K2 Uptime monitoring
- Area
- K — Observability
- Audience
- operator
- Kind
- feature
- Status
- accepted
- Maturity
- planned
- Labels
- observability, verification
- Requires
- B5
- Relates to
- G7
The feature in the specification · Read at
Requirements
Uptime monitoring MUST use an open-source, self-hostable monitor, and MUST NOT depend on a hosted uptime service.
- locked by
- 0.23.0 open
The monitor MUST be independent of lemonfiber's own health path, so a failure in the tool's reporting does not blind or falsify uptime coverage.
- locked by
- 0.23.0 open
Every critical service MUST have a live monitor, and every household-facing entry point MUST have one.
- locked by
- 0.23.0 open
Jobs expected to run on a schedule MUST be watched by heartbeat checks, and a missed check-in MUST itself raise an alarm.
- locked by
- 0.23.0 open
The tool MUST assert that each critical service and entry point has a monitor that is actively checking, not merely defined.
- locked by
- 0.23.0 open
The tool MUST assert that a deliberately induced failure is detected by the monitor and surfaced.
- locked by
- 0.23.0 open
Where the monitor lacks a stable configuration API, the tool MUST provision it through a supported mechanism and MUST state the limitation rather than implying an API exists.
- locked by
- 0.23.0 open
An unreachable monitor MUST be reported as blind coverage and MUST NOT be rendered as all-clear.
- locked by
- 0.23.0 open
A monitored target MUST fail consistently before alarming, so a single slow response does not raise a false positive.
- locked by
- 0.23.0 open
An intentionally stopped service MUST be treated as expected-down and MUST NOT raise an outage alarm.
- locked by
- 0.23.0 open
Alert routing MUST be reconciled with the stack's notifications (B5) so a single event does not produce duplicate alerts.
- locked by
- 0.23.0 open
An entry point MUST be monitored over the path the household actually uses, and internal reachability MUST NOT be reported as household reachability.
- locked by
- 0.23.0 open
An alert that could not be delivered MUST be surfaced as undelivered rather than recorded as delivered.
- locked by
- 0.23.0 open
Provisioning monitors and running the detection proof MUST each be reachable non-interactively.
- locked by
- 0.23.0 open