Skip to content

Uptime checks

Observability → Uptime checks the public URLs your users depend on - a storefront, an API’s health endpoint - on an interval, from the Seylo server. When a check is down it raises an alert, which reaches people through your notification channels and routes like any other alert. When it passes again, the alert resolves.

Uptime checks do not need the agent or the observability chart: they run from the Seylo server against public addresses, and the page header says where that server is.

Action Role
See checks and their results viewer
Create, edit, pause and delete checks; import them releaser
Turn the Uptime checks detector off or on admin
  1. Open Observability → Uptime and choose New check.

  2. Fill in the form:

    Field What it does Default
    Name How the check, and its alert, are named
    URL A public http or https address. Private and internal addresses are refused, and so is a user name or password in the URL
    Every How often it is checked: 1, 2, 5, 10, 15 or 30 minutes, or 1 hour 5 min
    Timeout How long to wait for an answer: 5, 10, 20 or 30 seconds 10s
    Up when status is The status codes that count as up, for example 200 or 200-299,301 2xx and 3xx
    Alert severity critical, warning or info: the level its alert is raised at. Routes and silences match it like any alert’s critical
    Down after How many failed checks in a row make it down and raise an alert, 1 to 10. Passing once resolves it 2 failed checks
    Body must contain Optional text the answer must contain, looked for in its first 64 KiB. The body is never stored
    Enabled Off pauses the check On
  3. Choose Save check. The first check runs within a minute of saving.

An organisation can have up to 50 checks. Results are kept for 30 days; the page header says so.

What a check does: one GET request to the URL. It does not follow redirects - a redirect is reported as its own 3xx status - and it does not send credentials, custom headers or a body. It checks every address the name resolves to and refuses one that is not public, at every run, so an uptime check cannot be used to reach inside a private network.

Each row shows the check’s URL, how it alerts (for example Alerts at critical after 2 failed checks in a row), its last result (up, down, paused or not checked yet), its uptime over the last 24 hours and 7 days, its p95 latency, and for https the days left on its certificate, highlighted under 14 days. While its alert is open, the row links to it.

Choose a check’s name to see its last 24 hours: one mark per check, and each time it was down with the last error.

Seylo records the certificate’s expiry with each https result, but raises no alert for it.

A check is down after its Down after number of failed checks in a row: 2 by default. One failure is not enough unless you set it to 1.

  1. Down: Seylo raises one alert for the check, reason UptimeDown, at the check’s severity. It notifies once, like any alert: while the check stays down, the alert’s count and message keep up and nobody is told again, unless your organisation turned on reminders.
  2. Up: the first passing check resolves the alert, with the reason Check passed again at <time>, and tells every channel that was told it fired.
  3. Down again later: the same alert reopens, with its history, and notifies again. If the check went down again within 15 minutes of recovering by itself, the alert reopens only once the new outage has lasted 5 minutes, so a check that flaps does not post a down-and-up pair every few minutes. It is delayed, never silent.

In a channel, it reads like this:

[DOWN] Sales portal - 503 from portal.example.com, since 18:11 UTC
[UP] Sales portal - passed again at 18:20 UTC (portal.example.com)

An alert never carries the full URL. It names the check, the URL’s host, the status code or error (with any URL inside it cut to its host), the latency of the last answer, since when, and a link to the check. A query string is where a token in a health URL usually lives, so it never leaves Seylo.

Everything else on Alerts applies: acknowledge, snooze, assign, comment, resolve with a reason, and silences. On an uptime alert, Check now runs the check at once. A check you resolved by hand while it is still down reopens at the next failed check: the check has the last word.

Uptime checks run from the Seylo server, so their alerts belong to no cluster and no namespace. In a route or a silence:

  • reason UptimeDown matches every uptime alert;
  • workload matches the check’s name;
  • severity matches the check’s Alert severity;
  • a route or silence that names a cluster never matches an uptime alert.

A route with no matchers catches uptime alerts too. See notification channels and routes.

Nightshift does not investigate uptime alerts on its own: they name no cluster or workload for it to read.

Pausing or deleting a check resolves its open alert, in the name of the person who did it: nothing would ever see it pass again. Enabling a paused check that is still down raises a new alert.

Observability → Alerts → Detectors lists Uptime checks with the other built-in detectors. It is on by default. An admin can turn it off for the whole organisation; it cannot be set per cluster, because uptime checks do not run in a cluster. Off, a down check raises and reopens nothing, its open alerts are resolved, and the checks keep running and recording results. To quiet one check instead, pause it, or match it with a silence.

Import on the Uptime page brings in checks from blackbox exporter probes, Datadog HTTP tests and Uptime Kuma, with a preview before anything is created. See import from Uptime Kuma and import.