Uptime checks (HTTP, keyword, ping, port)
How uptime checks work, what counts as down, and every setting.
Kinds of uptime check
- HTTP requests a URL and checks the status code and response time. Redirects are followed by default.
- Keyword is an HTTP check that also needs a word or phrase to be on the page (or not on it). It reads up to the first 1 MB of the page.
- Ping checks the host is reachable. Many servers block ICMP ping, so if it's blocked we try connecting on port 443, then 80.
- Port (TCP) connects to a host and port, like
mail.example.com:587.
HTTP settings
| Setting | Default | What it does |
|---|---|---|
| Method | GET | The HTTP method. HEAD is lighter but some servers answer it differently. |
| Expected status codes | 200-399 | Codes that count as up. Use ranges and lists, like 200-299, 301. |
| Keyword / keyword absent | none | Text that must be on the page, or must not be. |
| Alert when slower than | off | Responses slower than this, for as long as you choose, mark the site degraded. See slow responses. |
| Follow redirects | on | Follow redirects and judge the final page. |
| Headers | none | Extra request headers, for example to get past a staging password. |
| Interval | your plan's shortest | How often to check. |
When is a site down?
A check fails when the server doesn't answer within the time limit, the connection or TLS handshake fails, the status code isn't one you expect, or the keyword rule isn't met. One failure only makes the monitor suspect: it's re-checked right away from other regions (or, on a single-region plan, from the same region). It's marked down, an incident opens and you're alerted only when the failures are confirmed. It's marked up again after two successful checks in a row, and you get a recovery alert with the total downtime.
Slow responses
HTTP, ping and port checks can alert when they get slow. Choose a limit, like 2 seconds, and how long it must last: a number of checks in a row, or the average over the last few minutes. Once the check has a few days of history, the form suggests a limit from its usual response times. A slow site is degraded, not down: it shows amber, linked status page components show “degraded performance”, and it doesn't count against your uptime. It's back to normal once responses are under 90% of the limit for as long, so a site hovering at the limit doesn't flip back and forth. Slow alerts go to channels whose rules include “Slow or degraded”.
Regions
Each monitor is checked from as many regions as your plan includes, taking turns. The more regions, the faster a real outage is confirmed and the less likely a local network problem sets off an alert.
How uptime is calculated
Uptime is the share of checks that passed in the period. Checks during a maintenance window are left out.
Still stuck? Contact us and a person will reply.