Skip to Content

Open source server monitoring, no subscription

Prometheus, Grafana and Loki: what the owner sees, what the IT person tunes, and the lessons of a real outage.
November 15, 2026 by
Open source server monitoring, no subscription

Open source server monitoring: find out before your customers do

Field: IT operations, monitoring, business continuity

For: SME owners who host their own tools, and their IT person

Experience: the monitoring that watches over my own servers, day and night

A server that is running is not necessarily a server that is healthy. Open source server monitoring, with Prometheus, Grafana and Loki, tells you that a site is down, a disk is filling up or a backup did not run, with no per-machine or per-volume subscription. Here is what the business owner sees, what the IT person tunes, and what I learned running it.

Background

When an SME brings its tools in-house, as I describe in the map of open source building blocks, it also takes back a responsibility: knowing that everything works. Online monitoring services exist, but they are usually billed per machine or per volume of data sent, and they export the inner life of your IT to a third party. The same functions exist as free software, proven by thousands of companies, and can run on your own servers.

The problem: the silent failure

Serious outages are rarely spectacular. They start silently:

  • a disk that slowly fills up until the day the database refuses to write;
  • a certificate that expires on a Sunday, and a website that shows your customers a security warning;
  • a backup that has been failing every night for three weeks, without anyone knowing;
  • a site that goes down at two in the morning and is discovered when the office opens.

Without monitoring, it is the customer who warns you. With well-tuned monitoring, you warn the customer, or better still, nobody notices.

The solution: four free building blocks

1. Measure: Prometheus

Prometheus regularly queries each server and each application, and keeps the history of the measurements: processor, memory, disk, network, service status. It is the numerical memory of your IT, which lets you see a trend before it becomes an outage.

2. Centralise the logs: Loki

Every piece of software writes what it does in a log. Loki gathers them all in one place, so you can find in seconds what happened at 3:12 a.m. on a given service, instead of searching machine by machine.

3. Display: Grafana

Grafana turns these measurements into readable dashboards. I always build two levels: one dashboard for the business owner, a single page in green or red; and technical dashboards for the IT person, where every curve can be read in detail.

4. Alert, just enough

Alert rules turn a measurement into a message: in the team chat, by email or on the phone. And an external probe tests your websites the way a visitor would, from the outside: if the page stops responding, you know within minutes.

What the owner sees, what the IT person tunes

  • The owner sees the essentials at a glance: the sites respond, the last backup is from last night, there is space left on the disks, the certificates are valid. And they receive an alert only when a decision or an action is needed.
  • The IT person tunes thresholds, delays before alerting, recipients and how long history is kept, and has the curves and logs to understand the cause of an incident rather than guess at one.

What I learned running it

Installing these tools takes a day. Tuning them well takes months. Three lessons, paid for on my own servers:

  • A drowned alert is useless. In September, my ERP stopped after an update of the container engine. The alert did go out within five minutes. But when several sites went down at the same time, the message arrived empty, reduced to its title, among repetitive and harmless alerts. Nothing distinguished a real outage from a false positive: it took hours to react. Since then, every alert is tested in the worst case, and an alert that rings for nothing is fixed the same day.
  • Monitor the result, not the process. One day, a change at a provider made my backups fail, while the service involved showed as "active". What matters is the age of the last successful backup, not the status of a program.
  • Check what you measure. An audit counted thousands of blocked connections per day as so many attacks. Broken down, most of it was normal local network noise. Thresholds set on that noise would have produced useless alerts, and therefore ignored ones.

Under the hood (for your IT person)

  • Collection: node_exporter for the system, cAdvisor for containers, an Alloy agent pushing logs to Loki. The blackbox_exporter probes the websites over HTTP from the outside.
  • Custom metrics: textfile collectors publish the age of the last backup, certificate expiry and critical disk space. That is often where the real value lies.
  • Exposure: no monitoring port open on the internet; everything goes through the VPN. A new server adds its agents and reports to the same Grafana.
  • Alert templates: test the rendering when several targets fail together; common annotations can then be empty.
  • Memory: for a container, occupied memory is not consumed memory (cache is counted in it). Watch anonymous memory and limit overruns, not the displayed total.
  • Licences: Prometheus is under Apache 2.0; Grafana and Loki are under AGPLv3, with no constraint for internal use.

What I take away

Monitoring is not a screen full of curves. It is a promise: to be warned in time, and only when it is useful. Free software provides everything needed to keep it, without a subscription; the work is in the tuning, and in the discipline of fixing every false alert. It is also what gives confidence in the two other pillars: backups that actually restore and a server you know how to rebuild.

Going further

I can install this monitoring on your servers, tune the alerts with you and hand over a readable dashboard. See Your tools and your data, or book a 20-minute call.

The step-by-step guide

If your IT person prefers to do it themselves, my guide no. 8, monitoring and observability (CHF 39), takes them through it step by step: metrics, logs and traces brought together in Grafana dashboards, for one server or several. It is part of a complete stack, documented and tested in production, gathered in the Guides Pack (CHF 149).

Sources

Tags: server monitoring, observability, Prometheus, Grafana, Loki, alerting, open source, SME

Open source software for SMEs: the building blocks
ERP, files, passwords, remote access, backups, monitoring, local AI: what an SME can keep in-house.