feat: More logging for command stream

server/internal/grpc/server.go

@@ -221,18 +221,37 @@ func (s *vantageServer) CommandStream(stream pb.Vantage_CommandStreamServer) err
 	ping := time.NewTicker(pingInterval)
 	defer ping.Stop()
 
+	// Beats are counted and reported periodically rather than logged one by
+	// one: at one every 20s per agent, a fleet of any size would drown every
+	// other line in the log. What is worth a line of its own is the first beat
+	// (it tells the operator this stream's watchdog is now armed on the agent
+	// side) and any failure to send one.
+	var beats int
+	summary := time.NewTicker(pingSummaryInterval)
+	defer summary.Stop()
+
 	ctx := stream.Context()
 	for {
 		select {
 		case <-ctx.Done():
 			return nil
+		case <-summary.C:
+			log.Printf("agent %s command stream healthy, %d beats in the last %s",
+				srv.ServerID, beats, pingSummaryInterval)
+			beats = 0
 		case <-ping.C:
 			// A failed send is the point: it is how this side learns the stream
 			// is gone, which runs the deferred release and frees the agent's
 			// presence claim for whichever pod it reconnects to.
 			if err := stream.Send(&pb.ServerCommand{Ping: &pb.PingCmd{}}); err != nil {
+				log.Printf("agent %s command stream beat failed after %d beats: %v",
+					srv.ServerID, beats, err)
 				return err
 			}
+			beats++
+			if beats == 1 {
+				log.Printf("agent %s command stream beating every %s", srv.ServerID, pingInterval)
+			}
 		case cmd, ok := <-ch:
 			if !ok {
 				return nil
@@ -249,6 +268,11 @@ func (s *vantageServer) CommandStream(stream pb.Vantage_CommandStreamServer) err
 // reconnect.
 const pingInterval = 20 * time.Second
 
+// How often an otherwise silent healthy stream says so. Long enough that a
+// large fleet does not fill the log, short enough that "this pod is still
+// serving that agent" is answerable from the log rather than by inference.
+const pingSummaryInterval = 5 * time.Minute
+
 // StartGRPC serves the agent API until stop is called.
 //
 // It returns a stop function rather than serving forever because an abrupt exit