func (a *ConsumerActor) Running() dfa.Letter {
ticker := time.NewTicker(30 * time.Second)
defer ticker.Stop()
w := condition.NewNameWatch(a.grid.Etcd(), a.grid.Name(), a.flow.Name(), "finished")
defer w.Stop()
n := 0
finished := w.WatchUntil(ring.New(a.flow.NewContextualName("producer"), a.conf.NrProducers))
for {
select {
case <-a.exit:
return Exit
case <-a.chaos.C:
return Failure
case <-ticker.C:
if err := a.started.Alive(); err != nil {
return Failure
}
case <-finished:
if err := a.SendCounts(); err != nil {
return SendFailure
} else {
return EverybodyFinished
}
case err := <-w.WatchError():
log.Printf("%v: error: %v", a, err)
return Failure
case m := <-a.rx.Msgs():
switch m := m.(type) {
case DataMsg:
a.state.Counts[m.Producer]++
n++
if n%1000000 == 0 {
log.Printf("%v: received: %v", a.ID(), n)
}
if n%1000 == 0 {
if err := a.SendCounts(); err != nil {
return SendFailure
}
}
}
}
}
}
func (a *ProducerActor) Running() dfa.Letter {
ticker := time.NewTicker(30 * time.Second)
defer ticker.Stop()
a.state = NewProducerState()
s := condition.NewState(a.grid.Etcd(), 30*time.Minute, a.grid.Name(), a.flow.Name(), "state", a.ID())
defer s.Stop()
if err := s.Init(a.state); err != nil {
if _, err := s.Fetch(a.state); err != nil {
return FetchStateFailure
}
}
log.Printf("%v: running with state: %v, index: %v", a.ID(), a.state, s.Index())
// Make some random length string data.
data := NewDataMaker(a.conf.MsgSize, a.conf.MsgCount-a.state.SentMessages)
defer data.Stop()
r := ring.New(a.flow.NewContextualName("consumer"), a.conf.NrConsumers)
start := time.Now()
for {
select {
case <-a.exit:
if _, err := s.Store(a.state); err != nil {
log.Printf("%v: failed to save state: %v", a, err)
}
return Exit
case <-a.chaos.C:
if _, err := s.Store(a.state); err != nil {
log.Printf("%v: failed to save state: %v", a, err)
}
return Failure
case <-ticker.C:
if err := a.started.Alive(); err != nil {
return Failure
}
if _, err := s.Store(a.state); err != nil {
return Failure
}
case <-data.Done():
if err := a.tx.Flush(); err != nil {
return SendFailure
}
if err := a.tx.Send(a.flow.NewContextualName("leader"), &ResultMsg{Producer: a.ID(), Count: a.state.SentMessages, From: a.ID(), Duration: a.state.Duration}); err != nil {
return SendFailure
}
if _, err := s.Store(a.state); err != nil {
log.Printf("%v: failed to save state: %v", a, err)
}
return IndividualFinished
case d := <-data.Next():
if a.state.SentMessages%100 == 0 {
a.state.Duration += time.Now().Sub(start).Seconds()
start = time.Now()
}
if a.state.SentMessages%10000 == 0 {
if _, err := s.Store(a.state); err != nil {
log.Printf("%v: failed to save state: %v", a, err)
return Failure
}
}
if err := a.tx.SendBuffered(r.ByInt(a.state.SentMessages), &DataMsg{Producer: a.ID(), Data: d}); err != nil {
if _, err := s.Store(a.state); err != nil {
log.Printf("%v: failed to save state: %v", a, err)
}
return SendFailure
}
a.state.SentMessages++
}
}
}
func main() {
runtime.GOMAXPROCS(runtime.NumCPU())
flag.Parse()
go func() {
log.Println(http.ListenAndServe("localhost:6060", nil))
}()
etcdservers := strings.Split(*etcdconnect, ",")
natsservers := strings.Split(*natsconnect, ",")
conf := &Conf{
GridName: "flowgrid",
MsgSize: *msgsize,
MsgCount: *msgcount,
NrProducers: *producers,
NrConsumers: *consumers,
}
hostname, err := os.Hostname()
if err != nil {
log.Fatalf("error: failed to discover hostname: %v", err)
}
m, err := newActorMaker(conf)
if err != nil {
log.Fatalf("error: failed to make actor maker: %v", err)
}
g := grid.New(conf.GridName, etcdservers, natsservers, m)
exit, err := g.Start()
if err != nil {
log.Fatalf("error: failed to start grid: %v", err)
}
j := condition.NewJoin(g.Etcd(), 30*time.Second, g.Name(), "hosts", hostname)
err = j.Join()
if err != nil {
log.Fatalf("error: failed to regester: %v", err)
}
defer j.Exit()
go func() {
ticker := time.NewTicker(15 * time.Second)
defer ticker.Stop()
for {
select {
case <-exit:
return
case <-ticker.C:
err := j.Alive()
if err != nil {
log.Fatalf("error: failed to report liveness: %v", err)
}
}
}
}()
w := condition.NewCountWatch(g.Etcd(), g.Name(), "hosts")
defer w.Stop()
started := w.WatchUntil(*nodes)
select {
case <-exit:
log.Printf("Shutting down, grid exited")
return
case <-w.WatchError():
log.Fatalf("error: failed to watch other hosts join: %v", err)
case <-started:
}
for i := 0; i < *flows; i++ {
flow := NewFlow(i)
rp := ring.New(flow.NewContextualName("producer"), conf.NrProducers)
for _, def := range rp.ActorDefs() {
def.DefineType("producer")
def.Define("flow", flow.Name())
err := g.StartActor(def)
if err != nil {
log.Fatalf("error: failed to start: %v, due to: %v", def, err)
}
}
rc := ring.New(flow.NewContextualName("consumer"), conf.NrConsumers)
for _, def := range rc.ActorDefs() {
def.DefineType("consumer")
def.Define("flow", flow.Name())
err := g.StartActor(def)
if err != nil {
log.Fatalf("error: failed to start: %v, due to: %v", def, err)
}
}
def := grid.NewActorDef(flow.NewContextualName("leader"))
def.DefineType("leader")
def.Define("flow", flow.Name())
err = g.StartActor(def)
if err != nil {
log.Fatalf("error: failed to start: %v, due to: %v", "leader", err)
}
time.Sleep(5 * time.Second)
}
go func() {
sig := make(chan os.Signal, 1)
signal.Notify(sig, os.Interrupt, syscall.SIGTERM)
select {
case <-sig:
log.Printf("shutting down")
g.Stop()
case <-exit:
}
}()
<-exit
log.Println("shutdown complete")
}
