ucd/net-SNMP lmSensors -- update Feb 10, 2007

[ehall]

I have been working to get different kinds of hardware monitoring working in a consistent manner. This is the ucd/net-snmp lmSensors package.

Here is what the final result looks like for one of my systems, and shows the fan speeds, thermal probe temperatures, and voltage levels for all of the available sensors on that server:



This model uses a script query which uses SNMP to ask the target for all known sensors of the requested type. Add the appropriate data query to a device that has the ucd-net/snmp lmSensors extensions, and then create new graphs for each of the sensors that you want. Each selected sensor will be created as an individual data-source and graph, like with interfaces.

Graphs can be consolidated by editing and/or using one of the bundled "8x" graph templates, once the underlying data-sources and graphs have been created. If you have 8 items on a host, simply use the Graph Management menu to "add" a new graph, choose the target system, and then point to the data sources you created with the data query. If you have fewer or more than eight items, you can make a copy of the 8x template and add/delete however many more rows you want. Once the consolidated graphs are working, you can delete the sensor-specific one-off graphs, but DO NOT DELETE THE DATA SOURCES.

The screenie below uses the 8x template for voltages, but as you can see in the large image above, the other sensor groups on this host use 2x graphs.



There are some caveats here. Obviously, you must have lm_sensors installed and working, and you must have the ucd-snmp/lmSensors extension compiled into your version of net-snmp. Alos note that net-snmp uses textual string parsing to determine what kind of sensor you are using---sensor names with a "V" are mapped to voltage sensors, sensor names with "fan" are mapped to fans, and sensor names with "temp" are mapped to thermals. If you don't use that naming, your sensors will end up in the miscellaneous pool, and this script doesn't look in that pool.

If there are any additional updates to this, I will replace the attachment below.

This was exported from cacti 0.8.6i, using PHP 5. Other combinations may not work.

Other helpful links:

lm_sensors home page
net-snmp home page

[gandalf]

Wow, nice! Was just googling around for this bit. But to be honest, I'm currently not getting lmSensors to give me data. net-snmp-config tells me it is included. Is there anything else to be aware of? (Forgive me to be off-topic)
For used software, please see my signature.
Reinhard

[ehall]

Do you have lm_sensors configured and working? That's the first step and can be a kind of a pain to get fully working automatically.

First things first: do you have the "sensors" command in your path, and does it display any meaningful output?

If not, you should have a "sensors-detect" command installed somewhere that will probe for the different kinds of sensors available. If it it detects them, it will spit out a series of modprobe lines that tell you which sensors were discovered. Once those modules are loaded you can tweak /etc/sensors.conf to manipulate things like labels, fan multipliers, and so forth.

sensors-detect will also offer to create an /etc/sysconfig/lm_sensors control file (or something like it, depending on platform). I usually expirement with those and other availalbe modules, and put them into the sysconfig/kernel file so that they are loaded on boot too.

All of the above comes from installing the appropriate lm_sensors package, of course (check the links in top message). Once that's all working, adding support to net-snmp usually just requires tweaking the configure script and building a new package, but that's not even needed on some platforms.

Once you spend enough time with it to wrap your head around the architecture, it is pretty straightforward.

[gandalf]

Thanks for the links. IBM systems are supposed not to work with current lm_sensors ...Damn it
Reinhard

[ehall]

lm_sensors talks to regular kernel modules, and there are some independent modules floating around separate from the ones that are bundled into lm_sensors itself. Not saying that you'll find one for your boxes but it's worth googling for.

[rony]

From what I remember when I was playing with lm_sensors, you have to load a module for the base, and then each componet you want to monitor. Or was that APCI?

...

[ehall]

More or less, but on the lesser end of the scale it's so generic as to not matter. Some of the higher end sensor modules can have a lot of dependencies but the lower end sensors just need an SMBus or ISA driver (both of which are bundled into lm_sensors already).

[gandalf]

[Cornfed]

I'm confused what to do with the files. I was able to import the XML files, and that looks fine.

What kind of values do I put in for Index Type, Index Value, and Output Type ID?

What do I do with the PHP file? In the scripts directory?

I'm used to the older verions of cacti (before the xml), and haven't worked with the xml files before. (I'm now using the most current version)

[ehall]

From the main console you can use the "import templates" menu on the left to pull in the XML files. That will create the graph templates, the data query, and so forth. Then go into the device screen, and add the data query to a device that has lmsensors and net-snmp (compiled with lmsensors support). If everything is working on the back-end system the sensors should show up.

[Cornfed]

I'm almost there. When I add the data query to the device, click create, I get this message. I don't know what it's asking.

Is the Index Type = LM-SENSORS-MIB::lmFanSensorsValue
Is the Index Value = (1,2,3, number of fans)
(as in LM-SENSORS-MIB::lmFanSensorsValue.1)

No idea what Output Typ ID is. I'm guessing that's a number that I would assign to make this fan different from other fans I might track on this device.

[ehall]

You shouldn't be getting prompted for that information, since the script will populate the data as needed.

One possibility is that you are not using 0.8.6h which is what the XML and scripts were written to work with.

Although actually it kind of looks like you tried to add the graph template or data template to the device. You need to add the data query to the device and let it do the work. Once that is done, click the "Verbose Query" link in the Associated Data Query list, and you should get some debug output. Mine shows:

[Cornfed]

Ahhh, Maybe that's the source of my confusion. Should the tar file have contained lmsensors_fan.xml?

The contents look like all graph templates:

[ehall]

I'm pretty sure that the 0.8.6h graph templates include all the other templates. I exported them with all dependencies. The 8x templates don't (they are just additional graphs) but the regular ones should have all the data query templates and everything.

Look in the Data Queries screen. Does it have three "ucd/net-snmp - lmSensors - Get XXX Sensors" queries?

[Cornfed]

First, I tried to re-import the xml file.