pub fn spawn_streaming_response(service_name: &str, continuation: Continuation)Expand description
Spawn the server-side half of a streaming RPC — the task that keeps pushing
data after the StreamInfo reply has been sent — onto the current
LocalSet, bounded by a per-service admission permit.
Replaces the former “transport front-end spawns the continuation it got back
from process_request” model (#186): the dispatch core now spawns this task
itself and returns only the reply bytes, so the streaming lifecycle is no
longer threaded through every transport’s request/response path. This is the
M1 shape; in M2 the streaming transport (StreamChannel) moves to a
service-owned moq broadcast and this coarse per-service cap is replaced by
the StreamOpt backpressure axes (#134).
service_name keys the per-service permit pool (see
DEFAULT_MAX_CONCURRENT_STREAMS_PER_SERVICE and
install_max_concurrent_streams_per_service). When the pool is saturated
the task waits for a permit rather than being dropped — and the wait is
logged so saturation is observable rather than a silent stall.
§Invariant
MUST be called from within a tokio::task::LocalSet: the task is ?Send
(like every RequestService), and
spawn_local panics outside a LocalSet. Every
process_request caller already
runs on a LocalSet (ZMQ RequestLoop, the WebTransport server, and the
generic plane’s LocalServiceBridge), which is the only place this is
invoked.