What this guide covers
WatchZit is a free, open-source (AGPLv3) monitoring system for networks and servers. Next to ping, SNMP, web and database checks it has a small agent for Windows and Linux. Since version 1.4 there are two check types that need nothing installed on the machine: Linux over SSH (no agent) and Windows over WinRM (no agent).
What is measured
Both checks read the same values from a machine where nothing is installed:
- CPU load (measured over one second on every run),
- memory in use,
- swap on Linux or the page file on Windows,
- the fullest disk,
- the uptime,
- and on Linux also the load.
The values carry the same names as the agent's values. That is why the built-in triggers for CPU, memory and disks, and forecasts, work on these checks without any extra setup.
Prepare a Linux server (SSH)
WatchZit signs in over SSH and runs one fixed command that only reads /proc, df and uname. Nothing is written on the server.
- User: any user that may sign in over SSH is enough — no root, no sudo. A separate user, for example
monitor, keeps it apart from people's accounts. - Password or private key: either one. The key may be in OpenSSH or PEM format and must not have a passphrase. WatchZit stores the password or key encrypted and never shows it again.
- Port: 22 unless you set another one. The WatchZit server (or the remote collector that runs the check) must be able to reach it.
Host key: trust on first use
The first time the check connects, it accepts the server's host key and saves it. On every run after that the key is checked. When the key changes, the check goes down and names the new key — so a server that was swapped or a connection that was intercepted does not go unnoticed. If the change was expected, for example because the machine was reinstalled, open the check, tick Forget the saved host key and accept the next one and save.
If you do not want to rely on the first connection, enter the fingerprint yourself in the Host key field. On the server, this command shows it:
ssh-keygen -lf /etc/ssh/ssh_host_ed25519_key.pub
The field takes a SHA256 fingerprint like SHA256:…; empty means the first key is accepted.
Prepare a Windows server (WinRM)
On Windows, WatchZit reads WMI over WinRM. Windows Server has WinRM turned on; on other Windows versions turn it on with:
winrm quickconfig
- HTTP, port 5985: WatchZit signs in with NTLMv2 and encrypts every message with the sign-in key, as Windows requires. Nothing travels in clear text, even without a certificate.
- HTTPS, port 5986: needs a WinRM HTTPS listener on the server. TLS protects the messages and the sign-in is bound to the certificate (channel binding). If the listener uses a self-signed certificate, tick Accept certificates that are not trusted (self-signed).
- User: written as
name,DOMAIN\nameorname@domain. It must be an administrator, or a member of Remote Management Users with WMI rights onroot\cimv2. The second way is the one to prefer for a monitoring account. The password is stored encrypted.
Add the check in the console
- Add the machine under Devices with a name and its address, if it is not there yet.
- Open the device page and click Add a check.
- Under Type choose Linux over SSH (no agent) or Windows over WinRM (no agent).
- Fill in User name and Password. For Linux you can paste a Private key instead of the password and, optionally, the Host key. For Windows tick HTTPS (port 5986) if you use the HTTPS listener. Port is optional in both: empty means 22, or 5985 / 5986.
- Set Interval (seconds) and Timeout (ms). A run takes a few seconds, so give it a timeout of at least 15 seconds (
15000); an interval of a minute or more is enough. - Click Add check. After the first run the device page shows the values, and graphs fill from then on.
When you edit the check later, leave the password field empty to keep the stored one.
Remote collectors: a check that runs through a remote collector at a branch office cannot save the host key by itself. Enter the key in the check; until you do, the check shows the key the collector saw, so you can compare and copy it.
Triggers and forecasts
Because the values have the agent's names, the built-in triggers under Problems → Triggers watch these checks as soon as they run:
- High CPU load and Memory almost full — above 90 % for 5 minutes,
- Disk almost full and Disk full — above 90 % and 97 %.
To be warned before a disk is full, add a trigger of the kind Value will reach a threshold (forecast): choose the disk value, a Threshold and Within (days). WatchZit draws a straight line through the last week of the value and opens a problem when that line reaches the threshold within the days you set. The problem tells you the value now, how fast it changes per day and the date it gets there.
Problems go to the notification channels you set up — email, Slack, Microsoft Teams, Discord, Telegram, ntfy, PagerDuty, SMS or webhooks — and dependencies and maintenance windows work for these devices like for any other.
When the agent is the better choice
The agentless checks cover the basics. The WatchZit agent sees much more, so install it wherever you are allowed to:
| Without an agent (SSH / WinRM) | WatchZit agent | |
|---|---|---|
| CPU, memory, swap / page file, uptime | Yes (Linux also load) | Yes |
| Disks | The fullest disk | Every drive |
| Interfaces, services, watched processes | — | Yes |
| Hyper-V machines, Docker containers | — | Yes |
| Windows event logs, systemd journal | — | Errors are sent to Logs |
| Connection | WatchZit connects to the machine (SSH or WinRM) | One outbound connection from the agent to port 8444 |
| On the machine | Nothing installed; a user with sign-in rights | MSI for Windows, deb and rpm for Linux; updates itself from the console |
A good rule: the agent on your own servers, the agentless checks on machines you may not or cannot change.
Also new in 1.4: network path and learned triggers
Network path (traceroute) is a check that traces the way to a device every five minutes, like MTR: for every router on the way its address, its loss and the best, average and worst time (3 probes per hop by default, up to 30 hops). Every run is kept for 30 days; the device page shows the last path, each hop over the last 24 hours and the last ten route changes. The trigger Loss on the network path (over 20 % for 10 minutes) comes with it. Put it on the line to a branch office or a provider to see where loss or delay begins; from a remote collector it traces from that site.
Value unusual for the time (learned) is a trigger that needs no threshold. It learns from the last four weeks what is usual for a value at each hour of the week and opens a problem when the value is far from that — higher, lower or either. Sensitivity is High (more problems), Medium or Low (only big changes). It starts to compare after one week of history. It works on the values of the agentless checks too, for example a CPU load that is normal on Monday morning but not on Sunday night.
Limits
- Only one disk is reported: the fullest one. Interfaces, services, processes and logs need the agent.
- SSH keys with a passphrase are not supported; use a key without one, or a password.
- The Linux check reads
/proc, so it is meant for Linux, not for other Unix systems. - WinRM over HTTPS needs a listener that you set up on the Windows server; the WinRM user must be an administrator or in Remote Management Users with WMI rights.
- A run takes a few seconds; very short intervals are not useful. Through a remote collector the host key has to be entered by hand.
Next steps
- Draw the servers on a network map next to routers and switches — see the guide A modern alternative to MikroTik The Dude.
- Install the agent on the servers where you may, and keep the agentless checks for the rest.
- Add a forecast trigger for the disks that matter most.
See everything WatchZit does on the features page, get it on the download page, or read the full documentation. Questions are welcome in GitHub Discussions; companies can get commercial support from MaxZit.