Status page

This page describes what a status page for a dynamo-core deployment can show, using only the core's own monitoring: its Prometheus metrics, the alert rules in deploy/observability/alerts.yml, and the GET /healthz probe. It shows the state of each component now and in the past. It makes no promise of uptime or response time; the targets behind the states are on service levels and limits, and they are design targets, not commitments.

Components

ComponentWhat it covers
Core APIWhether the deployment answers at all: the owner API, the plugin wire and, on a multi-tenant deployment, the front door
AdmissionOpening streams, metering and reserving: how fast those requests are answered, and whether they are answered without a server error
SchedulerThe background work that signs vouchers, takes checkpoints, finalises settlements, renews credit card holds, expires approvals and freezes, and delivers webhooks
SettlementHow long signed work waits for a checkpoint, shown per funding mode the deployment runs (Control Mode, stablecoin escrow, credit card rail)
Adopted-rail claimsClaims of payees' vouchers on an adopted payment rail; shown only on a deployment that has one

States

StateWhat it means for a user
OperationalThe component is within its targets
DegradedThe component works, but slower or later than its target. Requests are still answered, and signed work is still settled, just later
OutageThe component does not work. Requests may fail, or background work has stopped
Paused by the operatorThe operator paused the deployment with the main switch. New work is refused on purpose, and work that was already signed stays settleable. This is not a fault

A component is never shown as better than its worst current alert.

How each state is derived

Each state comes from an alert rule in deploy/observability/alerts.yml, so the status page and the on-call alerts never disagree.

ComponentOperationalDegradedOutage
Core API/healthz and the scrape answernot usedDynamoCoreDown: the scrape fails for 2 minutes
Admissionp95 admission latency at most 250 ms and server errors at most one request in a thousandDynamoAdmissionLatencyHigh: p95 above 250 ms for 10 minutesDynamoAvailabilityLow: more than one request in a thousand answers 5xx for 10 minutes
Schedulera scheduler holds the lease and no duty is overdueDynamoDutyBacklog or DynamoDutyFailingDynamoSchedulerLeaseLost or DynamoSchedulerNotRunning
Settlement (lag shown per mode)lag at most 15 minutesDynamoSettlementLagHigh: the largest lag over every mode above 15 minutes for 10 minutes (one alert; the per-mode lag shows which mode)not derived from metrics; the operator posts it
Adopted-rail claimsno failed or lost claimnot usedDynamoRailClaimFailed: a claim failed or was lost in the last 30 minutes

"Paused by the operator" is not in the metrics. The main switch's state is read with the owner token at GET /owner/v1/main-switch, so the operator posts it by hand.

What the page never shows

Notes for operators