So, I switched my main server from a dual core atom + 4 GB of memory and 900 GB of disk, to an @OpenBSDAms VM with 1 Xeon core + 1 GB of memory and 50 GB of disk
I had to change a few things...
I thought munin was a lightweight monitoring solution, but not really.
- getting data requires running many commands for each value
- gathering data requires a lot of process on the server and take a lot of time and ~40 MB of memory per process
- rendering the graphs costs a lot of CPU and memory every 5 minutes
- using fast cgi for the graphs is a huge pain
I switched to collectd for collecting data, it uses nearly no CPU and 1 MB of memory :flan_aww:
sending data to a victoria metrics server using the graphite API, it doesn't use much data to transfer things as they can be transferred in bulk
victoria metrics uses like 20 MB of memory to handle 3 systems sending metrics, and almost no CPU
and data can be visualized with grafana which is the worse part of the stack, it draws 52 MB of memory and uses nearly no CPU.
The most boring thing was to create the dashboards on grafana, and that I can't find how to automatically detect the different instances in the datasets...
In the end, this stack is a lot more lightweight and scalable than munin :flan_disappointed:
@solene c'est fini le serveur à la maison par VPN ?