According to
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=prairiedog&dt=2014-07-21%2022%3A36%3A55
prairiedog saw a crash in "make check" on the 9.4 branch earlier tonight;
but there's not a lot of evidence as to why in the buildfarm report,
because the postmaster log file is truncated well before where things got
interesting. Fortunately, I was able to capture a copy of check.log
before it got overwritten by the next run. I find that the place where
the webserver report stops matches this section of check.log:
[53cd99bb.134a:158] LOG: statement: create index test_range_gist_idx on test_range_gist using gist (ir);
[53cd99bb.134a:159] LOG: statement: insert into test_range_gist select int4range(g, g+10) from generate_series(1,2000)
g;
^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^\
@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@^@[53cd99ba.1344:329] LOG: statement: INSERT INTO num_exp_div VALUES
(7,8,'-1108.80577182462841041118');
[53cd99ba.1344:330] LOG: statement: INSERT INTO num_exp_add VALUES (7,9,'-107955289.045047420');
[53cd99ba.1344:331] LOG: statement: INSERT INTO num_exp_sub VALUES (7,9,'-58101680.954952580');
The ^@'s represent nul bytes, which I find runs of elsewhere in the file
as well. I think they are an artifact of OS X buffering policy caused by
multiple processes writing into the same file without any interlocks.
Perhaps we ought to consider making buildfarm runs use the logging
collector by default? But in any case, it seems uncool that either the
buildfarm log-upload process, or the buildfarm web server, is unable to
cope with log files containing nul bytes.
Anyway, to cut to the chase, the crash seems to be from this:
TRAP: FailedAssertion("!(FastPathStrongRelationLocks->count[fasthashcode] > 0)", File: "lock.c", Line: 2957)
which the postmaster shortly later says this about:
[53cd99b6.130e:2] LOG: server process (PID 5230) was terminated by signal 6: Abort trap
[53cd99b6.130e:3] DETAIL: Failed process was running: ROLLBACK PREPARED 'foo1';
[53cd99b6.130e:4] LOG: terminating any other active server processes
So there is still something rotten in the fastpath lock logic.
regards, tom lane