VMware: How to achieve a perfect metric collection interval with Telegraf, InfluxDB and Grafana

Greetings friends, I have told you in the past how to monitor your VMware environment using InfluxDB, Telegraf, and Grafana, and according to Grafana’s dashboard, you are using it with about 4300 people in your Datacenters. All this is great, and I thank you very much for all the support and that so many people have the solution deployed.

Now, I have read for a long time that in some cases you were having incidents when monitoring the datastores in a correct way, which generated that the Dashboard will not display correctly, creating a lot of frustration, but this entry will ensure you have everything configured correctly.

vSphere vCenter – real-time metrics

VMware collects information every 20 seconds from the ESXi and stores this information in vCenter RAM. This makes it really easy and fast to access this information by telegraph or any other information collector. All this information is stored only for one hour in the vCenter, in the diagram we would be number 3, accessing directly to the Performance Manager:

This information is limited however to the resources we can obtain from the ESXi hosts and the VMs, which are the following:

Knowing this information now, InfluxDB Engineers have managed to achieve metrics of up to 7000VMs in about 20 seconds, so I am sure you will be able to deploy this solution in your environments without any problems.

vSphere vCenter – historical metrics

These metrics are stored in the VMware vCenter Database and are much heavier and more expensive to perform, as well as taking longer to return the response, etc. These metrics are stored in 5 minutes, 30 minutes, 2 hours, and 24-hour intervals. But we will use the range of every 5 minutes to have updated information. In the diagram, what we want here is to access the information in number 2:

This information is limited to the metrics we can get from the Clusters, Datacenter and Datastores. As we can see, if we don’t modify the telegraph configuration, we won’t be able to monitor as we should these two intervals that are different and limited to different resources.

Recommended telegraph configuration for your VMware vSphere environments

Having seen the theory, and knowing how the different elements to monitor, I recommend as always not to touch the telegraf.conf, better to create a file in the telegraf path, something like this – /etc/telegraf/telegraf.d/vsphere-monitor.conf and inside the following:

If we look at the configuration, of the first interval, shorter, we will have information in groups of 20 seconds about VMs and Hosts, and we exclude everything else like Datacenters, Datastores, and Cluster.

In the second part, what we do is to exclude elements such as VMs and Hosts, and monitor the elements with the statistics in the database, as well as upload the collection tasks to three to speed up the process.

Final result

The final result since a few weeks is that I do not stop receiving information about the Datastores, or other elements of my VMware when I monitor them with Telegraf, and the result can be seen here:

vSphere Overview Dashboard

vSphere Hosts Overview Dashboard

vSphere Datastore Overview

vSphere VM Overview

That’s all folks, if you want to follow the full Blog series about Grafana, InfluxDB, Telegraf, please click on the next links:

Note: If facing the error “Task Name: Remote View Manager, Status: The request refers to an unexpected or unknown type” please read the next Blog entry.

Author: jorgeuk

Father, writing in https://www.jorgedelacruz.es and https://jorgedelacruz.uk Blogger, Systems Engineer @veeam - vExpert 2014/2020 & NTC 2018/19

6 Thoughts

  1. None really, MySQL will not be able to handle the billion of metrics you pull and save from vCenter, for that reason VMware just saves a short period of them before flush them.

  2. Very nice post, thanks. Can you be more specific about the vCenter statistics config level for the 5 min interval duration please (staying at level 1 or level 2) ?

  3. It collects only only the local esx datastore. Shared like 3par datastores are reflected any idea on that?

  4. Hello,
    This specific article is for VMware Datastores, yes, whatever it is behind, like 3par, Netapp, etc, that does not matter, as VMware sees it.


  5. Hi, we implemented the dashboard. But the values on cluster level are different than in Vsphere. It shows 100% CPU in the dashboard, but in VMware the value is below 100%

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.