Server monitoring for a small org

If you run a handful of servers - a VPS, a box in the office, something in a closet - you don't lack monitoring options, you lack a way to choose between them. This is the map we wish we'd had. The tools below are grouped by the job they do, not by feature lists, because that's how the choice actually goes: Netdata and Prometheus want to show you everything. healthchecks.io and Uptime Kuma answer liveness questions - did the job run, does the endpoint answer. Monit supervises processes. And microSRE, the tool we make, tries to do one narrow thing: tell you when a server needs you, fast, and stay out of the way the rest of the time.

Written by the microSRE author, who is obviously biased. Every "pick them if" below is meant seriously - if another tool fits your job better, take it. Spotted something stale or unfair? hello@microsre.io.

First: when microSRE is the wrong tool

Concessions up front, because that's the useful order:

You want dashboards or historical graphs - take Netdata or Grafana.
You need per-second or per-application metrics: query latency, request rates, JVM stats. That's APM territory; microSRE won't get there.
You run Kubernetes-first fleets with autodiscovery and pod-level views.
You need automatic remediation - restarts, failover. microSRE tells you; systemd, monit, or your orchestrator do it.
You need compliance-grade long-term archival across years of full-resolution data.

Tool	Best at	Where microSRE fits instead
Netdata	Per-second dashboards, hundreds of app collectors	A few dozen health signals, minutes not seconds, tens of MB of RAM
Prometheus + Grafana	The standard metrics platform, scales to thousands	One binary, one config, no TSDB to run
healthchecks.io	Dead man's switches, done perfectly	Heartbeats plus the machine's actual health
Uptime Kuma	Self-hosted uptime + pretty status pages	Reachability isn't disk, memory, or pressure
Monit	Process supervision, restarts what dies	Modern signals, chat-native alerts
Datadog-class SaaS	APM, tracing, enterprise everything	Nothing leaves your machine by default

Netdata

The closest in spirit: an open-source agent that collects locally and can skip the cloud.

Where Netdata wins

Granularity. Per-second collection and per-app collectors - databases, web servers, containers, systemd services - where microSRE reads only the few dozen signals the kernel exposes in /proc.
Dashboards. If your troubleshooting starts from a graph, Netdata's local dashboards are excellent. microSRE deliberately has none.
Kubernetes. Helm charts, autodiscovery, container-aware views. microSRE is a per-host tool.
Retention. Longer local history out of the box; microSRE keeps no metric history at all - it evaluates the current check against your thresholds. If you want trends, that's Netdata or Prometheus.

Where microSRE wins

Footprint. Netdata's agent collects everything, all the time - RAM use reflects that. microSRE checks once a minute at a few tens of MB, and publishes its own cost in the metrics.
Where it runs. microSRE reads /proc directly - no format-string parsing, no dependencies - so it works in rescue shells, minimal containers, and on a 512MB Pi Zero.
Nothing to operate. Alerts go straight from your machine to Slack, Discord, or Telegram. No cloud account, no notification-agent config, no room full of charts to maintain.
Pressure and OOM. PSI stall signals and OOM-kill events arrive in chat - no dashboard to configure first, nothing else to run.

Pick Netdata if you want to watch graphs and drill into per-application behavior. Pick microSRE if you want to be told the moment it's time to log in - and not look at a dashboard the rest of the year.

Prometheus + node_exporter + Grafana + Alertmanager

The industry standard. A stack of moving parts, not a program.

Where the stack wins

Scale and queries. Thousands of hosts, years of history, PromQL - nothing else matches it for "show me CPU of fleet X versus last quarter".
Ecosystem. Every exporter imaginable, dashboards for everything, battle-tested alerting.
Ownership. You run it, you own the data, no vendor involved at all.

Where microSRE wins

One Python zipapp, one TOML file. No TSDB sizing, no retention policy, no Alertmanager routing trees, no Grafana to keep fed.
No central server to run, feed, or keep reachable. Every host checks itself and alerts on its own - add a hundred machines and you've added zero infrastructure.
Alerting without the ceremony. Threshold in config, message in chat. The whole alert path is one file you can read top to bottom.

Pick the Prometheus stack if you have a platform team, real scale, or questions only a query language can answer. Pick microSRE if you have servers and a chat room, and the question is "are they okay?"

healthchecks.io

Dead man's switches, perfected. Open source, self-hostable, generous free tier.

Where healthchecks.io wins

Focus. "Did the job run?" is its entire world, and it does that one thing beautifully - grace periods, schedules, per-check channels.
Maturity and integrations. A long list of ready notification integrations, a proven hosted service, and you can self-host it.
Zero machine footprint on the monitored side. A curl in your cron job is all it takes.

Where microSRE wins

Heartbeats are one knob, not the whole product. The same config that watches cron silence also watches memory, disk, pressure, and log error rates.
Log-line heartbeats. Match a pattern in a log file, not just HTTP pings - batch jobs that only write files are covered.
Local alerts, optional fleet. Alerts are generated on your machine; the hosted dead man's switch is opt-in, not the product.

Pick healthchecks.io if all you need is "did the job run?" - genuinely, no hard feelings. Pick microSRE if the job's silence and the box's health are one question. The two compose fine; nobody stops you running both.

Uptime Kuma

Self-hosted uptime monitoring with status pages and a big notification menu.

Where Kuma wins

Status pages. Public incident communication built in - microSRE has nothing like it.
Monitor variety. HTTP, TCP, ping, DNS, push, game servers - a long list of ways to ask "does it answer?"
Nice UI, self-hosted, free. A pleasant dashboard you'll actually open.

Where microSRE wins

Reachability isn't health. Kuma tells you the site answers; microSRE tells you the disk is 96% full, swap is thrashing, and the OOM killer just fired - before the site stops answering.
Runs on the box itself. No separate monitor host needed; the agent lives on the machine it watches.

Pick Kuma if you need public status pages or many protocol checks from one place. Pick microSRE if the 3am question is "why is it slow?" rather than "is it up?"

Monit

The veteran. In every distro's repos for two decades, and it restarts what dies.

Where Monit wins

Supervision. Monit restarts a crashed service. microSRE deliberately doesn't - it tells you, then you decide.
Ubiquity and age. It's everywhere, it's tiny, and its config language is well understood by twenty years of admin answers online.

Where microSRE wins

Modern signals. PSI pressure, OOM kills, log error rates, per-container docker logs - out of the box, not through a custom check program.
Chat-native alerts. Slack, Discord, and Telegram webhooks auto-detected from the URL - no glue scripts to write.
Optional fleet view. One place to see all your machines, if you want it.

Pick monit if you want supervision with restart actions and email alerts. Plenty of boxes run both happily - monit restarts, microSRE explains.

Datadog, New Relic, Grafana Cloud, and friends

The hosted platforms. Different league, different price.

APM, distributed tracing, RUM, log pipelines, per-host pricing that scales with ambition. If you have an on-call rotation and a monitoring line item in the budget, you already know these - and microSRE won't pretend to play in that league. The one-sentence contrast: those platforms move your data to them and charge you to look at it; microSRE sends nothing anywhere by default, and the hosted fleet view is summaries only, priced for fleets of pets, not herds of cattle.