Service map
Who calls whom, drawn from your traces.
The map #
The service map draws every service as a node and every observed call between them as an edge. Node size reflects traffic; edge colour reflects errors.
How edges are found #
An edge is drawn when a traced call goes from one service to another. Where a dependency is not instrumented, the peer is still shown as a node so the dependency is visible.
- Instrumented target — the edge is exact, from the caller and callee spans.
- Uninstrumented target — the dependency is drawn from the caller's declared peer.
Grouping #
Group the map by:
- service
- host
- team
- env
- namespace
- cluster
- pod
Grouping by team is the most useful view when you are reviewing ownership: it shows which teams are coupled.
Reading the map #
- A thick edge is a high-volume dependency.
- A red edge is carrying errors.
- An isolated node has no traced traffic in the window.
- A cluster of nodes around one edge is usually one dependency many services share.
Start grouped by service and switch to team. The first view explains the system; the second explains who to talk to.
Onward #
A node on the map leads to Services and latency for that service's numbers. The datastore and queue edges continue in Data streams monitoring.
Finding an unexpected dependency #
A dependency nobody documented usually shows as a node with no team, or as a node pointing at a datastore nobody claimed. Both are worth resolving while the map is in front of you.