Live metrics¶
Live metrics show per-pipeline volume and latency in the UI — how much data each pipeline is processing right now, what its queue depth looks like, where back-pressure is building up.
The data comes from the collectors' own metrics (receivers/processors/exporters instrumentation), exported via the ownmetrics capability of the OpAMP extension and ingested by Ampora's metrics endpoint.
Where it appears¶
- Configuration → Pipeline visualizer — overlay mode shows the per-component throughput as a thickness on the edges and a small inline number.
- Agent details → Live metrics — per-pipeline charts for the agent.
- Fleet flow view — aggregate flow thickness across the fleet reflects live throughput, not just topology.
Aggregation¶
For per-agent views, the data is the agent's own — no aggregation. For Fleet flow view and Configuration overlay, the data is summed across the matching agents:
- Per pipeline edge: total points/sec across agents using that edge.
- Per exporter: total bytes/sec sent.
- Per receiver: total points/sec ingested.
Cardinality¶
Live metrics deliberately stay coarse. Per-component, per-signal, per-pipeline counters — yes. Per-trace-attribute breakdowns — no. Those belong in your observability backend.
Latency¶
Cached for 30 seconds at the aggregate level. Per-agent views are fresher (events are pushed live via the bus). The toolbar shows "Last refreshed: 12 s ago" so you know what you are looking at.
Click Refresh to force a recompute.
Enabling¶
Live metrics are on whenever the agent reports ReportsOwnMetrics. You do not need to configure anything on the Ampora side — the endpoint is always listening.
For agents that do not report own metrics, the UI shows static topology only (no thickness, no numbers).
What the numbers mean¶
- Receiver
points/sec— points ingested by the receiver, before any processing. - Processor
accepted/refused/dropped— counters of the processor's per-batch outcomes. - Exporter
sent/failed/queued— counters of the exporter's outbound delivery. - Pipeline latency p95 — receive-to-export wall clock, sampled at the
batchboundary.
These are the same metric names the collector emits via Prometheus. If you cross-reference with your own scrape, the numbers should match.
What this is not¶
- Not a replacement for your APM. Your real-time analysis on traces / metrics / logs lives in Tempo, Jaeger, Prometheus, Datadog, Honeycomb, etc.
- Not stored long-term. Ampora keeps a rolling 24-hour window for the UI; analysis past that should query your scraped Prometheus.
The point of live metrics is the question "is the change I just rolled doing what I expected?" — answered in the same UI as the rollout, without context-switching to your observability tool.