Why is the pull model bollocks? I’ve been building monitoring for stuff since nagios and zabbix were the new hot tools; and I can’t really imagine preferring the oldschool ways vs the pretty much industry standard of promstack these days…
Prometheus as a time series DB is great, I even like its QL. What I don't like is pull. Sure there is agent mode or telegraf/grafana agent. But the idea that I need to hold my state and wait for Prometheus to collect it is utterly stupid. The biggest annoyance is that I need to have a webserver somewhere, with a single god instance(s) that can reach out and touch it.
Great if you have just one network, but a bollock ache if you have any kind of network isolation.
This means that we are using influxdb and its shitty flux QL (I know we could upgrade, but thats hard)
Were all kubernetes these days so i guess i didn’t think about it a lot in recent years.