Servers rarely die out of nowhere at 2 a.m. There is usually a warning: a disk filling up over days, memory draining hour by hour, a service quietly restarting itself in a loop. Server monitoring exists so you see those signals early, instead of hearing about the outage from an angry customer.
What is server monitoring?
Server monitoring is the practice of collecting metrics from your machines on a short cycle, comparing them against safe thresholds and paging whoever is on call when something crosses the line. A complete setup watches from two directions:
- From the inside: an agent running on the server reads CPU, RAM, disk usage, load, service status and container health.
- From the outside: HTTP, TCP, ping and SSL checks call your website or API exactly the way a real user would.
Drop either side and you get a blind spot. A website can return 200 while the disk sits at 2 percent free space. A server can be perfectly healthy while the domain points somewhere wrong.
Why do small teams need monitoring more, not less?
Large companies have an ops team on rotation around the clock. In a small business or agency, one person often owns both the code and the infrastructure, and incidents have a talent for arriving while that person sleeps. Three reasons the 3-minute setup pays for itself:
- Every minute of downtime costs money and trust. A customer who cannot place an order rarely comes back to try again.
- Incidents leave a trail before they happen. A 24-hour chart showing RAM climbing steadily since noon lets you fix things at 5 p.m. instead of 2 a.m.
- You need proof of uptime. A public status page shows customers exactly how transparent your operation is.
Which metrics should you actually watch?
You do not need the hundreds of metrics that enterprise platforms ship. For most web and application servers, this short table catches the vast majority of real incidents:
| Metric | Suggested threshold | Typical incident |
|---|---|---|
| CPU | Above 85% for 5 minutes | Runaway loop, attack, undersized machine |
| RAM | Above 90% | Memory leak, misconfiguration |
| Disk | Above 85% used | Logs ballooning, backups filling the disk |
| Services | Process stopped | Crash, OOM kill |
| HTTP check | Wrong status code or missing keyword | Broken deploy, expired SSL certificate |
Just as important as the metrics is where the alert lands. An alert sitting on a dashboard nobody opens might as well not exist. AgentWatch delivers alerts straight to Zalo or email, the channels people in Vietnam genuinely check every day, with Telegram support coming soon.
How is agent-based monitoring different from an uptime checker?
A pure uptime checker such as UptimeRobot only calls your website from the outside. It answers “is the site alive” but never explains why the site died, and it has no idea your disk is about to fill up.
An agent running inside the server supplies the missing half: machine resources, the state of each service and Docker container health assessed through three layers, combining Docker HEALTHCHECK, an internal HTTP probe and process status. AgentWatch merges both views into a single dashboard, with check intervals as short as 30 seconds on the Team and Business plans. If external checks are your main concern, the uptime monitoring page covers that side in depth.
How do you get started with AgentWatch?
The whole process takes about 3 minutes and no credit card:
- Create an account from the AgentWatch homepage with just an email address.
- Run one install command on your server. The agent is a static binary for linux amd64 and arm64. It registers itself and starts reporting right away.
- Connect an alert channel, Zalo or email, and add HTTP checks for your websites if you want the outside view too.
From that point on, metrics flow into the dashboard, history is retained from 7 up to 365 days depending on your plan, and you can publish a status page on its own URL to share uptime with your customers.
How much should server monitoring cost?
The AgentWatch Free plan covers 2 servers and 10 service checks, enough for one website plus a few internal APIs. Paid plans are priced in Vietnamese dong starting at 49,000 VND per month for 5 servers and 30 checks at a 1-minute interval, then Team at 149,000 VND with 15 servers and Business at 399,000 VND with 50 servers, both at 30-second intervals. The differences come down to server count, check interval and history retention. Full details are on the pricing table.
During the beta, every new account runs on the Business plan for free, which makes right now the cheapest possible moment to try it: create your account and let your servers start telling you their story.