Health dashboard
Description
The Health dashboard view reports the operational state of the gateway across its failure surfaces for the last 24 hours. The view is the Health tab of Overview › Analytics (route /health), available to the admin role only.
The view reads its data in a single call. Empty cells mean the gateway is healthy in that surface. When every surface is green, the view shows the banner All clear — every surface is green.
The view groups the data into the sections below.
Synthetic probes
The Synthetic probes section lists the scheduled end-to-end probes. Each row shows the probe name, the pass count, the fail count, the consecutive-fail count, the last run time, and a status. A probe with three consecutive failures is marked as down.
Request log non-2xx
The Request log non-2xx (24h, excluding 499) section lists the non-2xx responses recorded in the request log. Each row shows the status code, the count, and the number of distinct tenants. Status 499 (client-aborted) is excluded from this section.
Client-aborted requests
The Client-aborted requests (24h, status 499) section lists the requests that the client aborted before completion. Each row shows the count, the number of distinct tenants, and the number of distinct users.
Client errors phoned home
The Client errors phoned home (24h) section lists the errors reported by the browser client. Each row shows the message prefix and the count.
Provider health
The Provider health section lists the most recent health snapshot per provider. Each row shows the provider, the status, the latency, the check time, and a message.
Open queues
The Open queues section shows the count of open content reports, the count of scrub-preview failures for the last 24 hours, and the state of the Platform web-search key (Linkup): set (with the number of gateways using it), unset — no gateway depends on it yet, MISSING (the key is unset while production gateways carry a keyless Linkup web-search block — their web search is off; set the key under Feature Flags), or unknown when the status could not be read. MISSING and unknown clear the all-clear state and also raise a platform alert.
Plan-aware model taglines
The Plan-aware model taglines counter reports the health of the plan-aware model copy shown in the model picker (see plan_model_copy_enabled). It reads ok when the copy is current, STALE with a count when generation was rejected for one or more model sets and the picker shows fallback text, an indistinct count when a plan offers two or more models that no fact tells apart, so the picker could show the same tagline for both, or unknown when the status could not be read. A stale, indistinct, or unknown state clears the all-clear state.
Recent probe runs
The Recent probe runs (last 24h, up to 15 per probe) section lists the most recent probe runs — capped per probe, so one frequently-writing probe cannot push the others out of the list. Each row shows the time, the probe name, the status, and the failure reason. The consecutive-failure count in the probes table above is computed by the server over every run in the window, not from this list. Probe history is retained for 30 days.