Live topology — ring colour is health (alerts + error rate), edge weight is throughput. Click a service to open its dashboard.

Sampled and slow-endpoint transaction traces: controller → service → repository, method by method. Click a row to expand.

Everything every service recorded about this request id — the end-to-end path.

Deduplicated by exception + failing frame; click a row for the stack and request ids.

Threshold breaches across all services; recently resolved shown muted.

5xx responses count as errors; hover for exact values.

Restarts, alert transitions and first-seen exceptions — what changed, in order.

The monitor, monitored — ingest pipeline, history writer, state persistence and the collector fleet.

Automatically derived from the last hours of metrics, call graphs, exceptions and runtime data — what deserves attention and why.

Click a column to sort, a row to see its trend. Most-accessed first by default.

Captured requests ≥ the collector's slow threshold — exact call, exact time.

1-minute candles of requests/sec (open · high · low · close) with transferred bytes as volume; last 30 minutes, ticking every 5s.

Captured automatically when process CPU exceeds 80% — what the busiest threads were doing at that moment.

Broker totals and per-queue backlog from the management API — backlogged or consumer-less queues are the pipeline's earliest warning.

Live server stats over the wire protocol.

Change a service's log level live — delivered with its next snapshot (≤15s), applied in-process, and auto-reverted when the duration expires.

Captured in-process (WARN/ERROR/INFO) and shipped with the metric snapshots — request ids link into the end-to-end waterfall; lines with a stack expand on click.

Role-based access: ADMIN manages operators; VIEWER sees dashboards only. Every operator enrols their own authenticator at first sign-in.

Every sign-in and administrative action — who, when, from where.