Why can't I understand 1 minute polling, or is it just me?

stormonts · Post by **stormonts** » Wed Jun 24, 2009 9:45 pm

I inherited maintenance of our Cacti 0.8.7b install which runs on Linux. All of our data sources had a "Step" set at "300" and a "Heartbeat" set at "600". The "Poller Interval" on the server is set to "Every minute".

We wanted to start graphing every minute. I'm not sure how many of these data sources/graphs are default Cacti ones vs. the ones that the previous administrator created, but here is what I did. I went to each data source template and duplicated it. I then changed the Step to "60" and the "Heartbeat" to "120". I then made a duplicate of each graph template and changed the "Data source" for each graph item to the ones with the new step of 60. I then went to each device, duplicated each data source, and then changed the "Selected Data Template" field to point to the new one with the step of 60. Finally, I duplicated each graph for that device and then pointed it to the updated "Selected Graph Template" and changed the "Graph Item Fields" to the updated "Step" of 60 ones.

The attached images show the results that I am getting for "Linux - CPU Utilization Percentage / 4 - nice" (uses ucd/net CPU Usage - User, System, Wait, Nice, and Idle) and "Linux64 - Load Average" (uses Linux64 - Load Averages - load1, load15, and load5).

I understand that the "Minimum", "Maximum", and "Average" values are all different because the top graph in each image existed for a full day while the second graph only existed for 7 hours, but why don't the actual percentages shown in each graph sync up? What did I screw up?

khufure · Post by **khufure** » Thu Jun 25, 2009 4:58 pm

Did you check your CDEFs? Almost looks like its dividing an extra time, or that the values got mixed in CPU. Load looks ok so I think the interval upgrade is actually good.

stormonts · Post by **stormonts** » Thu Jun 25, 2009 7:15 pm

khufure wrote:Did you check your CDEFs? Almost looks like its dividing an extra time, or that the values got mixed in CPU. Load looks ok so I think the interval upgrade is actually good.

The CDEFs are exactly the same on both the original 5 minute graph template and the 1 minute on that I created as a duplicate. Maybe they were never setup correctly, or I just need to tweak something for the new one minute polling?

Attached are shots from the 4 CPU template. The previous admin set the CDEF up to divide by 4 (I'm assuming so that the graph would show essentially per CPU instead of the total load returned from that specific OID?). Regardless, that is the CDEF used for each item in each template.

Do I maybe need to change the "Consolidation Function" to "Last" (instead of "Average" which is what was being used in the original 5 minute template?) since I am now polling it every one minute?

The "Divide by 4" CDEF below is: cdef=CURRENT_DATA_SOURCE,400,MIN,4,/

The cpu_user data source below is polling OID .1.3.6.1.4.1.2021.11.50.0 which is UCD-SNMP-MIB::ssCpuRawUser.0

cpu_nice polls OID .1.3.6.1.4.1.2021.11.51.0 which is UCD-SNMP-MIB::ssCpuRawNice.0

cpu_system polls OID .1.3.6.1.4.1.2021.11.52.0 which is UCD-SNMP-MIB::ssCpuRawSystem.0

cpu_wait polls OID .1.3.6.1.4.1.2021.11.54.0 which is UCD-SNMP-MIB::ssCpuRawWait.0

Post by **TheWitness** » Tue Jun 30, 2009 5:33 pm

Before you go down the 1 minute road, you first need to understand that there are a few translations issues in the current code:

1) Step is in second, so 300 is 5 minutes and 60 is 1 minute
2) The 1 Minute and 5 Minute RRA's conflict with one another. Make sure you repair them first.

If you don't understand 2, goto RRDtool.org and read up on RRDtool. Else wise, add my SpikeKill plugin and perform and Analysis from the WebUI and you will see the issue.

If you combine the current 1 Minute (stock) and 5 Minute (stock) RRA definitions, you will only get a few months of graphs and then, all of a sudden, the data will stop graphing.

Sound familiar? You need to change a few things to get 1 Minute to run seamlessly.

It wasn't out intention to confuse everyone, but it's too late for that. Sorry.

TheWitness

EDIT: SpikeKill and not SuperLinks. Need more coffee.

stormonts · Post by **stormonts** » Tue Jun 30, 2009 5:59 pm

I understand #1 but not #2. What version of Cacti does Superlinks work on?

One of these two things fixed my issue: changed the cdef to that is just the source divided by 4 with no mention of the MIN or 400 variables and I changed the max value of the cpu data sources from 100 to 10000 (I assume this could only max out at 100 x the number of cores, but I wanted to be safe)

TheWitness wrote:Before you go down the 1 minute road, you first need to understand that there are a few translations issues in the current code:

1) Step is in second, so 300 is 5 minutes and 60 is 1 minute
2) The 1 Minute and 5 Minute RRA's conflict with one another. Make sure you repair them first.

If you don't understand 2, goto RRDtool.org and read up on RRDtool. Else wise, add my SuperLinks plugin and perform and Analysis from the WebUI and you will see the issue.

If you combine the current 1 Minute (stock) and 5 Minute (stock) RRA definitions, you will only get a few months of graphs and then, all of a sudden, the data will stop graphing.

Sound familiar? You need to change a few things to get 1 Minute to run seamlessly.

It wasn't out intention to confuse everyone, but it's too late for that. Sorry.

TheWitness

Post by **TheWitness** » Tue Jun 30, 2009 6:16 pm

Sorry, SpikeKill. Need another hobby

TheWitness

Post by **TheWitness** » Tue Jun 30, 2009 6:21 pm

Reltive to your other issue, keep in mind that some of the net-snmp MIBS are based upon timetics and not cpu percent, so it's a game of add um all up, do some math and you get a percentage. This is one of the reasons that they don't work too good with pure SNMP, unless your'e a CDEF genious.

I don't know if that's your issue or not. Don't have the cycles to spend. Best bet is to get closer to the data and then work your way out. Aka, start with the MIB explanation, then RRDtool data storage (guage vs. something else).

Then finally to RRDtool graph magic.

TheWitness

stormonts · Post by **stormonts** » Tue Jun 30, 2009 7:45 pm

We are only running version 0.8.7b, so now Spikekill for me yet.

Post by **TheWitness** » Tue Jun 30, 2009 7:48 pm

You can run it from the command line too. Make sure you have a wide display.

TheWitness

stormonts · Post by **stormonts** » Wed Jul 08, 2009 1:01 pm

TheWitness wrote:You can run it from the command line too. Make sure you have a wide display.

TheWitness

I installed Spikekill and did an analysis on a graph, but I'm not quite sure what I should be looking for.

Why can't I understand 1 minute polling, or is it just me?

Why can't I understand 1 minute polling, or is it just me?

Who is online