Update to LameDuck mode

- Increase grace period to 10sec
- Make default 2min
- Reject config with value less than 30sec
- Don't wait more than 1sec between clients if there are much
  less than alloted time
- Stop after last one (was still sleeping after last client was
  closed)

Signed-off-by: Ivan Kozlovic <ivan@synadia.com>

server/configs/test.conf

@@ -45,4 +45,4 @@ ping_max: 3
 # how long server can block on a socket write to a client
 write_deadline: "3s"
 
-lame_duck_duration: "4s"
+lame_duck_duration: "4m"

server/const.go

@@ -135,5 +135,5 @@ const (
 
 	// DEFAULT_LAME_DUCK_DURATION is the time in which the server spreads
 	// the closing of clients when signaled to go in lame duck mode.
-	DEFAULT_LAME_DUCK_DURATION = 30 * time.Second
+	DEFAULT_LAME_DUCK_DURATION = 2 * time.Minute
 )

server/opts.go

@@ -479,6 +479,11 @@ func (o *Options) ProcessConfigFile(configFile string) error {
 				errors = append(errors, err)
 				continue
 			}
+			if dur < 30*time.Second {
+				err := &configErr{tk, fmt.Sprintf("invalid lame_duck_duration of %v, minimum is 30 seconds", dur)}
+				errors = append(errors, err)
+				continue
+			}
 			o.LameDuckDuration = dur
 		case "trusted":
 			switch v.(type) {

server/opts_test.go

@@ -89,7 +89,7 @@ func TestConfigFile(t *testing.T) {
 		PingInterval:     60 * time.Second,
 		MaxPingsOut:      3,
 		WriteDeadline:    3 * time.Second,
-		LameDuckDuration: 4 * time.Second,
+		LameDuckDuration: 4 * time.Minute,
 	}
 
 	opts, err := ProcessConfigFile("./configs/test.conf")

server/server.go

@@ -42,7 +42,7 @@ import (
 )
 
 // Time to wait before starting closing clients when in LD mode.
-const lameDuckModeDefaultInitialDelay = int64(time.Second)
+const lameDuckModeDefaultInitialDelay = int64(10 * time.Second)
 
 // Make this a variable so that we can change during tests
 var lameDuckModeInitialDelay = int64(lameDuckModeDefaultInitialDelay)
@@ -1853,7 +1853,13 @@ func (s *Server) lameDuckMode() {
 		// use a tiny sleep interval that will result in yield likely.
 		si = 1
 		batch = int(numClients / dur)
+	} else if si > int64(time.Second) {
+		// Conversely, there is no need to sleep too long between clients
+		// and spread say 10 clients for the 2min duration. Sleeping no
+		// more than 1sec.
+		si = int64(time.Second)
 	}
+
 	// Now capture all clients
 	clients := make([]*client, 0, len(s.clients))
 	for _, client := range s.clients {
@@ -1873,6 +1879,9 @@ func (s *Server) lameDuckMode() {
 	}
 	for i, client := range clients {
 		client.closeConnection(ServerShutdown)
+		if i == len(clients)-1 {
+			break
+		}
 		if batch == 1 || i%batch == 0 {
 			// We pick a random interval which will be at least si/2
 			v := rand.Int63n(si)