Basic Prometheus, with vanilla storage engine, is quite easy to set up (IMO). On a Debian system it takes about 15-30 minutes to install and do a basic enough

Basic Prometheus, with vanilla storage engine, is quite easy to set up (IMO). On a Debian system it takes about 15-30 minutes to install and do a basic enough configuration to get it running. Another 30 minutes to get a Grafana instance sitting in front of it - and then it's incremental to get the node_exporter (also packaged up) out to your systems. The two time consuming bits are a) customising the metrics that you can scrape, though there's plenty of templates out there, and if you're not trying to do any fancy application-layer metric scraping you're probably going to be happy out of the box. And b) evaluating and then setting up longer / larger storage engine on the backend - but if you've committed to Prometheus after doing your low-effort PoC, then this is a simple cost benefit evaluation. Metrics-as-a-service are an interesting proposition, but I'm very cautious about them > Speaking of that I'm still missing the easy middle ground of monitoring which is modern, easy to use and works for a higher number of machines (and even something like Docker Swarm). What's modern mean in this context, and just how easy do you want? In my experience monitoring systems are relatively easy to deploy, unless you're huge. The difficult bit is working out (with platform owners) everything they need to monitor, what their key health indicators are, and what thresholds to apply.