cfengine process restart hang, runscript, and open files.

[Demosthenex]

I'm experimenting with using cfengine under Gentoo and am having
problems when restarting processes with the scripts in /etc/init.d.

cfengine can check for specific processes under varying criteria on a
given schedule, and among other things, restart that process using a
restart command if the process is missing.

The problem that occurs is when cfengine uses a script in /etc/init.d
to restart a failed/missing process, cfengine hangs. This is due to a
known issue with bad daemons leaving file descriptors open.

Here's the relevant portion of my config:

processes:
"/usr/sbin/syslog-ng"
restart "/etc/init.d/syslog-ng zap start"
matches=1
useshell=false
inform=true

The intent is that if syslog-ng isn't running, cfengine will restart
it using the restart command. Upon its scheduled execution cfengine
does detect the lack of syslog-ng, and calls the restart command. At
that point, cfengine hangs, waiting for the restart command to finish
and return.

This is very similar to when you start certain programs in the
background over an ssh connection, and though you try to logout, ssh
is still waiting for something.

So, I've varied the restart command. It has worked successfully
calling the syslog-ng binary directly, and using the line from
/etc/init.d/syslog-ng, that uses start-stop-daemon to launch
syslog-ng.

It seems the culprit is in /sbin/runscript, which is the interpreter
for all the /etc/init.d/* scripts. I believe if it would simply close
all open files before exiting, this would work cleanly. In the mean
time, all scripts in /etc/init.d are broken like this.

Here is a link to a relevant article on the same problem.

http://www.openldap.org/lists/openldap-devel/200001/msg00014.html

Comments and suggestions are welcome.

Demo

[delta407]

[Demosthenex]

[wdsl]

Hi,
can your solution cause any Problem?
It works fine on our test stystem but i'm a little bit afraid that this cause any not tested problems.

thx
wdsl