ESX LiteStats - Detail?

[ttmader]

Hi,

I have sexigraf implemented but when I try to see more details (30 days ago) the graph is is different.

Here is a comparison from vSphere Client and Sexigraf.

Can you put more detail on this graph?

Thank you very much.

Greetings.

[rschitz]

Hi, it is expected for small spikes to be averaged for several reasons:

• we wanted a lightweight appliance even with 100000+ vms monitored
• we wanted the be able to keep 5y old data
• we wanted the appliance to stay fast even with 100000+ vms monitored
• we wanted to be able to display several tiny graphs in the dashboards and not having trouble to read it like this:
  

For all those reasons we choose this storage schema : 5m:24h,10m:48h,60m:7d,240m:30d,720m:90d,2880m:1y,5760m:2y,17280m:5y
you'll find the details of how it works here : https://graphite.readthedocs.io/en/latest/config-carbon.html

if you need more granularity, i can give you the method to change it and reconfigure the existing files but the size of the appliance will grow so if you have a lot of vms you might need to check the free space

[rschitz]

FYI i'm working on a way to solve that in the future without having to scarifying the space. something like mixing avg and max values

[ttmader]

I was really blown away by the ability to collect so many metrics with very little storage. It's amazing. Everyone talks about Sexigraf, at first they laugh but then they realize how powerful it is.

However, many times, our bosses want metrics and more metrics. This is what I need :(

I am going to check what granunality we need but if you can tell me how to change it I would be grateful. We will have to see how much storage grows. I have a very small environment, about 2500 vms

A new version? Cool!

Thank you very much for answering so quickly!

[rschitz]

Thanks for your support.
Changing granularity is easy, there is the graphite configuration (for future whisper files) and the change we must apply on the existing whisper files.
Just let me know what you'd like and i'll tell you how to do it. As of now, in the last 30d period there is 720 minutes between points

[ttmader]

Hi,

Have you been able to review the steps I must follow to change this granularity?

Thanks.

[rschitz]

You need to give me what you'd need instead of this : 5m:24h,10m:48h,60m:7d,240m:30d,720m:90d,2880m:1y,5760m:2y,17280m:5y

[ttmader]

At the moment, I only need to modify the value of 24h:

1m:24h,10m:48h,60m:7d,240m:30d,720m:90d,2880m:1y,5760m:2y,17280m:5y

Thanks.

[rschitz]

You cant get 1m since we are collecting every 5min. Some of the metrics comes from the perfmanager so we could mimic 1min granularity but lot of metrics comes from property that we only collect every 5min so that's not possible

[ttmader]

Isn't it possible in any way? It's a shame.

We will have to find an alternative monitoring for this.

Thank you very much too.

[rschitz]

there is no monitoring tool that gives you 1min granularity for vsphere because none are cable of a 1min cycle on a medium infra. the only thing we could do is take metrics every 5min but split it into 1min batches. you'd have granularity but not frequency. May i ask why you need this?

[ttmader]

No problem you can ask. :) We have many machines that have spikes that are less than 5 minutes and we need to detect them. We had looked for an alternative like Telegraf but we had seen Sexigraf could be worth it.

[rschitz]

Thanks. On what metrics you'd need that?

[ttmader]

For ESX LiteStats.

Thanks.

[rschitz]

so cpu/ram of the ESXs?

[ttmader]

Exacly

[rschitz]

Sorry but that wont be possible since we are using overallCpuUsage and overallMemoryUsage every 5min for ESX and VM https://vdc-repo.vmware.com/vmwb-repository/dcr-public/c476b64b-c93c-4b21-9d76-be14da0148f9/04ca12ad-59b9-4e1c-8232-fd3d4276e52c/SDK/vsphere-ws/docs/ReferenceGuide/vim.host.Summary.QuickStats.html

[ttmader]

It's a shame.

Well then, I'll talk to my managers and tell them the news.

Thanks again for your help!

[rschitz]

i'm working on way to switch between average and max values so you could check peaks this way, i'll let you know if you're interested to test.
in the mean time, check how much time it takes to collect right now using the Pull Exec Time dashboard please

[ttmader]

It would be interesting if you get it. I would like to try it.

This is the Pull Exec Time graph:

[rschitz]

that would be low enough to do a 1min polling exept for one but that would be point less IMO

[rschitz]

see #349