Infrastructure / Monitoring
A green light needs evidence.
Service checks, host metrics, disk health, and operational events become useful when they explain what is wrong—and admit when the data is missing.
Last reviewed
On this page
Several views of health
Uptime Kuma checks service availability. Scrutiny tracks disk SMART information. Pulse brings host, guest, network, storage, and event data together. Hearth presents household-relevant monitoring inside Home Assistant.
These answer different questions. An HTTP endpoint responding does not prove that a database migration succeeded. A disk being present does not prove it is healthy. A collector returning yesterday’s result does not prove that a service is up now.
My integration work
The Home Assistant collector and custom frontend translate the estate’s signals into a usable interface. Compact snapshots carry the overview; larger detail payloads are fetched when a view actually needs them. That reduces repeated data movement while retaining drill-down capability.
The interface distinguishes stale, unavailable, and healthy states. It also keeps control positions stable when events arrive. Monitoring should help someone act, not make the screen unpredictable.
Alerts that deserve attention
Operational events flow through push notifications and an event feed. The useful unit is a meaningful change: a failure, an escalation, or a recovery. Repeating the same condition every minute makes important messages easier to miss.
The thin-pool monitor is a concrete example. It tracks both data and metadata pressure, uses severity thresholds, and applies hysteresis to recovery. It also limits repeat alerts while a condition remains elevated.
Read the capacity incident for the failure that motivated it.
Check the observer
A backup job needs failure reporting and a freshness check. A monitoring collector needs an age limit. The application should know the difference between an empty result and a failed fetch.
The browser tests exercise these states with synthetic data. A test that simulates a failed collector is useful evidence about presentation behavior; it is not a claim about current infrastructure health.
Public status stays appropriately small
The family homepage shows a selected service-health summary. Detailed operator inventory and household events belong in the authenticated operational interfaces. A portfolio screenshot is a staged illustration, not a public live monitoring feed.