Forking raises a race condition, where the parent might run the
test before the child has had time to fork. If that happens,
an error similar to this is produced:
Failed test 'Output correct'
at ./t/check_procs.t line 32.
'PROCS OK: 0 processes with args 'sleep 7' | processes=0;;;0;'
doesn't match '/^PROCS OK: 1 process?/'
Sleeping a bit should avoid the problem. It might be enough to
sleep less than a second, but perl's built-in sleep function only
supports integer seconds.
In our build environment, the build failed 3 of 4 times before
this patch. After the patch it failed 0 of 7 times.
Signed-off-by: Mikael Falkvidd <mfalkvidd@op5.com>
This fixes an issue that appears when running check_procs over NRPE,
where the default shell is configured to (for example) dash, as is the
case on Debian.
dash (and tcsh, and mksh, and probably others), when invoked with -c forks an additional process
to execute the argument string. Contrast this with bash, which does not
do this, provided that the argument string simply can be exec()'d as-is.
To demonstrate:
$ bash -c pstree
init─┬ ..
...
├─sshd─-─sshd───pstree
versus
$ dash -c pstree
init─┬ ..
...
├─sshd─-─sshd───dash───pstree
The consequence of this fork is that the following invocation:
/opt/plugins/check_procs -a init
will result in this output:
PROCS OK: 2 processes with args 'init' | processes=2;;;0;
because the check_procs, in addition to finding the actual init process,
finds its parent shell as well.
This example is a bit contrived, but I think it illustrates the
point.
This wouldn't really be a problem, and normally isn't, if it weren't
for the fact that NRPE uses a call to popen() which does exactly the
above (executes '/bin/sh -c ...'), causing inconsistent behaviour
between distributions and much confusion for end users.
The argument may be made that the dash process spawned by NRPE is just a
process like any other, and should therefore be included in the process
count just like any other. However, this is not very intuitive, because
of the previously mentioned inconsistencies.
The argument might also well be made that we're _never_ interested in the
immediate ancestor of the plugin, and while it is unknown how many
installations have already made the necessary modifications to their
setups to make up for the fact that the plugin behaves the way it does,
it is not deemed worthwhile to entertain such workarounds.
Thus, this patch ignores the parent process.
See also these bug reports:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=626913http://sourceforge.net/p/nagiosplug/bugs/512/https://github.com/nagios-plugins/nagios-plugins/issues/999https://bugs.op5.com/view.php?id=4398
