> Dnia wtorek, 21 lutego 2012 22:10:59 Riccardo Magliocchetti pisze: >> Il 21/02/2012 22:04, Łukasz Mierzwa ha scritto: >> > Dnia wtorek, 21 lutego 2012 21:56:24 Riccardo Magliocchetti pisze: >> >> Lukasz, >> >> >> >> Il 21/02/2012 21:50, Łukasz Mierzwa ha scritto: >> >>> Dnia wtorek, 21 lutego 2012 20:29:54 Łukasz Mierzwa pisze: >> >>>> Program received signal SIGFPE, Arithmetic exception. >> >>>> 0x000000000043a1eb in carbon_master_cycle () >> >>>> (gdb) bt >> >>>> #0 0x000000000043a1eb in carbon_master_cycle () >> >>>> #1 0x000000000041c61a in master_loop () >> >>>> #2 0x000000000042f190 in uwsgi_start () >> >>>> #3 0x00007fd06906670d in clone () at >> >>>> ../sysdeps/unix/sysv/linux/x86_64/clone.S:112 >> >>>> #4 0x0000000001b6cde0 in ?? () >> >>>> #5 0xae23d9d52ba2b100 in ?? () >> >>>> #6 0x000000000000000c in ?? () >> >>>> #7 0x00000000006595e8 in uwsgi () >> >>>> #8 0x0000000000000000 in ?? () >> >>>> >> >>>> I'll try to install more dbg packages. >> >>> >> >>> I can't get anything more, I installed debug packages for most of >> the >> >>> libs >> >>> used (I can' find all of them). But it turns out that this is not >> just >> >>> php, >> >>> python apps are also afected so this is something in master code. >> >>> Maybe new cheaper implementation doesn't like idle option? >> >>> I didn't had this issue before but I didn't had idle enabled. I'll >> look >> >>> at >> >>> 1.0.4 tommorow. >> >> >> >> The stackstrace suggests it's a bug in carbon plugin, you may want to >> >> rebuild uwsgi with debug set to true in the config file and post a >> `bt >> >> full` to help Roberto debug the issue. >> >> >> >> hth, >> >> riccardo >> > >> > Thanks, I have no idea about debugging C code, #0 is the function >> that >> > had >> > exception? I should read that bottom-up? >> >> Yes, #0 is the last stack frame, beware it may not be always the >> culprit. See the last mail i sent for a revised patch to try. >> >> riccardo > > Thanks, I tried to read it like python backtrace and I thought that the > error > is at the bottom so I kept trying ot decrypt "?? ( )" into something > readable. > I guess that looking at the 0x0000000000000000 address should ring a bell > that > I might do it wrong ;) > Anyway it seems to help, but I will be sure after few hours. Of course I > should be the one to blame since the buggy code code was from my patch > that > went into the carbon plugin. > >
I think i could map uwsgi_backtrace() to SIGFPE, this should help addressing similar problems in the future. Regarding carbon plugin, simply add a check in each division you have in the code, for zeroed values. By the way, it looks like Riccardo's patch is just enough -- Roberto De Ioris http://unbit.it _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
