// Publish sends a single MQTT PUBLISH message to the server. On completion, the
// supplied OnCompleteFunc is called. For QOS 0 messages, onComplete is called
// immediately after the message is sent to the outgoing buffer. For QOS 1 messages,
// onComplete is called when PUBACK is received. For QOS 2 messages, onComplete is
// called after the PUBCOMP message is received.
func (this *Server) Publish(msg *message.PublishMessage, onComplete OnCompleteFunc) error {
if err := this.checkConfiguration(); err != nil {
return err
}
if msg.Retain() {
if err := this.topicsMgr.Retain(msg); err != nil {
glog.Errorf("Error retaining message: %v", err)
}
}
if err := this.topicsMgr.Subscribers(msg.Topic(), msg.QoS(), &this.subs, &this.qoss); err != nil {
return err
}
msg.SetRetain(false)
//glog.Debugf("(server) Publishing to topic %q and %d subscribers", string(msg.Topic()), len(this.subs))
for _, s := range this.subs {
if s != nil {
fn, ok := s.(*OnPublishFunc)
if !ok {
glog.Errorf("Invalid onPublish Function")
} else {
(*fn)(msg)
}
}
}
return nil
}
// onPublish() is called when the server receives a PUBLISH message AND have completed
// the ack cycle. This method will get the list of subscribers based on the publish
// topic, and publishes the message to the list of subscribers.
func (this *service) onPublish(msg *message.PublishMessage) error {
if msg.Retain() {
if err := this.topicsMgr.Retain(msg); err != nil {
glog.Errorf("(%s) Error retaining message: %v", this.cid(), err)
}
}
err := this.topicsMgr.Subscribers(msg.Topic(), msg.QoS(), &this.subs, &this.qoss)
if err != nil {
glog.Errorf("(%s) Error retrieving subscribers list: %v", this.cid(), err)
return err
}
msg.SetRetain(false)
//glog.Debugf("(%s) Publishing to topic %q and %d subscribers", this.cid(), string(msg.Topic()), len(this.subs))
for _, s := range this.subs {
if s != nil {
fn, ok := s.(*OnPublishFunc)
if !ok {
glog.Errorf("Invalid onPublish Function")
return fmt.Errorf("Invalid onPublish Function")
} else {
(*fn)(msg)
}
}
}
return nil
}
func main() {
svr := &service.Server{
KeepAlive: keepAlive,
ConnectTimeout: connectTimeout,
AckTimeout: ackTimeout,
TimeoutRetries: timeoutRetries,
SessionsProvider: sessionsProvider,
TopicsProvider: topicsProvider,
}
var f *os.File
var err error
if cpuprofile != "" {
f, err = os.Create(cpuprofile)
if err != nil {
log.Fatal(err)
}
pprof.StartCPUProfile(f)
}
sigchan := make(chan os.Signal, 1)
signal.Notify(sigchan, os.Interrupt, os.Kill)
go func() {
sig := <-sigchan
glog.Errorf("Existing due to trapped signal; %v", sig)
if f != nil {
glog.Errorf("Stopping profile")
pprof.StopCPUProfile()
f.Close()
}
svr.Close()
os.Exit(0)
}()
mqttaddr := "tcp://:1883"
if len(wsAddr) > 0 || len(wssAddr) > 0 {
addr := "tcp://127.0.0.1:1883"
AddWebsocketHandler("/mqtt", addr)
/* start a plain websocket listener */
if len(wsAddr) > 0 {
go ListenAndServeWebsocket(wsAddr)
}
/* start a secure websocket listener */
if len(wssAddr) > 0 && len(wssCertPath) > 0 && len(wssKeyPath) > 0 {
go ListenAndServeWebsocketSecure(wssAddr, wssCertPath, wssKeyPath)
}
}
/* create plain MQTT listener */
err = svr.ListenAndServe(mqttaddr)
if err != nil {
glog.Errorf("surgemq/main: %v", err)
}
}
func profile() {
var f *os.File
var err error
if cpuprofile != "" {
f, err = os.Create(cpuprofile)
if err != nil {
log.Fatal(err)
}
pprof.StartCPUProfile(f)
}
sigchan := make(chan os.Signal, 1)
signal.Notify(sigchan, os.Interrupt, os.Kill)
go func() {
select {
case sig := <-sigchan:
log.Printf("Existing due to trapped signal; %v", sig)
case <-quit:
log.Println("Quiting...")
}
if f != nil {
glog.Errorf("Stopping profile")
pprof.StopCPUProfile()
f.Close()
}
close(done)
os.Exit(0)
}()
}
func (this *service) start() error {
var err error
// Create the incoming ring buffer
this.in, err = newBuffer(defaultBufferSize)
if err != nil {
return err
}
// Create the outgoing ring buffer
this.out, err = newBuffer(defaultBufferSize)
if err != nil {
return err
}
// If this is a server
if !this.client {
// Creat the onPublishFunc so it can be used for published messages
this.onpub = func(msg *message.PublishMessage) error {
if err := this.publish(msg, nil); err != nil {
glog.Errorf("service/onPublish: Error publishing message: %v", err)
return err
}
return nil
}
// If this is a recovered session, then add any topics it subscribed before
topics, qoss, err := this.sess.Topics()
if err != nil {
return err
} else {
for i, t := range topics {
this.topicsMgr.Subscribe([]byte(t), qoss[i], &this.onpub)
}
}
}
// Processor is responsible for reading messages out of the buffer and processing
// them accordingly.
this.wgStarted.Add(1)
this.wgStopped.Add(1)
go this.processor()
// Receiver is responsible for reading from the connection and putting data into
// a buffer.
this.wgStarted.Add(1)
this.wgStopped.Add(1)
go this.receiver()
// Sender is responsible for writing data in the buffer into the connection.
this.wgStarted.Add(1)
this.wgStopped.Add(1)
go this.sender()
// Wait for all the goroutines to start before returning
this.wgStarted.Wait()
return nil
}
func main() {
svr := &service.Server{
KeepAlive: keepAlive,
ConnectTimeout: connectTimeout,
AckTimeout: ackTimeout,
TimeoutRetries: timeoutRetries,
SessionsProvider: sessionsProvider,
TopicsProvider: topicsProvider,
}
var f *os.File
var err error
if cpuprofile != "" {
f, err = os.Create(cpuprofile)
if err != nil {
log.Fatal(err)
}
pprof.StartCPUProfile(f)
}
sigchan := make(chan os.Signal, 1)
signal.Notify(sigchan, os.Interrupt, os.Kill)
go func() {
sig := <-sigchan
glog.Errorf("Existing due to trapped signal; %v", sig)
if f != nil {
glog.Errorf("Stopping profile")
pprof.StopCPUProfile()
f.Close()
}
svr.Close()
os.Exit(0)
}()
err = svr.ListenAndServe("tcp://:1883")
if err != nil {
glog.Errorf("surgemq/main: %v", err)
}
}
// ListenAndServe listents to connections on the URI requested, and handles any
// incoming MQTT client sessions. It should not return until Close() is called
// or if there's some critical error that stops the server from running. The URI
// supplied should be of the form "protocol://host:port" that can be parsed by
// url.Parse(). For example, an URI could be "tcp://0.0.0.0:1883".
func (this *Server) ListenAndServe(uri string) error {
defer atomic.CompareAndSwapInt32(&this.running, 1, 0)
if !atomic.CompareAndSwapInt32(&this.running, 0, 1) {
return fmt.Errorf("server/ListenAndServe: Server is already running")
}
this.quit = make(chan struct{})
u, err := url.Parse(uri)
if err != nil {
return err
}
this.ln, err = net.Listen(u.Scheme, u.Host)
if err != nil {
return err
}
defer this.ln.Close()
glog.Infof("server/ListenAndServe: server is ready...")
var tempDelay time.Duration // how long to sleep on accept failure
for {
conn, err := this.ln.Accept()
if err != nil {
// http://zhen.org/blog/graceful-shutdown-of-go-net-dot-listeners/
select {
case <-this.quit:
return nil
default:
}
// Borrowed from go1.3.3/src/pkg/net/http/server.go:1699
if ne, ok := err.(net.Error); ok && ne.Temporary() {
if tempDelay == 0 {
tempDelay = 5 * time.Millisecond
} else {
tempDelay *= 2
}
if max := 1 * time.Second; tempDelay > max {
tempDelay = max
}
glog.Errorf("server/ListenAndServe: Accept error: %v; retrying in %v", err, tempDelay)
time.Sleep(tempDelay)
continue
}
return err
}
go this.handleConnection(conn)
}
}
// receiver() reads data from the network, and writes the data into the incoming buffer
func (this *service) receiver() {
defer func() {
// Let's recover from panic
if r := recover(); r != nil {
glog.Errorf("(%s) Recovering from panic: %v", this.cid(), r)
}
this.wgStopped.Done()
glog.Debugf("(%s) Stopping receiver", this.cid())
}()
glog.Debugf("(%s) Starting receiver", this.cid())
this.wgStarted.Done()
switch conn := this.conn.(type) {
case net.Conn:
//glog.Debugf("server/handleConnection: Setting read deadline to %d", time.Second*time.Duration(this.keepAlive))
keepAlive := time.Second * time.Duration(this.keepAlive)
r := timeoutReader{
d: keepAlive + (keepAlive / 2),
conn: conn,
}
for {
_, err := this.in.ReadFrom(r)
if err != nil {
if err != io.EOF {
glog.Errorf("(%s) error reading from connection: %v", this.cid(), err)
}
return
}
}
//case *websocket.Conn:
// glog.Errorf("(%s) Websocket: %v", this.cid(), ErrInvalidConnectionType)
default:
glog.Errorf("(%s) %v", this.cid(), ErrInvalidConnectionType)
}
}
func AddWebsocketHandler(urlPattern string, uri string) error {
glog.Debugf("AddWebsocketHandler urlPattern=%s, uri=%s", urlPattern, uri)
u, err := url.Parse(uri)
if err != nil {
glog.Errorf("surgemq/main: %v", err)
return err
}
h := func(ws *websocket.Conn) {
WebsocketTcpProxy(ws, u.Scheme, u.Host)
}
http.Handle(urlPattern, websocket.Handler(h))
return nil
}
func startFanPublisher(t testing.TB, cid int, wg *sync.WaitGroup) {
now := time.Now()
runClientTest(t, cid, wg, func(svc *service.Client) {
select {
case <-done:
case <-time.After(time.Second * time.Duration(subscribers)):
glog.Infof("(surgemq%d) Timed out waiting for subscribe response", cid)
return
}
cnt := messages
sent := 0
payload := make([]byte, size)
msg := message.NewPublishMessage()
msg.SetTopic(topic)
msg.SetQoS(qos)
for i := 0; i < cnt; i++ {
binary.BigEndian.PutUint32(payload, uint32(cid*cnt+i))
msg.SetPayload(payload)
err := svc.Publish(msg, nil)
if err != nil {
break
}
sent++
}
since := time.Since(now).Nanoseconds()
statMu.Lock()
totalSent += int64(sent)
totalSentTime += int64(since)
if since > sentSince {
sentSince = since
}
statMu.Unlock()
glog.Debugf("(surgemq%d) Sent %d messages in %d ns, %d ns/msg, %d msgs/sec", cid, sent, since, int(float64(since)/float64(cnt)), int(float64(sent)/(float64(since)/float64(time.Second))))
select {
case <-done2:
case <-time.Tick(time.Second * time.Duration(nap*publishers)):
glog.Errorf("Timed out waiting for messages to be received.")
}
})
}
// sender() writes data from the outgoing buffer to the network
func (this *service) sender() {
defer func() {
// Let's recover from panic
if r := recover(); r != nil {
glog.Errorf("(%s) Recovering from panic: %v", this.cid(), r)
}
this.wgStopped.Done()
glog.Debugf("(%s) Stopping sender", this.cid())
}()
glog.Debugf("(%s) Starting sender", this.cid())
this.wgStarted.Done()
switch conn := this.conn.(type) {
case net.Conn:
for {
_, err := this.out.WriteTo(conn)
if err != nil {
if err != io.EOF {
glog.Errorf("(%s) error writing data: %v", this.cid(), err)
}
return
}
}
//case *websocket.Conn:
// glog.Errorf("(%s) Websocket not supported", this.cid())
default:
glog.Errorf("(%s) Invalid connection type", this.cid())
}
}
// For SUBSCRIBE message, we should add subscriber, then send back SUBACK
func (this *service) processSubscribe(msg *message.SubscribeMessage) error {
resp := message.NewSubackMessage()
resp.SetPacketId(msg.PacketId())
// Subscribe to the different topics
var retcodes []byte
topics := msg.Topics()
qos := msg.Qos()
this.rmsgs = this.rmsgs[0:0]
for i, t := range topics {
rqos, err := this.topicsMgr.Subscribe(t, qos[i], &this.onpub)
if err != nil {
return err
}
this.sess.AddTopic(string(t), qos[i])
retcodes = append(retcodes, rqos)
// yeah I am not checking errors here. If there's an error we don't want the
// subscription to stop, just let it go.
this.topicsMgr.Retained(t, &this.rmsgs)
glog.Debugf("(%s) topic = %s, retained count = %d", this.cid(), string(t), len(this.rmsgs))
}
if err := resp.AddReturnCodes(retcodes); err != nil {
return err
}
if _, err := this.writeMessage(resp); err != nil {
return err
}
for _, rm := range this.rmsgs {
if err := this.publish(rm, nil); err != nil {
glog.Errorf("service/processSubscribe: Error publishing retained message: %v", err)
return err
}
}
return nil
}
// FIXME: The order of closing here causes panic sometimes. For example, if receiver
// calls this, and closes the buffers, somehow it causes buffer.go:476 to panid.
func (this *service) stop() {
defer func() {
// Let's recover from panic
if r := recover(); r != nil {
glog.Errorf("(%s) Recovering from panic: %v", this.cid(), r)
}
}()
doit := atomic.CompareAndSwapInt64(&this.closed, 0, 1)
if !doit {
return
}
// Close quit channel, effectively telling all the goroutines it's time to quit
if this.done != nil {
glog.Debugf("(%s) closing this.done", this.cid())
close(this.done)
}
// Close the network connection
if this.conn != nil {
glog.Debugf("(%s) closing this.conn", this.cid())
this.conn.Close()
}
this.in.Close()
this.out.Close()
// Wait for all the goroutines to stop.
this.wgStopped.Wait()
glog.Debugf("(%s) Received %d bytes in %d messages.", this.cid(), this.inStat.bytes, this.inStat.msgs)
glog.Debugf("(%s) Sent %d bytes in %d messages.", this.cid(), this.outStat.bytes, this.outStat.msgs)
// Unsubscribe from all the topics for this client, only for the server side though
if !this.client && this.sess != nil {
topics, _, err := this.sess.Topics()
if err != nil {
glog.Errorf("(%s/%d): %v", this.cid(), this.id, err)
} else {
for _, t := range topics {
if err := this.topicsMgr.Unsubscribe([]byte(t), &this.onpub); err != nil {
glog.Errorf("(%s): Error unsubscribing topic %q: %v", this.cid(), t, err)
}
}
}
}
// Publish will message if WillFlag is set. Server side only.
if !this.client && this.sess.Cmsg.WillFlag() {
glog.Infof("(%s) service/stop: connection unexpectedly closed. Sending Will.", this.cid())
this.onPublish(this.sess.Will)
}
// Remove the client topics manager
if this.client {
topics.Unregister(this.sess.ID())
}
// Remove the session from session store if it's suppose to be clean session
if this.sess.Cmsg.CleanSession() && this.sessMgr != nil {
this.sessMgr.Del(this.sess.ID())
}
this.conn = nil
this.in = nil
this.out = nil
}
// processor() reads messages from the incoming buffer and processes them
func (this *service) processor() {
defer func() {
// Let's recover from panic
if r := recover(); r != nil {
//glog.Errorf("(%s) Recovering from panic: %v", this.cid(), r)
}
this.wgStopped.Done()
this.stop()
//glog.Debugf("(%s) Stopping processor", this.cid())
}()
glog.Debugf("(%s) Starting processor", this.cid())
this.wgStarted.Done()
for {
// 1. Find out what message is next and the size of the message
mtype, total, err := this.peekMessageSize()
if err != nil {
//if err != io.EOF {
glog.Errorf("(%s) Error peeking next message size: %v", this.cid(), err)
//}
return
}
msg, n, err := this.peekMessage(mtype, total)
if err != nil {
//if err != io.EOF {
glog.Errorf("(%s) Error peeking next message: %v", this.cid(), err)
//}
return
}
//glog.Debugf("(%s) Received: %s", this.cid(), msg)
this.inStat.increment(int64(n))
// 5. Process the read message
err = this.processIncoming(msg)
if err != nil {
if err != errDisconnect {
glog.Errorf("(%s) Error processing %s: %v", this.cid(), msg.Name(), err)
} else {
return
}
}
// 7. We should commit the bytes in the buffer so we can move on
_, err = this.in.ReadCommit(total)
if err != nil {
if err != io.EOF {
glog.Errorf("(%s) Error committing %d read bytes: %v", this.cid(), total, err)
}
return
}
// 7. Check to see if done is closed, if so, exit
if this.isDone() && this.in.Len() == 0 {
return
}
//if this.inStat.msgs%1000 == 0 {
// glog.Debugf("(%s) Going to process message %d", this.cid(), this.inStat.msgs)
//}
}
}
func (this *service) processAcked(ackq *sessions.Ackqueue) {
for _, ackmsg := range ackq.Acked() {
// Let's get the messages from the saved message byte slices.
msg, err := ackmsg.Mtype.New()
if err != nil {
glog.Errorf("process/processAcked: Unable to creating new %s message: %v", ackmsg.Mtype, err)
continue
}
if _, err := msg.Decode(ackmsg.Msgbuf); err != nil {
glog.Errorf("process/processAcked: Unable to decode %s message: %v", ackmsg.Mtype, err)
continue
}
ack, err := ackmsg.State.New()
if err != nil {
glog.Errorf("process/processAcked: Unable to creating new %s message: %v", ackmsg.State, err)
continue
}
if _, err := ack.Decode(ackmsg.Ackbuf); err != nil {
glog.Errorf("process/processAcked: Unable to decode %s message: %v", ackmsg.State, err)
continue
}
//glog.Debugf("(%s) Processing acked message: %v", this.cid(), ack)
// - PUBACK if it's QoS 1 message. This is on the client side.
// - PUBREL if it's QoS 2 message. This is on the server side.
// - PUBCOMP if it's QoS 2 message. This is on the client side.
// - SUBACK if it's a subscribe message. This is on the client side.
// - UNSUBACK if it's a unsubscribe message. This is on the client side.
switch ackmsg.State {
case message.PUBREL:
// If ack is PUBREL, that means the QoS 2 message sent by a remote client is
// releassed, so let's publish it to other subscribers.
if err = this.onPublish(msg.(*message.PublishMessage)); err != nil {
glog.Errorf("(%s) Error processing ack'ed %s message: %v", this.cid(), ackmsg.Mtype, err)
}
case message.PUBACK, message.PUBCOMP, message.SUBACK, message.UNSUBACK, message.PINGRESP:
glog.Debugf("process/processAcked: %s", ack)
// If ack is PUBACK, that means the QoS 1 message sent by this service got
// ack'ed. There's nothing to do other than calling onComplete() below.
// If ack is PUBCOMP, that means the QoS 2 message sent by this service got
// ack'ed. There's nothing to do other than calling onComplete() below.
// If ack is SUBACK, that means the SUBSCRIBE message sent by this service
// got ack'ed. There's nothing to do other than calling onComplete() below.
// If ack is UNSUBACK, that means the SUBSCRIBE message sent by this service
// got ack'ed. There's nothing to do other than calling onComplete() below.
// If ack is PINGRESP, that means the PINGREQ message sent by this service
// got ack'ed. There's nothing to do other than calling onComplete() below.
err = nil
default:
glog.Errorf("(%s) Invalid ack message type %s.", this.cid(), ackmsg.State)
continue
}
// Call the registered onComplete function
if ackmsg.OnComplete != nil {
onComplete, ok := ackmsg.OnComplete.(OnCompleteFunc)
if !ok {
glog.Errorf("process/processAcked: Error type asserting onComplete function: %v", reflect.TypeOf(ackmsg.OnComplete))
} else if onComplete != nil {
if err := onComplete(msg, ack, nil); err != nil {
glog.Errorf("process/processAcked: Error running onComplete(): %v", err)
}
}
}
}
}
func startMeshClient(t testing.TB, cid int, wg *sync.WaitGroup) {
runClientTest(t, cid, wg, func(svc *service.Client) {
done2 := make(chan struct{})
cnt := messages
expected := publishers * cnt
received := 0
sent := 0
now := time.Now()
since := time.Since(now).Nanoseconds()
sub := newSubscribeMessage("test", 0)
svc.Subscribe(sub,
func(msg, ack message.Message, err error) error {
subs := atomic.AddInt64(&subdone, 1)
if subs == int64(publishers) {
close(done)
}
return nil
},
func(msg *message.PublishMessage) error {
if received == 0 {
now = time.Now()
}
received++
//glog.Debugf("(surgemq%d) messages received=%d", cid, received)
since = time.Since(now).Nanoseconds()
if received == expected {
close(done2)
}
return nil
})
select {
case <-done:
case <-time.After(time.Second * time.Duration(publishers)):
glog.Infof("(surgemq%d) Timed out waiting for subscribe response", cid)
return
}
payload := make([]byte, size)
msg := message.NewPublishMessage()
msg.SetTopic(topic)
msg.SetQoS(qos)
go func() {
now := time.Now()
for i := 0; i < cnt; i++ {
binary.BigEndian.PutUint32(payload, uint32(cid*cnt+i))
msg.SetPayload(payload)
err := svc.Publish(msg, nil)
if err != nil {
break
}
sent++
}
since := time.Since(now).Nanoseconds()
statMu.Lock()
totalSent += int64(sent)
totalSentTime += int64(since)
if since > sentSince {
sentSince = since
}
statMu.Unlock()
glog.Debugf("(surgemq%d) Sent %d messages in %d ns, %d ns/msg, %d msgs/sec", cid, sent, since, int(float64(since)/float64(cnt)), int(float64(sent)/(float64(since)/float64(time.Second))))
}()
select {
case <-done2:
case <-time.Tick(time.Second * time.Duration(nap*publishers)):
glog.Errorf("Timed out waiting for messages to be received.")
}
statMu.Lock()
totalRcvd += int64(received)
totalRcvdTime += int64(since)
if since > rcvdSince {
rcvdSince = since
}
statMu.Unlock()
glog.Debugf("(surgemq%d) Received %d messages in %d ns, %d ns/msg, %d msgs/sec", cid, received, since, int(float64(since)/float64(cnt)), int(float64(received)/(float64(since)/float64(time.Second))))
})
}
func startFanSubscribers(t testing.TB, cid int, wg *sync.WaitGroup) {
now := time.Now()
runClientTest(t, cid, wg, func(svc *service.Client) {
cnt := messages * publishers
received := 0
since := time.Since(now).Nanoseconds()
sub := newSubscribeMessage("test", 0)
svc.Subscribe(sub,
func(msg, ack message.Message, err error) error {
subs := atomic.AddInt64(&subdone, 1)
if subs == int64(subscribers) {
now = time.Now()
close(done)
}
return nil
},
func(msg *message.PublishMessage) error {
if received == 0 {
now = time.Now()
}
received++
//glog.Debugf("(surgemq%d) messages received=%d", cid, received)
since = time.Since(now).Nanoseconds()
if received == cnt {
rcvd := atomic.AddInt64(&rcvdone, 1)
if rcvd == int64(subscribers) {
close(done2)
}
}
return nil
})
select {
case <-done:
case <-time.After(time.Second * time.Duration(subscribers)):
glog.Infof("(surgemq%d) Timed out waiting for subscribe response", cid)
return
}
select {
case <-done2:
case <-time.Tick(time.Second * time.Duration(nap*publishers)):
glog.Errorf("Timed out waiting for messages to be received.")
}
statMu.Lock()
totalRcvd += int64(received)
totalRcvdTime += int64(since)
if since > rcvdSince {
rcvdSince = since
}
statMu.Unlock()
glog.Debugf("(surgemq%d) Received %d messages in %d ns, %d ns/msg, %d msgs/sec", cid, received, since, int(float64(since)/float64(cnt)), int(float64(received)/(float64(since)/float64(time.Second))))
})
}