clear whether that's just a
random failure, or something I can't don't recognize without further
information.
The report is that the dumps differ. My bet is that this is another
instance of the LO ordering issue. I can handle that in the
cross-version tests because they are completely under my control (see
<https://github.com/PGBuildFarm/client-code/commit/9ca42ac1783a8cf99c73b4f7c52bd05a6024669d>)
, but not in the builtin test, which has this:
if diff "$temp_root"/dump1.sql "$temp_root"/dump2.sql >/dev/null; then
echo PASSED
exit 0
else
echo "Files $temp_root/dump1.sql and $temp_root/dump2.sql differ"
echo "dumps were not identical"
exit 1
fi
So that's all we know. Maybe we should change the script to output the
diff, or at least some of it, say the first 1000 lines.
In any case, I think we will probably need to bite the bullet and have
pg_dump render LOs in OID order. Not sure if that needs to be always on,
or an option.
cheers
andrew
--
Andrew Dunstan https://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services
