AutoMerge is generally available

Generated at:

Component schema

GetTopologyRequest

object1 direct reference

Definition

Request for POST /api/topology. Returns the service topology for the time range: one node per service.namespace + service.name pair, and one edge per observed direction of calls between two nodes.

Grouping is not client-selectable, and deliberately so. The connection metric carries only service.name and service.namespace from the calling side of a call, so no other grouping can identify a caller; offering a groupBy would return a silently wrong graph for every other value.

TimeReferenceRangerequired

A range of time between two time references.

Dataset

The slug of a dataset, such as default. Where a request may omit it, "Selecting a dataset" in the API introduction describes how the dataset is chosen.

allOf

Selects services by resource attributes. Applied to both halves of the graph, but they are not the same predicate, and the difference shows in the response. Read both limits below before relying on this field.

The edge half filters on the called service only. An edge's attributes are the callee's: the connection metric carries the caller as nothing but service.name and service.namespace, so there is no caller-side attribute to match on. An edge is returned when its target matches, whatever its source is. A caller outside the filter still comes back as a node — as inferred, since it is not in the filtered service list — while the matching service's own outbound edges are absent, their targets having not matched. Filtering by service.name returns that service's inbound traffic, not its neighbourhood.

Beyond resource attributes, two keys behave differently from the rest. dash0.operation.name and dash0.operation.type are carried on the connection metric, copied from the server span, so they do filter edges — by the callee's operation, keeping the edges whose inbound calls had that operation. The result is a graph of who calls that operation, which is a narrower thing than the node half returns for the same criterion.

Every other non-resource key — http.route and the span attributes generally — is accepted, is applied correctly to the node half, and matches no connection series, so the response is the filtered services with no edges at all. That is not rejected, and is indistinguishable from a tenant whose filtered services make no calls.

A 400 is returned only for a criterion that cannot be parsed, not for one that is merely inapplicable to edges.

ServiceStatistic[]

Statistics to compute per node, from the node's own RED metrics. Defaults to count, statusCounts and duration.p50. An empty array returns node identities only.

Same vocabulary and same cost rules as GetServiceCatalogRequest.statistics.

Statistics to compute per edge, from the connection metric. Defaults to count, statusCounts and duration.p50. An empty array returns edge endpoints only, which is the cheapest way to fetch the graph's shape.

This selects what each edge carries, never which edges exist: the same time range and filter return the same nodes and edges whatever is requested here. An edge whose duration cannot be determined — a connection observed with no completed calls in the range — is returned without that statistic rather than omitted.

integer

Maximum number of services read from the service list. When more match, the list is truncated and the response carries a warnings entry naming the limit.

This caps the services that carry statistics, not the size of the graph. It bounds neither the edges nor the nodes synthesized from them, so a response can contain more than nodeLimit nodes. Truncation punches no holes: a service dropped here still appears as an inferred node wherever an edge references it, without statistics of its own.

The cut is by identity, not by traffic. The service list is ordered by service.namespace then service.name, both ascending, so services with no namespace sort first and survive the cut. On a tenant whose un-namespaced peers — hosts, external endpoints — exceed the limit, the namespaced services are the ones that lose their statistics, however busy they are. Raise the limit or narrow filter; there is no way to ask for the busiest services instead.

Used by 1

These schemas and endpoints reference this type directly.