1.1beta9 - Incorrect Disk Space Values on Linux

[rob9988]

Has anyone figured out why Zabbix reports different values for disk space on Linux vs. Solaris? Here's what I am using for a common UNIX item in Zabbix

Disk - Free space on /
vfs.fs.size[/,free]
Numeric (float)
B
Custom multiplier
1048.576 (which is of course, 1024 * 1.024)

For Solaris, the results are right on the mark. However, for Linux, the numbers are off by a decent amount. I have tried using multipliers of 1000, 1024, and 1048.576, and can't get the Linux numbers to match any of them.

Here are the numbers that I am seeing with Zabbix and df:

Linux box
---------
Zabbix:
Disk - Percentage used on / 16 May 11:36:30 14.25 %
Disk - Free space on / 16 May 11:37:00 7.63 GB

user@linuxbox:~> df -k
/dev/mapper/system-root 9113316 1298216 7815100 15% /

user@linuxbox:~> df -h
/dev/mapper/system-root 8.7G 1.3G 7.5G 15% /

Solaris box
----------
Zabbix:
Disk - Free space on / 16 May 11:57:00 148.99 MB
Disk - Percentage used on / 16 May 11:57:30 90.81 %

[user@solarisbox /]$ /usr/local/bin/df -k
/dev/dsk/c0t0d0s0 1621767 1424125 148989 91% /

[user@solarisbox /]$ /usr/local/bin/df -h
/dev/dsk/c0t0d0s0 1.6G 1.4G 146M 91% /


Any suggestions? I have searched the forums extensively, and have read the docs and a blog or two about other people's experiences, but haven't found an answer yet.

Thanks,

=-Rob

[marius]

I can confirm this. I hope it gets fixed soon.

[dcrandall]

zabbix builtins not reliable

My experience so far:
I'm using zabbix-1.1beta9 on FreeBSD and Debian.

I've had a lot of problems with file system stats gathering as well, either being unsupported or returning bad data, and have resorted to writing my own tools and using UserParameter to get the info into zabbix.
For example: after fussing around with vfs.fs.* parameters, unsuccessfully trying to find a way to use them in a template that works for all of my *nix machines, I wrote a short perl script that uses df and works on all of my machines. Added UserParameter for it on each client and moved on.
Of course that's the rub: You have to maintain the zabbix_agentd.conf on each client.

Don't get me wrong... I like Zabbix a lot! It's a great tool, and I'm excited to see it grow.
The zabbix framework: front end, triggers, graphs, etc are great. But, the builtin zabbix_agentd parameters need a lot of work.

Daniel

[marius]

My experience so far:
I'm using zabbix-1.1beta9 on FreeBSD and Debian.

I've had a lot of problems with file system stats gathering as well, either being unsupported or returning bad data, and have resorted to writing my own tools and using UserParameter to get the info into zabbix.
For example: after fussing around with vfs.fs.* parameters, unsuccessfully trying to find a way to use them in a template that works for all of my *nix machines, I wrote a short perl script that uses df and works on all of my machines. Added UserParameter for it on each client and moved on.
Of course that's the rub: You have to maintain the zabbix_agentd.conf on each client.

Don't get me wrong... I like Zabbix a lot! It's a great tool, and I'm excited to see it grow.
The zabbix framework: front end, triggers, graphs, etc are great. But, the builtin zabbix_agentd parameters need a lot of work.

Daniel

I feel the same way. I ended up creating my own UserParameters for stuff like net.in/net.out, memory free, etc.