Service Map
The Service Map is a directed graph of how your services call each other, derived from trace spans.
Header
- Title:
Service Map · Visualize service-to-service communication. - Counter top-right:
<n> services, <n> connections. - Refresh button top-right.
Time picker
Presets: 15m | 1h | 6h | 24h | 7d plus explicit From/To datetime inputs.
Source scope
Respects the source switcher in the left rail. Pick one source to scope the map to traces from that source only. With All sources, the map shows every service across the org.
What the graph shows
- Nodes are services. The size of each node scales with the service's request volume.
- Color of each node maps to its error rate:
- Green: <5% errors (healthy).
- Yellow: 5–10% (warning).
- Red: >10% (critical).
- Edges are caller→callee relationships derived from
parent_span_idlinking. Edge width scales with call volume. - A label below each node shows the service name.
Legend
The bottom-right legend documents the visual encoding:
- Color → error rate band.
- Node size → request volume.
- Edge width → call volume.
Empty state
When no services have reported traces in the time range:
> "No services found. Ingest trace data to see the service topology."
The counter top-right reads 0 services, 0 connections.
Sending traces
Same OpenTelemetry setup as the Traces page. Set OTEL_SERVICE_NAME per service (this becomes the node label). Point the OTLP exporter at the agent on 4317 (gRPC) or 4318 (HTTP) and the agent's API key in X-API-Key.
Once spans flow, the map updates within seconds — but with a single service or services that don't traceably call one another, you'll see isolated nodes with no edges. Edges only appear when one service's span has a parent_span_id matching another service's span in the same trace.
What this page is not
- Not a service mesh. ObserveKit observes the graph; it doesn't proxy or route traffic.
- Not a CMDB. It's derived from real traced calls, not configuration. Services with zero traffic in the time window are absent.