Struct RunTraceSubscriber 

pub struct RunTraceSubscriber {
    pub host_client_id: String,
    /* private fields */
}

One live runs.trace.event subscriber — the producer side of the bounded channel whose drain task writes frames to the subscriber’s WebSocket (agent run tracing, U4).

Keyed in crate::session::ServerState::run_subscribers by (run_id, host_client_id) so two CarHost windows on the same run are independent streams (the explicit fanout the single-subscriber- per-method notification registry can’t provide), and so disconnect cleanup can drop exactly this connection’s subscriptions.

The producer holds ONLY the tx and calls RunTraceSubscriber::push — a non-blocking try_send. It never touches the WS socket, so a slow CarHost can never stall the recorder, the runs lock, or any other in-flight RPC (invariant #2). The dedicated drain task owns the socket write.

Fields

host_client_id: String

The connection that subscribed — its WS client_id.

Implementations

impl RunTraceSubscriber

pub fn spawn(host_client_id: String, channel: Arc<WsChannel>) -> Self

Spawn a drain task bound to channel and return the producer-side subscriber handle. The drain task serializes each runs.trace.event to a JSON-RPC notification frame and writes it to the subscriber’s WS; it exits when the tx is dropped (the subscriber is removed on unsubscribe / disconnect) — at which point the channel closes and recv() returns None.

pub fn push(&self, event: RunTraceEvent) -> bool

Non-blocking push of one event onto the drain channel. Returns false when the channel is full (a wedged subscriber) — the event is dropped rather than blocking the producer (invariant #2). The producer treats the drop as best-effort: the client detects the resulting cursor gap and re-subscribes to backfill (R8).