Generated at:
Topology Edge
Definition
One direction of calls between two nodes, aggregated over the whole
time range. source and target are service identities that each
match exactly one nodes[].service; join on (namespace, name).
There is no opaque edge or node identifier: the identity is the key. A pair of services calling each other both ways appears as two edges.
Identifies a service: an OpenTelemetry resource grouped by
service.namespace + service.name.
Field names are unprefixed because the property carrying this object
already qualifies them (service.name, not service.serviceName).
That is a deliberate divergence from StarredServiceRef, whose
prefixed names are already shipped to clients and cannot be renamed
without a breaking change, so the two stay separate rather than being
consolidated.
Identifies a service: an OpenTelemetry resource grouped by
service.namespace + service.name.
Field names are unprefixed because the property carrying this object
already qualifies them (service.name, not service.serviceName).
That is a deliberate divergence from StarredServiceRef, whose
prefixed names are already shipped to clients and cannot be renamed
without a breaking change, so the two stay separate rather than being
consolidated.
RED statistics for one edge: total count, the per-status breakdown
(statusCounts), and the duration distribution (duration). A field
the server declined to compute is absent: the key is omitted rather
than sent as null, so a strict client should test for presence.
The error rate is not included: derive it as
statusCounts.error / count. count may be 0, in which case there
is no rate to show.
These numbers are not comparable to the target node's. Two independent reasons, both expected rather than defects:
- Different histograms. An edge's
durationis read from an explicit-bucket histogram and interpolated within a bucket, while a node's comes from an exponential histogram and is near-exact. The same service's nodeduration.p50and its inbound edges'duration.p50will differ, sometimes substantially. - Different populations. A node's
countis every request the service served. An edge'scountis only the calls whose caller could be identified, and is scaled by the sampling compensation factor. Inbound edge counts therefore do not sum to the node count, and are usually lower.
Measured from the server side of each call.
Used by 1
These schemas and endpoints reference this type directly.