False-alarm protection

A single failed request is weak evidence. Networks drop packets, DNS resolvers time out, and a server can return one bad response during a deploy. Asniher verifies every failure before it alerts anyone.

The sequence

1. Primary check. Checks run from Europe (Germany) at the monitor's interval.
2. Re-check after 30 seconds. If a check fails, Asniher waits 30 seconds and checks again, resolving your host name through a different DNS resolver than the first attempt. If the re-check succeeds, no incident opens.
3. Remote confirmation. If the re-check also fails and remote probes are available, Asniher asks probes in other regions to run the same check.
   • If a remote probe reaches your service, no incident opens. The failure is recorded as a local issue.
   • If remote probes confirm the failure, an incident opens.
   • If remote probes do not answer in time, an incident opens based on the two local failures.
4. Incident and alert. The incident opens with the failure reason, and down alerts go to the monitor's channels.

Recovery needs a single successful check. The incident closes immediately and a recovery alert is sent.

What this means for detection time

Verification adds about 30 seconds, plus a few seconds when remote confirmation runs. For a monitor with a 1-minute interval, a real outage is typically reported within about 1.5 minutes of the first failed check.

What it filters out

• Transient DNS resolver failures or stale records on one resolver.
• Single dropped connections or one-off timeouts.
• Short blips during a deploy or restart that recover within 30 seconds.
• Network problems between Germany and your host that do not affect other regions (when remote probes are available).

What it does not filter

• Failures lasting longer than the re-check window. These are real outages and will alert.
• Problems visible from every region, such as an expired certificate or a 500 error from your application.

Heartbeat monitors

Heartbeat monitors do not use re-checks or remote confirmation. A missed ping after the grace period is already definitive, so the incident opens as soon as the deadline passes. Use the grace period to absorb normal variance in job duration.