func GenerateJobsCycle() {
hostname, err := os.Hostname()
if err != nil {
log.Fatalf("Could not get hostname: %s", err.Error())
}
log.Print("Initial select from RunQueue and starting launcher goroutines")
if err := selectRQAndNotify(); err != nil {
log.Fatalf("Could not do initial select run queue: %s", err.Error())
}
log.Print("Initial select from Timetable")
ttRows, err := selectTimetable()
if err != nil {
log.Fatalf("Could not do initial select timetable: %s", err.Error())
}
log.Print("Starting jobgen goroutines")
if err := notifyForFullTTSelect(ttRows, false); err != nil {
log.Fatalf("Could notify about timetable: %s", err.Error())
}
for {
res, err := db.LockCycle(getLockName(), hostname)
if err != nil || !res {
if err == nil {
log.Println("Could not get lock, another host holds it? Retrying in 10 seconds")
} else {
log.Warningf("Could not get lock, got DB error: ", err.Error())
}
time.Sleep(time.Second * 10)
continue
}
// timer := pinba.TimerStart(map[string]string{"group": "jobgenerator"})
startTs := time.Now().UnixNano()
db.LogCycleStart(CYCLE_CLASS_NAME, hostname, 0)
log.Debug("Cycle started")
success := doCycle()
log.Debug("Cycle finished")
successInt := 1
if !success {
successInt = 0
}
db.LogCycleStop(CYCLE_CLASS_NAME, hostname, 0, successInt)
passedMs := int64((time.Now().UnixNano() - startTs) / 1e6)
if passedMs < cycleMs {
time.Sleep(time.Duration(cycleMs-passedMs) * time.Millisecond)
}
}
}
func NewProcess(id uint64, bin string, cmd string, args []string, stdout_file, stderr_file string) (*Vproc, error) {
p := new(Vproc)
p.Id = id
p.Cmd = exec.Command(bin, cmd)
var a = []string{bin, cmd}
a = append(a, args...)
p.Cmd.Args = a
p.Cmd.Env = os.Environ()
log.Debug("%v\n", a)
stdout, err := os.OpenFile(stdout_file, os.O_CREATE|os.O_RDWR|os.O_TRUNC, 0666)
if err != nil {
return nil, err
}
p.Cmd.Stdout = stdout
stderr, err := os.OpenFile(stderr_file, os.O_CREATE|os.O_RDWR|os.O_TRUNC, 0666)
if err != nil {
return nil, err
}
p.Cmd.Stderr = stderr
return p, nil
}
func main() {
service.Initialize("conf/phproxyd.conf", &config)
q, w, e := syscall.Syscall(syscall.SYS_GETPRIORITY, 1, 0, 0)
log.Debug("GetPriority: %v %v %v", q, w, e)
sh = newShardedCache(32)
service.EventLoop(ports)
}
func FinalizeRestartWithSuccess() {
log.Debug("restarted successfuly")
currentRestart.Pidfile.CloseAndRemove()
currentRestart = nil
}
// Call this when you want to start your servers and stuff
func EventLoop(ports []Port) {
defer log.Debug("exiting")
initPhaseDuration = time.Since(startupTime)
daemonConfig := config.GetDaemonConfig()
// service-stats ports
ports = append(ports, GpbPort("service-stats-gpb", stats_ctx, badoo_service.Gpbrpc))
ports = append(ports, JsonPort("service-stats-gpb/json", stats_ctx, badoo_service.Gpbrpc))
// build map of ports and do some sanity checks
ph := make(map[string]*Port)
for i := 0; i < len(ports); i++ {
p := &ports[i]
ph[p.Name] = p
// json and gpb ports should have the same context
// so try and warn user about passing plain values in (as it makes a copy)
if reflect.ValueOf(p.Handler).Kind() != reflect.Ptr {
log.Infof("port[%d].Handler should be a pointer (you want gpbs and json to use the same context, right?) (now: %T)", i, p.Handler)
}
}
getRestartSocket := func(rcd *RestartChildData, portName, portAddr string) (*RestartSocket, *os.File) {
if rcd == nil {
return nil, nil
}
restartSocket, exists := rcd.GpbrpcSockets[portName]
if exists == false {
return nil, nil
}
restartFile := os.NewFile(restartSocket.Fd, "")
if restartSocket.Address != portAddr {
return nil, restartFile
}
return &restartSocket, restartFile
}
// start 'em all
for _, lcf := range daemonConfig.GetListen() {
portName, portAddr := lcf.GetProto(), lcf.GetAddress()
port := ph[portName]
if nil == port {
log.Warnf("ignoring unknown port: %s at %s", portName, portAddr)
continue
}
if port.IsStarted {
log.Warnf("ignoring double startup for port: %s at %s", portName, portAddr)
}
listener, err := func() (listener net.Listener, err error) { // it's important that this should be a function, see defer inside
restartSocket, restartFile := getRestartSocket(restartData, portName, portAddr)
// this whole fd/file affair is very inconvenient to
// since when getRestartSocket() returns fd - it can't close it yet, as it can be used by FileListener
defer restartFile.Close()
if restartSocket == nil {
listener, err = net.Listen("tcp", portAddr)
if err != nil {
log.Errorf("listen failed for server %s at %s: %s", portName, portAddr, err)
return
}
log.Infof("port %s bound to address %s", portName, listener.Addr())
} else {
listener, err = net.FileListener(restartFile) // this dup()-s
if err != nil {
log.Errorf("failed to grab parent fd %d for %s at %s: %s", restartSocket.Fd, portName, portAddr, err)
return
}
log.Infof("port %s bound to address %s (parent fd: %d)", portName, listener.Addr(), restartSocket.Fd)
}
return
}()
if err != nil {
os.Exit(1)
}
// enable pinba only for ports that explicitly request it
ps := func() gpbrpc.PinbaSender {
if !lcf.GetPinbaEnabled() {
return nil // explicit nil here
}
if pinbaSender == nil {
log.Warnf("pinba is not configured, but pinba_enabled IS set for port %s: %s", portName, portAddr)
return nil // explicit nil here
}
log.Infof("pinba configured for port %s:%s -> %s", portName, portAddr, pinbaSender.Address)
return pinbaSender
}()
// slow request log time
slowRequestTime := time.Duration(daemonConfig.GetSlowRequestMs()) * time.Millisecond
srv := &Server{
Name: lcf.GetProto(),
Address: lcf.GetAddress(),
Server: gpbrpc.NewServer(listener, port.Proto, port.Codec, port.Handler, ps, slowRequestTime),
}
go srv.Server.Serve()
port.IsStarted = true
StartedServers[port.Name] = srv // save it for laterz
}
// kill parent if this is a child of graceful restart
if restartData != nil {
syscall.Kill(restartData.PPid, syscall.SIGQUIT)
}
log.Infof("entering event loop")
exitMethod := wait_for_signals()
if exitMethod == EXIT_GRACEFULLY { // wait for established connections to close and then die
// FIXME: should stop servers from accepting new connections here!
const ATTEMPTS_PER_SEC = 2
maxAttempts := daemonConfig.GetParentWaitTimeout() * ATTEMPTS_PER_SEC
for i := uint32(0); i < maxAttempts; i++ {
for _, srv := range StartedServers {
currConn := atomic.LoadUint64(&srv.Server.Stats.ConnCur)
if currConn > 0 {
log.Debugf("%s still has %d connections", srv.Name, currConn)
time.Sleep(time.Second / ATTEMPTS_PER_SEC)
}
}
}
} else {
// do nothing for EXIT_IMMEDIATELY
}
// doing cleanups here
// XXX: can this be moved to defer at the start of this function?
if pidfile != nil {
pidfile.CloseAndRemove()
}
}