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

Try it free
Uptime monitoring in the PeakNode portal: header “1 uptime monitor down” with 7 of 8 reachable, 99.98% uptime and 133.93 ms average response time, below eight monitors with uptime, response time and availability over the last 24 hours.

PeakNode calls your website, your API or your service from outside at fixed intervals, just as a user would. If the target does not respond or returns an error instead of the right page, you get a notification by email, Teams or Slack. Nothing needs to be installed on the target system.

Screenshot from the portal with sample data (German interface).

What is Uptime monitoring?

Uptime monitoring with PeakNode checks from outside at fixed intervals whether a website, an API or a service is reachable and responds correctly. An HTTP(S) monitor compares the status code, the number that signals success or failure, as well as the body and headers of the response with what you expect. Ping and port monitors check reachability. If a check fails, a second one follows shortly afterwards. If that one fails too, PeakNode opens an incident and notifies the monitor’s channels.

Section of the dashboard in the PeakNode portal: availability of six services with response time and history over 24 hours, next to it the NodeGuard panel and the “Needs attention” list with one unreachable service and one expiring SSL certificate.
Screenshot from the portal with sample data (German interface).

Availability in the portal

  • Check intervals from 1 minute to 1 day, checked from two check locations. Your plan sets which intervals are available.
  • The free Proof of Concept plan includes 1 uptime monitor for testing.
  • Only publicly reachable targets are checked. For services in the internal network there are heartbeats or the on-premises edition.

What an uptime monitor checks for you

A service that is reachable but delivers an error page is not working for your users. That is why you define for each monitor what the correct response looks like.

The response counts, not just reachability

An HTTP(S) monitor only counts as online if the status code matches (default 200, several values possible). You can also define a text that must appear in the response, and headers with exactly the expected value. The text is searched exactly as you enter it, and case matters; regular expressions and JSON paths are not supported.

Endpoints with authentication too

Request method (HEAD, GET, POST, PUT, PATCH), custom headers, basic auth and request body can be set per monitor, so that an endpoint behind a bearer token can be checked as well. The monitor follows redirects if you want, at most five.

An invalid certificate gets noticed

For https targets the check fails if the certificate is invalid. The option “Verify secure connection” is switched on when you create a monitor. The approaching expiry is reported in advance by the monitoring of domains and SSL.

Two check locations against false alarms

For each monitor you choose the locations it is checked from. Every check runs from one of them. If it fails, the recheck comes from the other location where possible, so that a disruption along the way does not count as an outage of your service.

Response time and history

Every check stores the result, status code, response time and location. The detail page shows uptime, downtime, incidents and the average response time in the selected period, also per location. Failed checks are not included in the average response time.

Every outage with a reason

An outage becomes an incident with start, end, duration and the number of failed checks. The alert states the reason in plain language, such as “Connection refused”. When the check succeeds again, the incident closes and the all-clear goes out through the same channels, see alerts and escalation.

What the portal shows when a check fails

Messages of an HTTP(S) monitor

What happenedStatus codeMessage in the portal
The server responds with an error503Response status code does not match the expected one.
The response arrives, but the expected text is missing200Response body does not match the expected one.
No response within the timeout-Timed out after 10 seconds, no response received
The certificate cannot be verified-TLS certificate could not be verified
The web server does not accept the connection-Connection refused

Example with the wording as in the portal. The same reason appears in the check, in the incident and in the alert.

Which check type suits which service

What matters is what “running” means for your service: the server responds, the port accepts connections, or the application returns the correct response.

Check typeWhat is checkedTypical use
HTTP(S)Request to the URL with the selected method. Successful with the expected status code, optionally also with the expected text in the response and the expected headers, and for https with a valid certificate.Websites, APIs and web services, inference endpoints such as /v1/models or /api/tags
Ping (ICMP)One ICMP echo to the host or IP address. Successful if a reply arrives within the timeout.Reachability of a server, independent of a specific service
Host / portTCP connection to the specified port. Successful if the connection is established within the timeout. No protocol is spoken over it.Databases, mail servers and other services on a port, for example SMTP on port 25
“AI inference endpoint” templateHTTP(S) monitor with GET on the model list. Successful only with status 200 within the response time limit and the expected model name in the response.OpenAI-compatible servers such as vLLM, and Ollama. You check Triton with a regular HTTP(S) monitor on the ready route

Check intervals: 1, 3, 5, 10 and 30 minutes, 1, 6 and 12 hours, 1 day. Request timeouts: 1, 2, 3, 5, 10 and 25 seconds. The monitor requests one address. It does not execute JavaScript and does not go through multi-step flows such as a login. The template does not start any inference and does not consume any tokens.

Set up in three steps

The service only needs to be reachable from the internet.

  1. Create a monitor

    Under “Uptime monitors”, go to “Create uptime monitor”, choose the check protocol and enter the target, for example https://api.example.com/health. Set the check interval and request timeout.

  2. Define the correct response

    In the “Custom response” section, enter the status code and, if needed, a text that must appear in the response. The section only appears if a method other than HEAD is selected under “Custom request”, for example GET, because HEAD returns no content. For protected endpoints, add the Authorization header there.

    Response status code(s): 200
    Response body: "status":"ok"
  3. Choose locations and channels

    Under “Advanced settings”, select the check locations, tick the notification channels and save. Until the first check, the monitor shows “Pending first check”.

What is uptime monitoring?

Uptime monitoring checks from outside at fixed intervals whether a website, an API or a service is reachable and responds as expected. PeakNode does this via HTTP(S), ping or TCP port. If the response is missing or wrong, an incident is opened and you are notified.

What does uptime monitoring with PeakNode cost?

The Proof of Concept plan is free and includes 1 uptime monitor for testing. Premium at €149 per month includes 50, Enterprise at €490 per month 250. Data is kept for 14 days on the free plan and for 1 year on Premium and Enterprise.

Do I get an alert for every brief glitch?

No. If a check fails, PeakNode waits briefly and checks a second time, from a different location where possible. If the recheck succeeds, the monitor stays online. Only if it fails as well is an incident with an alert opened. Uptime monitors have no setting “alert only after several intervals”.

Can PeakNode monitor a service in the internal network?

Not with an uptime monitor in the cloud edition: the target must be publicly reachable, and private networks, loopback and link-local addresses are rejected. For internal services, a heartbeat that a script sends after a successful local check is a good fit. The on-premises edition checks internal addresses directly.

How do I monitor vLLM, Ollama or Triton?

For OpenAI-compatible servers such as vLLM and for Ollama there is the “AI inference endpoint” template: it queries the model list and checks whether the expected model name appears in the response. You check Triton with a regular HTTP(S) monitor. The page inference servers lists the right route for each server.

Which data does PeakNode store per check?

Time, result, status code, response time and check location. If the content check fails, the server’s response is stored as well (at most 1 MiB), so that you can see what it returned instead of the expected text.

How can I measure uptime?

As the share of successful checks among all checks in a period. PeakNode checks from outside at the chosen interval and shows uptime, downtime and incidents per monitor, and for a month also in the SLA report. The shorter the interval, the more precisely the measurement captures short outages.

What is the difference between uptime monitoring and server monitoring?

Uptime monitoring checks from outside whether a service responds, without installing anything on the target system. Server monitoring measures with an agent on the server how it is doing: CPU, RAM, drives, network and GPUs. The two complement each other and run in the same PeakNode portal.

Check your first service from outside

Create an account, enter the address of the service, choose channels. The Proof of Concept plan is free and includes 1 uptime monitor for testing.