troubles with forking program (not the usual probs though...)

[Markus Hoenicka <markus dot hoenicka at mhoenicka dot de>]

Hi all,

I've managed to screw up a forking program (refdbd, from  
http://refdb.sourceforge.net) in an attempt to install R which I need  
at work. This does not look like the widely reported fork failures as  
I do not get any diagnostic output - things just don't work as  
expected anymore. setup ran uneventfully, except that it failed to  
upgrade libatk the first time as it was probably in use, so I ran  
setup a second time after closing all programs. I've appended the log  
of both runs just in case.

refdbd had been working happily for years until an hour ago. I ran  
setup.exe to install R from cygwinports, following the well-known  
instructions, and accepted any updates that setup would suggest. Now  
there are a few seemingly unrelated ways that refdbd fails after the  
upgrade, therefore I have to apologize for the amount of information  
that follows. But maybe it rings a bell somewhere.

refdbd is a forking server and responds to queries sent from clients  
(e.g. refdbc). refdbd is usually started in "daemon mode", i.e. it  
detaches from its console. After I started it like this after the  
update, I got the following:

$ ./refdbd

$ refdbc -C whichdb
could not establish server connection

That is, there were no error messages (and no errors in refdbd's log  
file either) but the client couldn't talk to the server. Now I wanted  
to kill refdbd again:

$ ps ax
      PID    PPID    PGID     WINPID  TTY  UID    STIME COMMAND
     2296       1    2296       2296    ? 45021 18:50:55 /usr/bin/mintty
I    3112    2296    3112       3640    0 45021 18:50:55 /usr/bin/bash
     2968       1    2968       2968    ? 45021 18:51:00 /usr/bin/mintty
     2964    2968    2964       3692    1 45021 18:51:00 /usr/bin/bash
     4036       1    4036       4036    ? 45021 18:52:09  
/home/markus.hoenicka/prog/refdb/refdb/src/refdbd
      560    2964     560        600    1 45021 18:52:11 /usr/bin/ps

$ kill 4036
-bash: kill: (4036) - No such process

Otoh I can kill other processes, e.g. some less process, without problems.

I killed the refdbd process from Windows task manager. Then I  
restarted refdbd, this time having it run in the foreground:

$ ./refdbd -s

I was now able to talk to the server from a different console:

$ refdbc -C whichdb
[...]
999:1

As mentioned previously, refdbd is a forking server which forks one  
child per client request. The child is supposed to exit when done, and  
it did so for years until an hour ago. Now I see this:

$ ps ax
      PID    PPID    PGID     WINPID  TTY  UID    STIME COMMAND
     2296       1    2296       2296    ? 45021 18:50:55 /usr/bin/mintty
     3112    2296    3112       3640    0 45021 18:50:55 /usr/bin/bash
     2968       1    2968       2968    ? 45021 18:51:00 /usr/bin/mintty
     2964    2968    2964       3692    1 45021 18:51:00 /usr/bin/bash
     3852    3112    3852       3592    0 45021 18:53:27  
/home/markus.hoenicka/prog/refdb/refdb/src/refdbd
     3132    3852    3852       3132    0 45021 18:53:46  
/home/markus.hoenicka/prog/refdb/refdb/src/refdbd
     3404    2964    3404       2928    1 45021 18:53:58 /usr/bin/ps

That is, the child is still around. Any subsequent client calls create  
more of these zombies. Again, kill(1) maintains that these processes  
do not exist.

So, to sum it up: (1) detaching refdbd from the console fails (2)  
killing the process fails although it shows up in ps (3) child  
processes do not terminate but pile up.

I've rebooted the box, ran rebaseall to exclude simple glitches, and I  
even rebuilt refdbd from the sources but the problem persists.

Any clues? cygcheck output is attached too.


regards,
Markus


--
Markus Hoenicka
http://www.mhoenicka.de
AQ score 38

Attachment: cygcheck_output.txt
Description: cygcheck_output.txt

Attachment: updated_packages.txt
Description: updated_packages.txt

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple