Speed up leader/follower replication

This commit lowers the response time for a leader to send
updates to a follower. Should now be nearly instant.

internal/server/aof.go

@@ -130,6 +130,8 @@ func (s *Server) flushAOF(sync bool) {
 		if err != nil {
 			panic(err)
 		}
+		// send a broadcast to all sleeping followers
+		s.fcond.Broadcast()
 		if sync {
 			if err := s.aof.Sync(); err != nil {
 				panic(err)
@@ -165,11 +167,6 @@ func (s *Server) writeAOF(args []string, d *commandDetails) error {
 		s.aofsz += len(s.aofbuf) - n
 	}
 
-	// notify aof live connections that we have new data
-	s.fcond.L.Lock()
-	s.fcond.Broadcast()
-	s.fcond.L.Unlock()
-
 	// process geofences
 	if d != nil {
 		// webhook geofences
@@ -516,6 +513,7 @@ func (s *Server) liveAOF(pos int64, conn net.Conn, rd *PipelineReader, msg *Mess
 	if err != nil {
 		return err
 	}
+
 	b := make([]byte, 4096*2)
 	for {
 		n, err := f.Read(b)
@@ -525,7 +523,9 @@ func (s *Server) liveAOF(pos int64, conn net.Conn, rd *PipelineReader, msg *Mess
 			}
 		}
 		if err == io.EOF {
-			time.Sleep(time.Second / 4)
+			s.fcond.L.Lock()
+			s.fcond.Wait()
+			s.fcond.L.Unlock()
 		} else if err != nil {
 			return err
 		}

internal/server/server.go

@@ -135,7 +135,7 @@ type Server struct {
 	fcond    *sync.Cond
 	lstack   []*commandDetails
 	lives    map[*liveBuffer]bool
-	lcond    *sync.Cond
+	lcond    *sync.Cond // live geofence signal
 	fcup     bool       // follow caught up
 	fcuponce bool       // follow caught up once
 	aofconnM map[net.Conn]io.Closer
@@ -305,10 +305,19 @@ func Serve(opts Options) error {
 		nerr <- s.netServe()
 	}()
 
+	var fstop atomic.Bool
+	go func() {
+		for !fstop.Load() {
+			s.fcond.Broadcast()
+			time.Sleep(time.Second / 4)
+		}
+	}()
+
 	go func() {
 		<-opts.Shutdown
 		s.stopServer.Store(true)
 		log.Warnf("Shutting down...")
+		fstop.Store(true)
 		s.lnmu.Lock()
 		ln := s.ln
 		s.ln = nil