Application Observability
Process groups, services, applications
OneAgent discovers a process group (identical processes across a tier), then applies service detection rules to split processes into services - web request services (HTTP entry points), database services, and queue services. A single process group can host several services. Applications are the RUM side (web/mobile) and sit above services. Candidates regularly lose points by treating these three as synonyms.
Traces and attribution
A PurePath is the end-to-end distributed trace: every service, database call, and remote call in one request, stitched across hosts. When the same endpoint could belong to several tiers, backtrace decides which service owns the request by looking at the initiator. Request naming rules then collapse high-cardinality URLs (/orders/12345 → /orders/{id}) so aggregations are meaningful - do not confuse the two: backtrace = attribution, naming = aggregation.
Analysis surfaces
Failure detection (HTTP status, exceptions, custom rules) decides what counts as a failed request; response time hotspots and method hotspots show where time is spent; multidimensional analysis pivots service metrics by dimensions such as endpoint, geography, or Kubernetes namespace; SLOs express objectives with error budgets.
Worked rule of thumb
If a question asks how to stop thousands of REST paths flooding your service view, the answer is a request naming rule with a placeholder, not a management zone and not a new service. If it asks how spans from an uninstrumented OpenTelemetry source join the trace, the answer is OTel ingest - typically via ActiveGate/API - merged with OneAgent spans.
Trap: distributed tracing silently breaks when one tier has no agent, so the PurePath stops there rather than showing a partial remote call.