Monitor Vmware ESX

vladi · Guest

hi all,
Does somebody try to use cacti for VMware ESX
i have the mibs, but i can not to create any monitor
some help please

10x

vladi
vladi@we-can.co.il

Guest · Posted: Mon Nov 15, 2004 5:25 pm Post subject:

playing with it now, let you know how it works out

JosephA · Joined: 07 Oct 2004 Posts: 2

Replying to a rather old post, but in case anybody else is wondering...

The console OS for ESX is Linux at the moment, so you can monitor the ESX server just like you would any Linux server. By default snmpd is not running, so you will have to enable that and you should be good to go.

Hannu Liljemark · Joined: 16 Feb 2005 Posts: 4

here's something i made for vmware esx. it's a host template that includes some ucd/net snmp and generic stuff, but also things you can get from the VMWARE agent extensions.

perhaps you want to try the template on some test system... works fine for me but I don't give any guarantees. import the host template into cacti (i'm using 0.8.6d for this), drop the vmware_*.xml into your resource/script_queries directory and the php files into scripts directory.

there's graphs for each virtual machine's memUtil, cpuUtil, networking (packets and traffic) and disk (read/write operations and amount of bytes read/written). any improvements are welcome. especially the php scripts are pretty ugly... probably because i've never written php before.

don't forget the configure snmp in vmware. enable it from the web browser gui and make further configuration in /etc/snmp/snmpd.conf just like you do in any ucd-snmp/net-snmp system.

Hannu Liljemark · Joined: 16 Feb 2005 Posts: 4

here's a few more screenshots

ajeskey · Posted: Mon May 23, 2005 8:24 am Post subject:

ah neat this saves me a bit of time. I just found the MIB's last night and was going to work on this today.

Just one suggestion on the memory part, maybe doing an inverse stack on the free memory so that it comes from the top down and not bottom up, would help to be a bit more descriptive.

Hannu Liljemark · Joined: 16 Feb 2005 Posts: 4

the php scripts could use some work, there's way too much snmp queries being done. looking at other scripts they seem to do some fetches from the database and cache stuff - maybe that'd be a good idea. and instead of script query i guess they should be redone for the script server thing in 0.8.6.

i have no idea how well the graphing will work when you e.g. shutdown all virtual machine and restart them in another order than when you initially created the graphs. it's kinda messy figuring out which cpu, memory, disk and nic in the RESOURCE mib belongs to which VM in INFO mib. i'm not completely sure how the indexing should be done. virtual machine can be down and then it won't have anything in the RESOURCE mib and machines can have more than one disk, nic and cpu, i guess.

i haven't tested with two cpu systems. all the VMs here are single cpu.

that reversed memory is a good idea, but i couldn't figure out how to do it.

ajeskey · Posted: Tue May 24, 2005 8:37 am Post subject:

I am noticing that their poll-ability does suck. Each index should have a reference to the Virtual Machine, that right there would make things a lot smoother. It seems silly that a company that really bases their product off of scalability and consolidation (both things that really need to have well defined monitoring to know) has such a poor implementation of their monitoring.

I’ll email my contacts over at VMWare and see what their plans are for upcoming versions. If I hear anything I’ll let you know.

As for now, I am not even going to bother with this, not that my Virtual Servers reboot but the machines do, and if that causes an enumeration, what’s the point of graphing?

ajeskey · Posted: Tue Jun 21, 2005 8:26 am Post subject:

I think I have found away around the difficulties of losing the machine index when the VM PID changes

Table .1.3.6.1.4.1.6876.2.1.1
VM ID .1.3.6.1.4.1.6876.2.1.1.1
VM Name .1.3.6.1.4.1.6876.2.1.1.2

If you did something like BSOD2600 is doing in his Win32 Process monitor ( http://forums.cacti.net/viewtopic.php?t=6661 ) to base the indexing on the name, you could relate that back to the VM PID.

Just a thought.

laffen · Cacti User Joined: 13 Jul 2004 Posts: 52 Location: Norway

Hi,

has anyone played more with templates for the ESX server?

titan · Joined: 05 Jan 2006 Posts: 17

I too am very interested in monitoring our ESX servers.

I'm quite happy to pay for someone's time if need be. I'm a cacti noobie so haven't a clue on how to create the required entries for a set of graphs based on new OIDS.

bpowers7 · Joined: 04 Apr 2006 Posts: 7

how can I get the graphs to post the virtual server name instead of the 'vmIDX'?

ymartin59 · Joined: 17 Feb 2006 Posts: 40

Hello,
After a upgrade from Debian cacti 0.8.6c to current 0.8.6i, I'm now able to test that ESX template.
I have first updated scripts to use new 0.8.6i SNMP authentication string parsing - because I use snmp v3 ! I will post the result of my work when satisfied.

ymartin59 · Joined: 17 Feb 2006 Posts: 40

After investigation, available information through ESX SNMP is really poor compared to what vmkusage/vmktree collects - especialy for memory usage and CPU.

By the way, I updated the template:
- use script server method
- include ESX reserved console memory
- use VM display name as key order for CPU and Memory
- create a virtual reference (vmDisplayName/hwname) as key order for disk and NIC
- improve graph titles and data source names

I still have doubts about graph units for VM NIC traffic (kb/sec or bits/sec)
and for CPU ?? I kept the original unit/legend from the previous template.
Let me know if that is wrong.

--
Update: I have found that the "ESX Memory Usage" graph is buggy concerning the "GPRINT" legends. I will post an update next week after I check the units

fmangeant · Posted: Fri Nov 10, 2006 10:09 am Post subject:

Hi

thanks for these new templates !

However, here's what I get when I try to import them :