Re: [HTCondor-devel] 8.9.8 master getting into an infinite loop on startup

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

Also see my code review question on this at

https://htcondor-wiki.cs.wisc.edu/index.cgi/tktview?tn=7650

regards,
Todd

p.s. people should ideally not do a code review on their own code :)


On 7/21/2020 2:22 PM, Bockelman, Brian via HTCondor-devel wrote:
> Hi,
>
> Wasnât Jaime looking at this in terms of using /proc/self/fd to determine the highest value?
>
> Brian
>
> Sent from my iPhone
>
> > On Jul 21, 2020, at 2:19 PM, MÃtyÃs Selmeci <matyas@xxxxxxxxxxx> wrote:
> >
> > ïBe careful with that -- I was seeing slow startup times and general lack of responsiveness at 8 million.
> > 2 million seems fine.
> >
> > -Mat
> >
> >
> > > On 7/21/20 1:12 PM, Tim Theisen wrote:
> > > Thank you for figuring that out. I guess I just include a big number
> > > rather than infinity in the systemd .service file.
> > >
> > > Infinity should have worked everywhere.
> > >
> > > ...Tim
> > >
> > > > On 7/21/20 1:05 PM, MÃtyÃs Selmeci via HTCondor-devel wrote:
> > > > If I run it by hand instead of via systemd, everything works fine.
> > > >
> > > > If I edit the .service file and change LimitNOFILE=infinity to LimitNOFILE=524288, everything works fine.
> > > >
> > > > Interestingly, in a root shell, I get "operation not permitted" when trying to do `ulimit -n unlimited`.
> > > > The highest value I can set it to is 1073741816, which is the value of the fs.nr_open sysctl.
> > > >
> > > > -Mat
> > > >
> > > >
> > > > On 7/21/20 12:45 PM, Bockelman, Brian wrote:
> > > > > Ohh - is that in the middle of "close() all FDs possible" code?
> > > > >
> > > > > Does "strace" show a lot of close() followed by EBADF?  What's the process limit on FDs?
> > > > >
> > > > > Brian
> > > > >
> > > > > > On Jul 21, 2020, at 12:41 PM, MÃtyÃs Selmeci <matyas@xxxxxxxxxxx> wrote:
> > > > > >
> > > > > > Here's the pstack of the child:
> > > > > >
> > > > > >
> > > > > > #0  0x00007fd5c788fa17 in close () from /usr/lib64/libpthread.so.0
> > > > > > #1  0x00007fd5c83eb03e in CreateProcessForkit::exec() () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #2  0x00007fd5c83eb89c in CreateProcessForkit::fork_exec() () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #3  0x00007fd5c83f85cb in DaemonCore::Create_Process(char const*, ArgList const&, priv_state, int, int, int, Env const*, char const*, FamilyInfo*, Stream**, int*, int*, int, __sigset_t*, int, unsigned long*, int*, char const*, MyString*, FilesystemRemap*, long) () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #4  0x00007fd5c82da64b in ProcFamilyProxy::start_procd() () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #5  0x00007fd5c82db283 in ProcFamilyProxy::ProcFamilyProxy(char const*) () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #6  0x00007fd5c82d9e18 in ProcFamilyInterface::create(char const*) () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #7  0x00007fd5c83f9236 in DaemonCore::Create_Process(char const*, ArgList const&, priv_state, int, int, int, Env const*, char const*, FamilyInfo*, Stream**, int*, int*, int, __sigset_t*, int, unsigned long*, int*, char const*, MyString*, FilesystemRemap*, long) () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #8  0x0000000000416315 in daemon::RealStart() ()
> > > > > > #9  0x0000000000416f3a in Daemons::StartDaemonHere(daemon*) ()
> > > > > > #10 0x0000000000416fe3 in Daemons::StartAllDaemons() ()
> > > > > > #11 0x000000000040ebbe in main_init(int, char**) ()
> > > > > > #12 0x00007fd5c8403468 in dc_main(int, char**) () from /usr/lib64/libcondor_utils_8_9_8.so
> > > > > > #13 0x00007fd5c76d9042 in __libc_start_main () from /usr/lib64/libc.so.6
> > > > > > #14 0x000000000040b90e in _start ()
> > > > > >
> > > > > >
> > > > > > On 7/21/20 12:34 PM, Bockelman, Brian wrote:
> > > > > > > Hi Mat,
> > > > > > >
> > > > > > > Could you do a "pstack" of the child condor_master process?
> > > > > > >
> > > > > > > Unfortunately, from your traceback, it like the master is simply waiting for the child to do something (either exec or error out) -- not too much info there.
> > > > > > >
> > > > > > > Brian
> > > > > > >
> > > > > > > > On Jul 21, 2020, at 12:05 PM, MÃtyÃs Selmeci via HTCondor-devel <htcondor-devel@xxxxxxxxxxx> wrote:
> > > > > > > >
> > > > > > > > Hey folks,
> > > > > > > >
> > > > > > > > I've got a problem running 8.9.8 on my Fedora 32 laptop (I'm using an
> > > > > > > > RPM Tim gave me from an NMI build): when I start condor, the master
> > > > > > > > forks and the child master gets into an infinite loop, eating an entire
> > > > > > > > CPU and not responding to SIGTERM.  The last line in the MasterLog is:
> > > > > > > >
> > > > > > > > 07/21/20 11:46:56 (fd:1) (pid:233863) (D_DAEMONCORE) About to exec "/usr/sbin/condor_procd"
> > > > > > > >
> > > > > > > > SELinux is off.  I attached my MasterLog with D_ALL:2 and
> > > > > > > > condor_config_val -summary (that feature's great).  The traceback
> > > > > > > > at the end of MasterLog is me killing sending SIGABRT to both
> > > > > > > > condor_master processes.
> > > > > > > >
> > > > > > > > Any ideas?
> > > > > > > >
> > > > > > > > Thanks,
> > > > > > > > -Mat
> > > > > > > > <MasterLog.txt><summary.txt>_______________________________________________
> _______________________________________________
> HTCondor-devel mailing list
> HTCondor-devel@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/htcondor-devel

--
Todd Tannenbaum <tannenba@xxxxxxxxxxx> University of Wisconsin-Madison
Center for High Throughput Computing   Department of Computer Sciences
HTCondor Technical Lead                1210 W. Dayton St. Rm #4257
Phone: (608) 263-7132                  Madison, WI 53706-1685