I guess the os.wait() throws an OSError with errno = EINTR because the SIGALARM fires? If that is the case we should probably ignore it in use_fork.py
On Saturday, April 21, 2012 3:36:44 PM UTC-4, Stephen Montgomery-Smith wrote: > > I did some research, and I found that you are completely correct. I > don't know why FreeBSD is misbehaving, because the os.wait command > definitely triggers the OSError exception. signal.getsignal(SIGCHLD) > shows that it is correctly set to signal.SIG_DFL. So I am mystified. > > Thank you for setting me straight. > > > On 04/21/2012 08:58 AM, Volker Braun wrote: > > I don't have a FreeBSD box to try this on, but note that children > > shouldn't just disappear when they finish. They remain as zombies > > (<defunct> in ps) until they are collected by os.wait() > > > > > > > > On Saturday, April 21, 2012 12:39:13 AM UTC-4, Stephen Montgomery-Smith > > wrote: > > > > In parallel/use_fork.py, around line 120, there seems to be a race > > problem. This seems to be a genuine problem with my FreeBSD build: > > sage -t -force_lib "devel/sage/sage/parallel/decorate.py" > > goes into a permanent loop. > > > > Suppose that timeout has been set, and the forked process is too > slow. > > So then SIGALRM is sent by the alarm before the "os.wait()[0]" is > > completed. But now suppose that after the alarm has triggered, the > > forked process manages to finish before getting to the > "os.kill(pid,9)" > > command. Then "len(workers)" is still positive, so the while loop > > continues. But now the "os.wait()[0]" command has nothing to wait for > > (since it should be waiting for "pid" to die), and everything ends up > > stalling. > > > > Also if you do an os.kill command on a non-existent pid, you get an > > OSError. > > > > I really don't understand the full logic of this code, otherwise I > > would > > give you guys a patch. > > > > -- > > To post to this group, send an email to sage-devel@googlegroups.com > > To unsubscribe from this group, send an email to > > sage-devel+unsubscr...@googlegroups.com > > For more options, visit this group at > > http://groups.google.com/group/sage-devel > > URL: http://www.sagemath.org > > -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
