Server uptime monitoring without an agent
How to monitor a server's uptime from outside: ping for the host, TCP checks for each service, HTTP for web apps, and a heartbeat for what only the server can see.
· 5 min read · By the Spot Downtime team
“Is the server up?” is really several questions: does the machine answer, is each service on it accepting connections, does the app behind them work, and are its jobs still running? You can answer all four from outside, with nothing installed on the server, by stacking a few simple checks.
Want CPU, memory and disk too?
The four layers
| Layer | Check | Catches | Plan |
|---|---|---|---|
| The machine | Ping | Host off, network or routing down | Free |
| Each service | TCP port | Database, SSH, mail or Redis stopped or firewalled | Free |
| The app | Website or API | Errors, slow responses, wrong content | Free |
| Inside the server | Heartbeat | Cron jobs, backups, disk full, outbound network | Starter and up |
1. Ping the host
A Ping monitor sends three ICMP pings per check and records the round-trip time. It's the quickest signal that the machine itself is gone. Two caveats: many cloud firewalls drop ICMP by default (allow it from anywhere, or the check will always fail), and a host that answers ping can still have every service dead. That's what the next layer is for. Step-by-step ping setup.
2. A TCP check per service
A TCP port monitor opens a connection to a port and closes it. If the connection is refused or times out, the service is down even though the machine is up. Add one for each service that matters:
22SSH, so you know you can still get in5432PostgreSQL,3306MySQL,6379Redis,27017MongoDB25,587or993for a mail server- Any game server, queue or custom service on its own port
Databases usually shouldn't be open to the internet. If yours is firewalled, allow only our check addresses, or skip it and let the app check (layer 3) prove the database works. Step-by-step TCP setup. For UDP services like game servers and NTP, see UDP monitoring.
3. Check the app, not just the port
Port 443 accepting connections doesn't mean the site works. A Website monitor with a Page must contain keyword, or an API monitor with JSON assertions on a health endpoint, proves the whole stack answers correctly. Point it at a route that touches the database, so a database failure shows up here too.
4. A heartbeat for what only the server sees
Some failures are invisible from outside: a backup that stopped, a disk filling up, a server that can't reach the internet. Turn them around. Have the server ping a Heartbeat URL on a schedule, and get alerted when the pings stop. This crontab line pings every 5 minutes, but only while the root disk is under 90% full:
*/5 * * * * [ "$(df --output=pcent / | tail -1 | tr -dc 0-9)" -lt 90 ] && curl -fsS -m 10 --retry 3 https://YOUR-HEARTBEAT-URL > /dev/nullThe same pattern works for any condition you can test in a shell. More heartbeat examples.
Avoid false alarms
- Confirm before alerting. Spot Downtime retries a failed check from a second region before opening an incident, so a single dropped packet doesn't wake anyone.
- Check from where your users are. Pick regions near your audience in Check from.
- Schedule maintenance. Pause alerts during planned reboots and upgrades.
Network devices too
Get new guides by email
Practical posts on monitoring, outages, error fixes and on-call, when they're published. No spam.
Keep reading
- How-to · Monitoring · APIsHow to configure gRPC health checkingLearn how to implement and configure gRPC health checking for reliable service discovery and load balancing: server status, Kubernetes probes, client-side health checks, Envoy, cloud load balancers and Consul.November 7, 2026 · 7 min read
- How-to · Servers · MonitoringHow to monitor CPU, memory and disk usage on a Linux serverCheck CPU, memory and disk on Linux with top, free and df, learn which numbers actually matter, pick sensible alert limits, and get alerted before a server runs out.November 4, 2026 · 8 min read