Thanos, Prometheus and Golang version used:
Thanos 0.31.0
Grafana Tempo 2.1.1
Otel Collector 0.81.0

Object Storage Provider:
AWS S3

What happened:
Missing span of kind "client" for communications beginning in Frontend towards downstream url.
Frontend fetches fifocache with 0 keys found. Then it calls a "retry" routine to fetch the data from the downstream url (see attached image).
There is a span of kind "server" for the http url "/api/v1/query_range" on the Querier side (the downstream) but there is no matching "client" span from the Frontend side.

What you expected to happen:
Expected a "client" kind span from the Frontend when a request is made to the downstream url, something like na "HTTP Outgoing Request"

How to reproduce it (as minimally and precisely as possible):
Using the following architecture:

graph LR
A[Grafana] --http--> B(Frontend)
B --http--> C(Global Query)
N --grpc--> F
C --grpc--> N(Query pool)

N --grpc--> I
N --grpc--> J
D(Compactor) --http--> E{S3}
F(Storegateway Pool) --http--> E

I(SidecarX/PrometheusX) --http--> E
J(Receiver) --http--> E
M(PrometheusY) --http--> J
subgraph No grpc inbound
  M
end
subgraph Data at rest
  E
end

Frontend uses local memory for cache.

All Thanos components send tracing data to an otel collector that relays to a all-in-one Grafana Tempo instance (no special config).

Thanos tracing config:

type: OTLP
config:
  client_type: "grpc"
  insecure: true
  endpoint: "tempo.distributor.url:port"

Trace "ServiceName" is set via "OTEL_SERVICE_NAME" environment variable.

Full logs to relevant components:

Anything else we need to know:
For my architecture, I use a global Querier. I've tried pointing Frontend downstream directly to a Querier from the Query pool (to bypass global Querier) but It behaves the same.