Display open file handles

[stigmata]

Hi,

is there any way to monitor the number of open file handles on the linux system?

Regards,
Oliver

[gandalf]

If you have a script that simply returns this wanted value or an OID for that, the answer ist: YES. There's a way to implement this script at http://www.cacti.net/downloads/docs/htm ... thods.html
Reinhard

[stigmata]

Hi Reinhard,

currently I'm searching for this script
"lsof | wc -l" <-- I can't believe that this is the best way to get the open files...

Oliver

[stigmata]

I made this script:

--------------------------------------------------------------------------------
#!/bin/bash

TEMP=`ls /proc/`
TEMP=`echo $TEMP | sed "s/[a-zA-Z]//g" | sed "s/[.]//g"`
NR_HANDLES=0

for PID in $TEMP;
do
NR_PID=`/usr/sbin/lsof -p $PID | wc -l`
NR_HANDLES=$[$NR_HANDLES+$NR_PID]
done

echo $NR_HANDLES
--------------------------------------------------------------------------------

[gandalf]

Well, seems nice, but has problems with processes like sysrq-trigger (the "-" sign!). You'll have to define a Data Input Method for that, a Data Template and a Graph Template as described in http://www.cacti.net/downloads/docs/htm ... thods.html
Reinhard

[stigmata]

I changed to a new solution:

#!/bin/bash

TEMP=`ls /proc/`
PID_EXCLUDE=$[$$+1]
TEMP=`echo $TEMP | sed "s/[a-zA-Z]//g" | sed "s/[.]//g" | sed "s/$PID_EXCLUDE//g"`
NR_HANDLES=0


for PIDS in $TEMP;
do

FD_COUNTER=0

for FD in $(ls /proc/$PIDS/fd/);
do
FD_COUNTER=$[FD_COUNTER+1]
done

FD_COUNTER_ALL=$[FD_COUNTER_ALL+FD_COUNTER]

done
echo $FD_COUNTER_ALL


because lsof generates high load on the server....

[gandalf]

Yes, it's a lot faster. But three runs on my host (laptop) yielded the follwoing

ls: /proc/4962/fd/: No such file or directory
ls: /proc/4963/fd/: No such file or directory
ls: /proc/-/fd/: No such file or directory
1162
[root@gandalf scripts]# . lvm_file_handles.sh
ls: /proc/5193/fd/: No such file or directory
ls: /proc/5194/fd/: No such file or directory
ls: /proc/-/fd/: No such file or directory
2324
[root@gandalf scripts]# . lvm_file_handles.sh
ls: /proc/5422/fd/: No such file or directory
ls: /proc/5423/fd/: No such file or directory
ls: /proc/-/fd/: No such file or directory
3486

So there has to be some more error handling, I suppose ...
Reinhard

[stigmata]

hmm, strange - on our server this scripts works fine.
sorry thats not running on your laptop.

[gandalf]

Don't bother for that, just thought it would make the script more stable to work on that. My system runs:

Linux 2.6.9-5.EL #1 Wed Jan 5 19:22:18 EST 2005 i686 GNU/Linux
0.8.6g
CACTID 0.8.6f Copyright 2002-2005 by The Cacti Group
PHP 4.3.9 (cgi) (built: Aug 17 2005 10:22:19)
Copyright (c) 1997-2004 The PHP Group
Zend Engine v1.3.0, Copyright (c) 1998-2004 Zend Technologies
RRDtool 1.2.11 Copyright 1997-2005 by Tobias Oetiker <tobi@oetiker.ch>
mysql Ver 14.7 Distrib 4.1.10a, for redhat-linux-gnu (i386)
gcc (GCC) 3.4.3 20050227 (Red Hat 3.4.3-22.1)

Did not find time to debug this on my own yet...
Reinhard

[wytzevanderploeg]

Thank you for your script.

I have some entries in /proc/ which contain a '-' (dash) sign. These won't get replaced by sed. (sysrq-trigger, key-users) which result in a '-' sign as given directory and thus giving an error. If you add a dash to the sed script it works really well.

So the TEMP line would become like this:

Code: Select all

TEMP=`echo $TEMP | sed "s/[a-zA-Z-]//g" | sed "s/[.]//g" | sed "s/$PID_EXCLUDE//g"`

UPDATE:
I modified the script a bit. I hope you don't mind:

Code: Select all

#!/bin/bash

NR_HANDLES=0

for PIDS in $(ls /proc/ | sed "s/[^0-9]//g");
do
        if [ -d /proc/$PIDS ]; then
                NR_HANDLES=$[NR_HANDLES+$(sudo ls /proc/$PIDS/fd/ | wc -w)]
        fi
done

echo -n $NR_HANDLES

Because Cacti was running as www-data I had to add the sudo part to be able to list the contents of the fd dirs. The line in /etc/sudoers looks like this:

www-data ALL=(root) NOPASSWD:/bin/ls /proc/*/fd/

I attached the xml file for your convenience. (Exported with Version 0.8.6i)

[Cactus84]

So I tried this one and the command worked fine in the shell, but on the graph it was way off:

./file_handles.sh
1644

Does anything stand out?

Thanks