Metrics does not send information of all LXC/VM

[MisterY]

I have InfluxDB and Grafana running and pointed to them in the metrics settings.

Approximately 10-20 % of my LXCs/VMs are missing in Grafana/InfluxDB.

What I did to solve the problem:
- Upgraded from PVE7 to 8.1
- Installed new Grafana and InfluxDB
- Upgraded to InfluxDB 2 (what a pain in the ass)
- installed new LXCs/VMs
- Installed a new server

Everytime the same problem: Some of the LXCs/VMs are missing!

What is the problem?

 

[anderl1969]

Sadly there was never an answer to this. I have the same issue here. I'm running an InfluxDB2 in a LXC container. It's configured as metric server within the Proxmox datacenter. It works, but when I'm creating a new LXC or VM, there is a certain chance, that it's metrics are not arriving at the InfluxDB2.

Any ideas?

 

[anderl1969]

Was able to identify and solve the problem - at least for me!


The causes were the Proxmox tags:

• they are concatenated by Proxmox and passed to InfluxDB separated by semicolons: <tag1>;<tag2>;<tag3>...
• In the vast majority of cases, InfluxDB interprets this as string.
• Exception: If only a single tag is defined and this is a number, e.g. 12 - in this case the transferred value is interpreted as float.

• Depending on which type of tag (string or float) arrives first at InfluxDB, the data type is set in the InfluxDB bucket. As a result, all VMs or LXCs are dropped with the other type of tag because it does not match the data type of 'tag'.

So either add at least one alphanumeric tag to all VMs/LXCs with numeric tags or even better: avoid numeric tags completely!

 

[dcsapak]

Hi,

it seems the original report was overlooked at the time...

but thanks for finding the actual problem @anderl1969 , i now sent a patch that the tags field always gets sent as a string, regardless of content:
https://lore.proxmox.com/pve-devel/20241127111821.1649985-1-d.csapak@proxmox.com/