I recently setup the Influxdb and Grafana to get some better graphs. The setup works fine as is however the data only updates as fast as 5 seconds. I have changed Grafana to check for updates every 1 second but had no success with the faster updates. Do I need to change something in Proxmox or Influx?
Honestly its not much of need more of a desire if anything. Comparing the graphs to the native built in graphs Proxmox offers I could see a difference in accuracy which wasn't appealing to me. Although your response has me confused; If Polling stats at a faster rate isn't cheap, why does Proxmox update faster then the metric server?
The "PVE polls current performance/usage/.. data" and "PVE sends that data out to configured metric servers" is tightly coupled.
The same daemon-process is doing it in a loop, so poll+send is always done at the same Interval. The same daemon also broadcasts the data for the graphs to all the cluster. Or which specific graph do you mean?
The same daemon-process is doing it in a loop, so poll+send is always done at the same Interval. The same daemon also broadcasts the data for the graphs to all the cluster. Or which specific graph do you mean?
After writing my original reply I realized I could answer all the questions I came up with. To answer your question, I was referring to the CPU Usage graph on the summary page of the PVE and the Summary of all the stats in the first box of the summary page. ( See Image For Better Clarification )
At this point I am just more curious how harsh of increase in resources would it be to update the metrics faster and If its possible in the future, an option to define the rate at which the external metric server sends new data.
At this point I am just more curious how harsh of increase in resources would it be to update the metrics faster and If its possible in the future, an option to define the rate at which the external metric server sends new data.
It differs a bit by the type of metric. Memory and CPU metrics, no problem, those are relatively cheap to gather. Everything doing IO, maybe even through network (NFS, ..), relatively expensive.
also, the rrd graph on the summary page also only has a 10s update interval and the other stats get only pulled once per second when you have the webpage open and only for that vm
if one would query those for all vms/ct (potentially hundreds/thousands) this would get costly quick
if one would query those for all vms/ct (potentially hundreds/thousands) this would get costly quick