#171 uxmon reads part of uxmon-net twice on Solaris

[Anonymous]

uxmon is not reading it's config file properly from the
adm directory.

When started uxmon will exit with the following error:

bash-2.05# bin/bb_start start
Starting Big Sister ...
Monitor bsmon OK
Server bbd OK
Agent uxmon OK

Messages:
Monitor bsmon: Mon Dec 15 15:12:14 2003: bsmon:
RRDTool's perl module is not present - not using perl
bindings
Agent uxmon: uxmon: WARNING: host EM appearing in
uxmon config is unresolvable
...............
these go on forever (about 2 per second)
...............

If you look closely, you'll see the host "EM" is
unresolvable. I tracked the error to the uxmon-net
config file - the following lines:

# KEYWORD SYSTEM FEATURES
Apply To HOST
#--------- -------------------------- ----
--------
DESCR features=unix,sysv,solaris
localhost
# DESCR features=unix,linux
someotherhost

The word system is no longer seen as being commented
out. I changed the line to read "sysfeatures" and the
resulting error from uxmon is as follows:

Starting Big Sister ...
Monitor bsmon OK
Server bbd OK
Agent uxmon OK

Messages:
Monitor bsmon: Mon Dec 15 15:18:14 2003: bsmon:
RRDTool's perl module is not present - not using perl
bindings
Agent uxmon: uxmon: WARNING: host SYSFEATURES
appearing in uxmon config is unresolvable
Agent uxmon: Mon Dec 15 15:18:17 2003: uxmon:
check Apply failed to start up - ignored
Agent uxmon: uxmon: WARNING: host SYSFEATURES
appearing in uxmon config is unresolvable
Agent uxmon: Mon Dec 15 15:18:17 2003: uxmon:
check To failed to start up - ignored
Agent uxmon: uxmon: WARNING: host SYSFEATURES
appearing in uxmon config is unresolvable
Agent uxmon: Mon Dec 15 15:18:17 2003: uxmon:
check HOST failed to start up - ignored
Agent uxmon: uxmon: WARNING: host SYSFEATURES
appearing in uxmon config is unresolvable

I am reachable at drew.skinner@ca.pwc.com

Take care,

Drew.

[Thomas Aeby]

Logged In: YES
user_id=77961

This problem has been reported by many Solaris users (but
not all). I have not been able to reproduce this so far and
no one has answered my requests for either access to a
machine with this problem or sending me a system image of a
machine hit by the problem.

Anyway, I keel this report open.

A possible workaround is to install a Perl version that has
not been built by SUN :-)

[Thomas Aeby]

Logged In: YES
user_id=77961

Ok, finally I got access to an installation where I could
track down this behaviour.

The problem is that Solaris fork() does not copy file
handles but rather shares them between child and parent.
Even worse, Solaris exit() fiddles around with file handles.
So, in the uxmon case if a test is used in uxmon-net that
does a fork()/exit() during setup, the config file position
is reset to an arbitrary position by the child process and
therefore, parts of the config file are read more than one
time in the parent.

As a workaround, one can replace every occurrence of exit()
by POSIX::_exit() and add a "use POSIX;" to common.pm.

I am not sure of the scope of this problem. Actually I could
reproduce on Solaris 10 (with multiple Perl versions), only.
The fact that file descriptors are shared between parent and
child is SystemV behaviour(?) and will occurr on more
systems than on Solaris. However, the fact that exit() does
change the file position of a file not even touched in the
child is rather a surprise - I have not yet tracked it down
to Perl or Solaris and actually it does not even matter.