New: temperature, fans, power supplies and SSD health per server. View hardware health

Try it free

PeakNode shows whether your servers are running. NodeGuard shows who logs in, which ports are open, whether system files have been changed, who is computing on the GPUs and what the web server’s access logs reveal about attack attempts. It uses the agent that is already running and stays with notification: NodeGuard blocks nothing and changes nothing.

Sample data, wording as in the portal; addresses and names are placeholders. NodeGuard displays and reports, it blocks nothing.

What is NodeGuard?

NodeGuard is the part of PeakNode for the security of your servers. The Linux agent of the server monitor supplies SSH logins, open TCP ports, checksums of system files, pending security updates, the package list and an analysis of the Apache and nginx access logs. The portal evaluates the data and reports anything unusual through the server monitor’s channels. NodeGuard displays and reports, it blocks nothing. Included in Premium, Enterprise and On-Premise, switched on per server.

NodeGuard dashboard in the portal: “1 critical item”, nine panels for web attacks, SSH logins, open ports, system files, GPU users, updates, software, vulnerabilities and anomalies, below the history per server over 24 hours.
Screenshot from the portal with sample data (German interface).

NodeGuard in the portal

  • The same agent as for server monitoring, version 4 or later, installed as root. It only sends; there is no remote access.
  • Web attacks only become critical with an indication of success. Routine scans without success appear in the daily summary.
  • Only the Linux agent sends NodeGuard data, and only from servers for which NodeGuard is switched on.

What NodeGuard watches

Nine areas per server. A suspicion is an item for you to check, with evidence and numbers, not a verdict.

SSH logins

NodeGuard shows the failed attempts of the last 15 minutes, the most frequent senders and the last 20 successful logins. You get an alert from 60 failed attempts in 15 minutes, adjustable per server, and on a login from an address NodeGuard has not yet seen on this server. The portal checks senders against public lists without sending any address to third parties.

Attacks on the web server

The agent analyzes the Apache and nginx access logs on the server and detects, among other things, scan tools, searches for .env, .git and backups, exploit attempts and login series. PeakNode receives counters, conspicuous addresses and shortened evidence, never complete logs. An event only becomes critical if an attempt may have succeeded, for example because the server delivered a sensitive file.

Open ports and changed system files

NodeGuard lists every TCP port on which the server accepts connections, with its process. It reports a new port on a non-local interface, as well as a service without authentication of its own, such as Ollama, that is bound so it can be reached from outside. It also reports every change to the user list, sudoers, the SSH configuration, root’s SSH keys, /etc/crontab and root’s crontab.

Software status and known vulnerabilities

The software status lists the running versions of PHP, OpenSSL, nginx, Apache, OpenSSH, Docker, databases, the kernel and more, compared with the end of support per branch and with a list of critical security advisories maintained by Nelpx. The portal matches the installed packages against OSV.dev and the CISA list of known exploited vulnerabilities and names the package version that fixes a vulnerability.

Users on the GPUs

NodeGuard remembers which users run processes on your GPUs and reports when a previously unknown user occupies a card. If a process name matches known mining software, it is flagged as a suspicion. At this point NodeGuard checks nothing beyond the name.

Unusual behavior per server

Anomaly detection learns per server what is usual for GPU utilization, GPU power, CPU, upload and download at each hour, weekdays and weekends separately. After a seven-day learning phase it reports values that deviate clearly over three consecutive reports. Example in the portal’s format: “Anomaly: GPU usage 96%, usual at this time 2% (±3%), since 03:10.”

What NodeGuard items look like

Example items, worded as in the portal

AreaSeverityMessage
SSH loginsCriticalSSH brute force: 312 failed login attempts in 15 minutes from 9 addresses. Most frequent sender: 203.0.113.7 (140).
SSH loginsWarningSSH login from a new address: user root from 198.51.100.77 (password).
Web attacksCriticalshop.example.com: the server delivered /vendor/composer/installed.json to 203.0.113.7 (status 200, 48.2 KB). Please check whether the file may be publicly reachable and replace any credentials it contains.
Web attacksWarningshop.example.com: path scan from 203.0.113.7. So far 114,831 requests, 113,902 of them not found (404), up to 15,616 per minute. Tool according to its user agent: feroxbuster. Example: /.env
Open portsWarningOllama accepts connections on port 11434 (all interfaces). The service has no authentication of its own. The agent cannot see whether a firewall is in front of it.
Open portsWarningNew open port 8443 on all interfaces (java)
System filesWarningSystem file changed: /root/.ssh/authorized_keys
Users on the GPUsCriticalSuspected mining: process xmrig on GPU 0, user www-data. The process name matches known mining software.
Users on the GPUsWarningNew user on GPU: www-data on GPU 0 (python3)
Software statusCriticalSecurity advisory: PHP 8.3.30 (/opt/plesk/php/8.3/bin/php) is affected by php2shell (CVE-2026-109618, CVE-2026-109619). Fixed in version 8.3.36.
VulnerabilitiesCriticalKnown exploited vulnerability CVE-2023-4911 in glibc 2.35-0ubuntu3.1 (CISA KEV). Fixed in package version 2.35-0ubuntu3.4. glibc: buffer overflow in ld.so leading to privilege escalation
AnomaliesWarningAnomaly: GPU usage 97%, usual at this time 3% (±15%), since 03:07.

Addresses, names and version numbers are sample data.

What is checked and when it is reported

For NodeGuard the agent only runs read-only commands; service queries and the access log analysis run with a time limit and the lowest CPU priority. The evaluation happens in the portal.

AreaWhat is checkedWhen it is reported
SSH loginssshd lines from journalctl, otherwise auth.log or securefrom 60 failed attempts in 15 minutes, critical from five times that. Login from a new address or an address on a public list
Web attacksApache and nginx access logs, at most 16 MB and 10 seconds per runWarning on volume or series, for example from 1,500 “not found” from one address in 10 minutes. Critical only with an indication of success
Open portslistening TCP ports with process from ss or netstat, no UDPnew port on a non-local interface. Service without authentication of its own (Ollama, TensorBoard, Docker API without TLS) immediately
System filesSHA-256 checksums of passwd, group, shadow, sudoers, sshd_config, root’s SSH keys, /etc/crontab and root’s crontabon every change
Security updatesnumber of pending updates and reboot notice, Debian and Ubuntu onlyshown in the portal
Software statusversions of the server software with program path and package sourcecomponent affected by a security advisory or no longer maintained
Vulnerability matchingpackage list from dpkg, rpm or apk against OSV.dev and CISA KEVfor a vulnerability from the KEV list, or CVSS 9 and higher, as soon as a fixing package version exists
Users on the GPUsprocesses per GPU with user from nvidia-sminew user on a GPU, process name of known mining software
Anomaly detectionfive metrics against the median and spread of the same hour over 14 daysdeviation over three consecutive reports, at most one notification per metric in six hours
Fail2ban and load spikesFail2ban status with banned addresses. From 90% CPU, the five processes with the highest loadshown in the portal

During the first 24 hours NodeGuard learns which ports, login addresses and GPU users are usual and does not report them. Brute force, meaning series of login attempts, and changed system files are reported immediately.

Set up in three steps

No second agent, no rules on the server. Everything NodeGuard needs comes from the server monitor’s agent. It requires the Premium or Enterprise plan or the on-premises edition.

  1. Install the Linux agent as root

    NodeGuard needs a server monitor with agent version 4 or later, installed as root, and version 9 or later for the web analysis. Only the agent’s user may write to the agent’s directory.

  2. Switch on NodeGuard for the server

    You switch on NodeGuard per server in the server monitor’s settings. Only then does the agent send NodeGuard data from this server. There you also choose the warning threshold for failed SSH attempts and enter up to 50 trusted addresses or networks, such as your office.

  3. Check the notifications

    NodeGuard reports through the server monitor’s channels. “Send test emails” shows in advance what a notification looks like.

Frequently asked
questions about NodeGuard

Continue to:

What is NodeGuard?

NodeGuard is the part of PeakNode for the security of your servers. It shows SSH logins, attack attempts on the web server, open ports, changed system files, known vulnerabilities in installed packages and unusual behavior per server. NodeGuard reports, but does not intervene in the server.

Does NodeGuard cost extra?

No. NodeGuard is included in the Premium and Enterprise plans and in the on-premises edition, at no extra charge. It is not included in the free Proof of Concept plan. It is switched on per server.

Does NodeGuard block attackers, and does it replace Fail2ban?

No. NodeGuard displays and reports; it does not intervene in the server, changes nothing and blocks nothing. It does read whether the blocking tool Fail2ban is running and how many addresses are banned. If no rule there bans web attacks, the portal shows suitable rules to copy, including how to undo them.

Which data does the agent send for NodeGuard?

Only from servers for which NodeGuard is switched on, the agent sends: the sshd lines on logins with user name and source address, the open TCP ports with process name, checksums of the monitored system files, the number of pending updates, the package list and the software status, plus the Fail2ban status and, from 90% CPU utilization, the names of the five processes with the highest load and their owners. From the access logs, only counters, conspicuous addresses and a few shortened request lines of detected attacks leave the server, and for a sensitive file also whether it is in the web directory. The agent does not send file contents, complete logs or command lines. The full list is under security.

How does NodeGuard know which packages are vulnerable?

From public sources: OSV.dev with the distribution’s security data and the CISA list of known exploited vulnerabilities (KEV). Up to 3,000 packages per server are matched, on every package change and every six hours, on Ubuntu, Debian, AlmaLinux, Rocky Linux and Alpine. You get a notification for a vulnerability from the KEV list and for a severity of CVSS 9 and higher, as soon as a fixing package version exists. NodeGuard matches the kernel against the KEV list only.

Does NodeGuard also run on Windows or macOS?

No. Only the Linux agent sends NodeGuard data, and it must run as root, because otherwise the SSH log cannot be read. The agents for Windows and macOS send no NodeGuard data. NodeGuard counts security updates on Debian and Ubuntu only.

What does NodeGuard not see?

Anything that is not in the evaluated sources: contents of POST requests, attacks on the logic of an application, UDP services and the rules of an upstream firewall. Behind a CDN, NodeGuard sees the CDN’s address instead of the attacker’s. A suspicion is an item for you to check, not proof, and NodeGuard does not promise completeness.

How do I detect an SSH brute force attack?

By many failed SSH logins in a short time, often with changing user names. On Linux they are in the sshd log, which you can read with journalctl. NodeGuard counts the failed attempts per 15 minutes, reports a warning from 60 and a critical case from five times that, and the threshold can be set per server between 30 and 300. The notification states the number of attempts, the number of addresses and the most frequent sender.

What is intrusion detection?

Intrusion detection means recognizing attacks or their traces. Host-based intrusion detection evaluates what happens on the server itself: logins, logs, files and open ports. Network-based intrusion detection inspects the traffic on the network. NodeGuard is host-based, works through the agent of server monitoring and does not capture network traffic.

See who is accessing your servers

Create an account, create a server monitor, install the Linux agent as root and switch on NodeGuard for the server. NodeGuard is included from the Premium plan, not in the free Proof of Concept plan.