David,
As it stands we do not parse the java process name for information as it
quite often gets corrupted by the process size being over 2GB.
I've yet to find a solution that works for this.
Regards,
Mike Rowell
-----Original Message-----
From: Gore, David W (David) [mailto:user-3e5761c68b56@xymon.invalid] Sent: 12 February 2007 15:20
To: user-ae9b8668bcde@xymon.invalid
Subject: RE: [hobbit] Solaris ps
-----Original Message-----
From: Mike Rowell [mailto:user-63f3e97eb1de@xymon.invalid] Sent: Monday, February 12, 2007 09:45
To: user-ae9b8668bcde@xymon.invalid
Subject: RE: [hobbit] Solaris ps
I've monitored this discussion with considerable interest, one thing
that some people may not be aware of, is that over 2GB of memory size,
/usr/ucb/ps will corrupt the process list (i.e. it's not largefile
aware).
Now for the majority this will not be an issue, however on the majority
of my production systems we run java processes with a memory size of
about 4GB. We also run Databases with inexcess of 10GB resident memory
size.
Regards,
Mike Rowell
Mike, you must have short java process names or you do not need to
extract any identifying process name info past 80 characters then? Java
processes names are exactly why we have wanted /usr/ucb/ps for many
months. Almost all of our hobbit clients have to be updated to use
/usr/ucb/ps.
I also use a PERL script to extract CPU and memory info for these same
processes using /usr/ucb/ps. We have also had to preface /usr/ucb/ps
with sudo to accommodate Solaris 10 and Solaris 8 hosts with the latest
patches.
We have something that works now, so it's no big deal. Having a new
built-in feature that tracks CPU and memory would be nice, but I am
expecting it to use the current ps. So it will be of limited use to us,
unless it can track super long java process names.
~David
-----Original Message-----
From: Galen Johnson [mailto:user-d2ff723b6cb6@xymon.invalid] Sent: 12 February 2007 03:43
To: user-ae9b8668bcde@xymon.invalid
Subject: Re: [hobbit] Solaris ps
Galen Johnson wrote:
Henrik Stoerner wrote:
On Sat, Feb 10, 2007 at 06:22:52PM -0500, Galen Johnson wrote:
Is there any reason NOT to use '/usr/ucb/ps auxwww' in place of >>>>> the ps that hobbit-sunos.sh is using?...in other words, will it >>>>> break anything if I use it (yes, I know I need to run it as root >>>>> to get everything)? I prefer the output of what Henrik provides >>>>> but the solaris ps is braindead (truncates at 80 characters).
Yeah, but all of our systems are SPARC...which does...even with >>> Solaris 10. Question stands...is there a reason not to? (other >>> than upgrading, will it affect any of hobbit's builtin
functionality?)
If you look at the hobbitclient-sunos.sh script that runs on the
clients, you'll see that it currently runs
ps -A -o pid,ppid,user,stime,s,pri,pcpu,time,pmem,rss,vsz,args
If you can come up with a /usr/ucb/ps command that provides the same >> information - hopefully using almost the same column headers - then
I have no problem in switching.
Right now, Hobbit only looks at the command line from the "ps"
listing,
but I do have a plan to make it track the memory- and cpu-utilisation
of
processes; and for that Hobbit must be able to identify the
corresponding columns in the "ps" output. But that's in the future.
Thanks...I looked into trying just that. I can get using options > 'auxwww' gets me a lot of the fields and I can get others with > 'alxwww'...unfortunately, the 'l' and 'u' options appear to be > mutually exclusive...which is annoying. 'auxwww' does give you some > memory and cpu info (I _think_ it's the same as the pcpu and pmem). I
wouldn't necessarily suggest this as a permanent change but I need it > to get more information from the command line.
I can work up the changes to provide both as an option (it does > require that the -o args be rearranged since the output of /usr/ucb/ps
isn't configurable).
=G=
These two commands are equivalent...
/usr/ucb/ps auxwww
USER PID %CPU %MEM SZ RSS TT S START TIME COMMAND
/usr/bin/ps -A -o user,pid,pcpu,pmem,vsz,rss,tty,s,stime,time,args
USER PID %CPU %MEM VSZ RSS TT S STIME TIME
COMMAND
I've verified that SZ and VSZ are the same by comparing output...however, there is a bit of a shortcoming with the berkeley ps...it runs %MEM, SZ and RSS together. It looks like the percent memory is consistently 3 chars wide (%1.1f)...I wish I could say the same about SZ and RSS...
=G=
This email has been scanned for all viruses by the MessageLabs service.
This email has been scanned for all viruses by the MessageLabs service. ______________________________________________________________
This email has been scanned for all viruses by the MessageLabs service.
This email has been scanned for all viruses by the MessageLabs service. ________________________________________________________________________