metric server performance issues

[schoda]

Hi,

is there a way to configure what data is send to the metric server? I connected to our influxdb via cli and did an "show series" on the Proxmox database and if i counted corretly we have 2000+. There are a lot of network interfaces shown that Proxmox doesn't use at all. The influxdb server currently got 8 cores and 32GB RAM and the grafana proxmox plugin loads for ages when you increase the time range to e.g. 30 days (or even 7 days).

My proxmox cluster currently got 7 nodes and is running 170 VMs across them. It will get more and more in the future and i'm worried about the performance of the metric server getting worse and worse. I mean sure i can look at the last hour of data or the last 12hours of data and it loads in an acceptable time but for higher time ranges it just sucks.

Any advice besides throwing hardware at influxdb? Would graphite be an better option? (It isn't according to https://www.influxdata.com/blog/influxdb-outperforms-graphite-in-time-series-data-metrics-benchmark/)

Thanks & Cheers,
Daniel

 

[wolfgang]

Hi,

If you want to control the monitored parameter I would recommend an external metric server with a client in PVE.
Proxmox VE is not designed to make finely graded monitoring.
There are good OpenSource projects out there that are capable of this use case.