func NewClient(address string, p Protocol, c ClientCodec, connect_timeout, request_timeout time.Duration) *Client {
ips, err := dns.LookupHostPort(address)
if err != nil {
log.Errorf("dns.LookupHostPort() faield: %v", err)
// FIXME(antoxa): just reusing existing ips here, which actually sucks
// this only works because cli.Call() uses net.Dial() which resolves the name again
}
canUseClient := func(client *Client) bool {
// readLoop() might be modifying this conn
// but don't really need to lock for ips comparison, since ips are never modified for existing client
client.lk.Lock()
defer client.lk.Unlock()
// TODO(antoxa): can just use one ip for client and recheck not full equality
// but only if new ips contain old ip
if !util.StrSliceEqual(client.ips, ips) {
return false
}
if client.closed {
return false
}
return true
}
const max_tries = 3 // arbitrary limit, i know
for done_tries := 0; done_tries < max_tries; done_tries++ {
client := Pcm.GetClient(address)
if client == nil {
break
}
if !canUseClient(client) {
client.closeNoReuse()
continue
}
log.Debugf("reused existing client %p for %s (after %d tries)", client, address, done_tries)
return client
}
log.Debugf("creating new cli for %s", address)
return &Client{
address: address,
ips: ips,
Proto: p,
Codec: c,
connect_timeout: connect_timeout,
request_timeout: request_timeout,
}
}
func (d *DispatcherData) processAllJobs(jobs []*TimetableEntry) {
newJobs := make([]*TimetableEntry, 0)
toProcessFinished := make([]*FinishEvent, 0)
for _, row := range jobs {
if el, ok := d.waitingMap[row.id]; ok {
if row.finish_count >= el.finish_count && row.added_to_queue_ts.Valid && !row.finished_ts.Valid {
d.removeFromWaiting(el)
row.reportedDup = el.reportedDup
d.addToAdded(row)
log.Warnf("external waiting to added promotion OLD:%+v---NEW:%+v", el, row)
} else if row.finish_count > el.finish_count && !row.added_to_queue_ts.Valid {
d.removeFromWaiting(el)
d.addToWaiting(row)
log.Warnf("external finish_count promotion OLD:%+v---NEW:%+v", el, row)
}
continue
}
if el, ok := d.addedMap[row.id]; ok {
// External job notification can update timetable by incrementing it's finish count.
// We do not really care about any modifications made by the script except for finished_successfully field
if row.finish_count > el.finish_count {
log.Debugf("External finish for tt row %+v", row)
toProcessFinished = append(toProcessFinished, &FinishEvent{
timetable_id: row.id, success: row.finished_successfully != 0,
havePrevFinishCount: true, prevFinishCount: row.finish_count, errorCh: make(chan error, 1)})
}
} else if d.deletedIds[row.id] == 0 {
log.Debugf("External timetable row %+v", row)
newJobs = append(newJobs, row)
}
}
if len(newJobs) > 0 {
d.acceptNewJobs(newJobs)
}
for _, el := range toProcessFinished {
log.Debugf("processFinished: %+v", el)
d.processFinished(el)
}
for ttId, refCount := range d.deletedIds {
if refCount--; refCount <= 0 {
delete(d.deletedIds, ttId)
} else {
d.deletedIds[ttId] = refCount
}
}
}
func (client *Client) establishConnection(network string) error {
client.lk.Lock()
defer client.lk.Unlock()
// conn might've been closed in between Call*() invocations
if client.closed {
client.closed = false
client.conn = nil
}
// already established, can reuse
if client.conn != nil {
return nil
}
// we're creating new connection here; (closed == false && conn == nil)
// TODO(antoxa): rewrite this to use ips, that we've resolved earlier
conn, err := net.DialTimeout(network, client.address, client.connect_timeout)
if err != nil {
client.closed = true // connection attempt has failed -> do not reuse
return fmt.Errorf("connect error: %v", err)
}
log.Debugf("connected %s -> %s", conn.LocalAddr(), conn.RemoteAddr())
client.conn = conn
client.respch = make(chan response, 1) // needs to be buffered, to avoid stalling readLoop() on Write failure
client.numExpectedResponses = 0 // this is needed to reset counter in case last Write has failed
go client.readLoop()
return nil
}
func restartRunChild(childData *RestartChildData) (*RestartContext, error) {
if currentRestart != nil {
panic("can't call this function when restart is already in progress")
}
child_data_json, err := json.Marshal(childData)
if err != nil {
return nil, fmt.Errorf("can't json encode child data: %v", err)
}
os.Setenv(RESTART_ENV_VAR, string(child_data_json))
log.Debugf("%s = %s", RESTART_ENV_VAR, child_data_json)
// can start child now
child, err := forkExec(childData.files)
if err != nil {
return nil, fmt.Errorf("forkExec error: %v", err)
}
log.Debugf("started child: %d", child.Pid)
os.Setenv(RESTART_ENV_VAR, "") // reset env after child starts, just in case
// save state
rctx := &RestartContext{
Child: child,
ChildWaitC: make(chan RestartProcStatus, 1), // needs to be buffered (in case we drop restart state before goroutine has the chance to write)
ChildTimeoutC: time.After(5 * time.Second), // FIXME: make this configureable?
}
// start child wait goroutine, this goroutine never dies if child starts up successfuly
// but it doesn't matter, since this process will exit soon in that case
go func(rctx *RestartContext) {
status, err := rctx.Child.Wait()
rctx.ChildWaitC <- RestartProcStatus{status, err}
}(rctx)
return rctx, nil
}
func (d *DispatcherData) checkZero(src string) {
if len(d.addedMap) == 0 && d.waitingList.Len() == 0 {
log.Debugf("No rows left in class=%s, location=%s (%s)", d.className, d.location, src)
trigger(d.zeroTTCh, "zerott")
if d.killRequest != nil {
log.Printf("Killed all jobs in class=%s, location=%s, waiting on continue channel", d.className, d.location)
d.killRequest.ResCh <- nil
d.killRequest = nil
log.Printf("Can continue dispatching in class=%s, location=%s", d.className, d.location)
}
}
}
func loadFullState(funcs ...*LoadStateFunc) (err error) {
for _, funEntry := range funcs {
startTs := time.Now().UnixNano()
err = funEntry.fun()
if err != nil {
log.Errorf("Could not load %s: %s", funEntry.name, err.Error())
return err
}
log.Debugf("Selected from %s for %.5f sec", funEntry.name, float64(time.Now().UnixNano()-startTs)/1e9)
}
return nil
}
func ParseRestartDataFromEnv() (*RestartChildData, error) {
rcd := &RestartChildData{}
rcd_env := os.Getenv(RESTART_ENV_VAR)
if rcd_env == "" {
return nil, nil // ok, but no data is set
}
log.Debugf("ParseRestartDataFromEnv; %s: %s", RESTART_ENV_VAR, rcd_env)
if err := json.Unmarshal([]byte(rcd_env), rcd); err != nil {
return nil, err
}
if rcd.PPid <= 0 {
return nil, fmt.Errorf("restart_data.Ppid <= 0 (%d)", rcd.PPid)
}
if rcd.GpbrpcSockets == nil {
return nil, fmt.Errorf("restart_data.GpbrpcSockets == nil")
}
return rcd, nil
}
func initSyslogHook() error {
if syslogHookInited {
return nil
}
if DaemonConfig().SyslogIdentity != nil {
syslogIdentity := parseSyslogIdentity(DaemonConfig().GetSyslogIdentity())
log.Debugf("syslog_identity: <%s>", syslogIdentity)
syslogAddr := fmt.Sprintf("%s:%d", DaemonConfig().GetSyslogIp(), DaemonConfig().GetSyslogPort())
network := func() string {
if DaemonConfig().GetSyslogIp() != "" && DaemonConfig().GetSyslogPort() != 0 {
return "udp"
}
return ""
}()
hook, err := syslog_hook.NewSyslogHookNoFrozen(network, syslogAddr, syslog.LOG_LOCAL6, syslogIdentity)
if err != nil {
return err
}
log.AddHook(hook)
if DaemonConfig().GetSyslogSendBufferSize() != 0 {
log.Warn("unsupported config option syslog_send_buffer_size, ignored")
}
}
syslogHookInited = true
return nil
}
func doCycle() bool {
var (
jiRows map[string]map[string]*JobInfoEntry
scripts map[string]*ScriptEntry
flags map[string]*FlagEntry
scriptsRusage map[string]*ScriptRusageEntry
classLocTTRows map[string]map[string][]*TimetableEntry
)
unifiedStartTs := time.Now().UnixNano()
startTs := time.Now().UnixNano()
err := loadFullState(
&LoadStateFunc{name: "Scripts", fun: func() (err error) { scripts, err = getGroupedScriptsForPlatform(); return }},
&LoadStateFunc{name: "JobInfo", fun: func() (err error) { jiRows, err = getGroupedJobInfo(); return }},
&LoadStateFunc{name: "Flags", fun: func() (err error) { flags, err = getFlags(); return }},
&LoadStateFunc{name: "ScriptsRusage", fun: func() (err error) { scriptsRusage, err = getScriptRusageStats(); return }},
&LoadStateFunc{name: "ScriptTimetable", fun: func() (err error) { classLocTTRows, err = selectTimetable(); return }})
if err != nil {
log.Errorf("Failed to select state in doCycle: %s", err.Error())
return false
}
log.Debugf("Loaded for %.5f sec", float64(time.Now().UnixNano()-startTs)/1e9)
startTs = time.Now().UnixNano()
err = loadSettingsFromRows(jiRows, scripts)
if err != nil {
log.Errorf("Could not load settings from rows: %s", err.Error())
return false
}
func() {
allSettingsMutex.Lock()
defer allSettingsMutex.Unlock()
for _, row := range scripts {
row.settings = allSettings[row.settings_id]
}
}()
scriptsMap.Lock()
scriptsMap.v = scripts
scriptsMap.Unlock()
log.Debugf(" Selected %d rows from flags", len(flags))
log.Debugf(" Selected %d rows from scripts rusage", len(scriptsRusage))
log.Debugf("Load settings for %.5f sec", float64(time.Now().UnixNano()-startTs)/1e9)
startTs = time.Now().UnixNano()
// We should not try to generate jobs for scripts that are not present in Script table
// But we should not forget settings (e.g. last generation_id) for that script
for class_name := range jiRows {
if _, ok := scripts[class_name]; !ok {
delete(jiRows, class_name)
}
}
log.Debugf("Selected all for %.5f sec", float64(time.Now().UnixNano()-unifiedStartTs)/1e9)
startTs = time.Now().UnixNano()
updateLoadEstimates()
log.Debugf("Load estimates updated for %.5f sec", float64(time.Now().UnixNano()-startTs)/1e9)
func() {
rusageInfo.Lock()
defer rusageInfo.Unlock()
log.Debugf("Group hosts: %+v", rusageInfo.groupHosts)
}()
startTs = time.Now().UnixNano()
failedLocationsMutex.Lock()
failedLocations = make(map[string]bool)
failedLocationsMutex.Unlock()
success := true
if len(scripts) > 0 {
throttle.setIntervalCh <- time.Second / time.Duration(len(scripts))
}
trigger(throttle.c, "throttle, start of cycle")
for className, script := range scripts {
<-throttle.c
tx := new(db.LazyTrx)
err := tx.Begin()
if err != nil {
log.Errorf("Could not start transaction in job generate: %s", err.Error())
success = false
continue
}
have := make(map[string]bool)
locTtRows := classLocTTRows[className]
if locTtRows != nil {
for rawLoc, v := range locTtRows {
loc, err := getLocationIdx(script.settings.location_type, rawLoc)
if err != nil {
log.Warningf("Broken settings for class %s: %s", className, err.Error())
loc = rawLoc
}
if len(v) > 0 {
have[loc] = true
}
}
}
add_to_timetable, err := generateJobs(tx, className, script.settings, jiRows[className], have, flags[className])
if err != nil {
log.Errorf("Could generate jobs for class %s: %s", className, err.Error())
tx.Rollback()
success = false
continue
}
err = tx.Commit()
if err != nil {
log.Errorf("Could not commit generate jobs for class %s: %s", className, err.Error())
success = false
continue
}
per_location := make(map[string][]*TimetableEntry)
for _, row := range add_to_timetable {
allSettingsMutex.Lock()
row.settings = allSettings[row.settings_id]
allSettingsMutex.Unlock()
if row.settings == nil {
log.Warningf("Internal inconsistency error: Invalid settings for generated row: %+v", row)
continue
}
key := DEFAULT_LOCATION_IDX
if row.settings.location_type == LOCATION_TYPE_EACH {
key = row.location
}
if _, ok := per_location[key]; !ok {
per_location[key] = make([]*TimetableEntry, 0)
}
per_location[key] = append(per_location[key], row)
}
for location, rows := range per_location {
notifyAboutNewTTRows(className, location, rows, true)
}
}
notifyForFullTTSelect(classLocTTRows, true)
log.Debugf("Processed %d classes for %.5f sec", len(scripts), float64(time.Now().UnixNano()-startTs)/1e9)
log.Debugf("Total %.5f sec", float64(time.Now().UnixNano()-unifiedStartTs)/1e9)
return success
}
func (d *DispatcherData) printTrace() {
trace := make([]byte, 8192)
n := runtime.Stack(trace, false)
log.Debugf("class:%s location:%s\nStack trace: %s\n", d.className, d.location, trace[0:n])
}
func (server *Server) serveConnection(conn net.Conn) {
log.Debugf("accepted connection: %s -> %s", conn.LocalAddr(), conn.RemoteAddr())
defer func() {
atomic.AddUint64(&server.Stats.ConnCur, ^uint64(0)) // decrements the value, ref: http://golang.org/pkg/sync/atomic/#AddUint64
log.Debugf("closing connection: %s -> %s", conn.LocalAddr(), conn.RemoteAddr())
if server.onDisconnect != nil {
server.onDisconnect(RequestT{
Server: server,
Conn: conn,
})
}
conn.Close()
}()
atomic.AddUint64(&server.Stats.ConnCur, 1)
atomic.AddUint64(&server.Stats.ConnTotal, 1)
for {
// need these as var here, since goto might jump over their definition below
var (
result ResultT
request RequestT
startTime time.Time
pinbaReq pinba.Request
requestTime time.Duration
)
msgid, msg, bytes_read, status, err := server.Codec.ReadRequest(server.Proto, conn)
if err != nil {
if status == ConnEOF {
// connection was closed gracefully from client side
break
}
if status == ConnOK {
// misunderstanding on protobuf level
result = server.Proto.ErrorGeneric(err)
// FIXME(antoxa): remove this goto, it's annoying to code around
goto write_response
}
// IO error or similar
log.Infof("aborting connection: %v", err)
break
}
// FIXME(antoxa): stats will not be incremented when ReadRequest returns err != nil and status == ConnOK
atomic.AddUint64(&server.Stats.BytesRead, uint64(bytes_read))
atomic.AddUint64(&server.Stats.Requests, 1)
request = RequestT{
Server: server,
Conn: conn,
RequestId: atomic.AddUint64(&globalRequestId, 1),
MessageId: msgid,
Message: msg,
PinbaReq: &pinbaReq,
}
if msgid < MaxMsgID {
atomic.AddUint64(&server.Stats.RequestsIdStat[msgid], 1)
}
startTime = time.Now()
result = server.Proto.Dispatch(request, server.Handler)
requestTime = time.Since(startTime)
// log slow request if needed
if server.slowRequestTime > 0 && server.slowRequestTime <= requestTime {
if msg != nil {
requestInfo := func() string {
reqName := MapRequestIdToName(server.Proto, msgid)
body, err := json.Marshal(msg)
if err != nil {
return fmt.Sprintf("%s %v", reqName, err)
}
return fmt.Sprintf("%s %s", reqName, body)
}()
log.Warnf("slow request (%d ms): %s; addr: %s <- %s", requestTime/time.Millisecond, requestInfo, conn.LocalAddr(), conn.RemoteAddr())
}
}
server.sendToPinba(&pinbaReq, server.pinbaReqNames[msgid], requestTime)
write_response:
if result.Action == ACTION_RESULT {
writeln, err := server.Codec.WriteResponse(server.Proto, conn, result.Message)
if err != nil {
// write error: can't recover
log.Infof("unrecoverable error while writing: %v", err)
break
}
atomic.AddUint64(&server.Stats.BytesWritten, uint64(writeln))
}
if result.Action == ACTION_CLOSE {
break
}
}
}
func Initialize(default_config_path string, service_conf Config) {
flag.StringVar(&flags.ConfFile, "c", default_config_path, "path to config file")
flag.StringVar(&flags.LogFile, "l", "", "path to log file, special value '-' means 'stdout'")
flag.StringVar(&flags.PidFile, "p", "", "path to pid file. if empty, pidfile will not be created")
flag.BoolVar(&flags.Testconf, "t", false, "test configuration and exit")
flag.BoolVar(&flags.Version, "v", false, "print version")
flag.BoolVar(&flags.FullVersion, "V", false, "print full version info")
flag.BoolVar(&flags.Debug, "debug", false, "force DEBUG log level")
flag.Parse()
if flags.Version {
fmt.Printf("%s\n", VersionInfo.GetVersion())
os.Exit(0)
}
if flags.FullVersion {
data, _ := json.MarshalIndent(VersionInfo, "", " ")
fmt.Printf("%s\n", data)
os.Exit(0)
}
var err error
config = service_conf // save a pointer to service's config (NOT a copy, mon!)
commandLine = strings.Join(os.Args, " ") // XXX(antoxa): couldn't think of a better way
hostname = getHostname() // get hostname early
// moved here from init(), just importing a package should not publish expvars
initExpvars()
// current executable full path (symlinks and shit sometimes complicate things)
binaryPath = func() string {
path, err := osutil.GetCurrentBinary()
if err != nil {
// log as notice, non-critical error (only stats affected)
log.Infof("couldn't get current binary (using argv[0] = %q): %v", os.Args[0], err)
return os.Args[0]
}
return path
}()
// config path
confPath := func() string {
if flags.ConfFile != "" {
return flags.ConfFile
}
return default_config_path
}()
// resolve absolute config path, convenient for stats
configPath = func(path string) string {
var err error
if path, err = filepath.Abs(path); err != nil {
return path
}
if path, err = filepath.EvalSymlinks(path); err != nil {
return path
}
return path
}(confPath)
// parse config and construct final config merged with command line flags
// use path as supplied to us in args (i.e. unresolved), just to avoid 'too smart, outsmarted yourself' gotchas
err = ParseConfigFromFile(confPath, service_conf)
if err != nil {
err_message := func(err error) string {
switch real_err := err.(type) {
case nil:
return "syntax is ok"
case *json.SyntaxError:
return fmt.Sprintf("%v at offset %d", real_err, real_err.Offset)
case *os.PathError:
return fmt.Sprintf("%v", real_err)
default:
return fmt.Sprintf("(%T) %v", real_err, real_err)
}
}(err)
stderrLogger.Fatalf("Error in config: %s", err_message)
} else {
if flags.Testconf {
fmt.Printf("testconf %s: syntax is ok\n", configPath)
}
}
mergeCommandlineFlagsToConfig(flags, config)
daemonConfig := config.GetDaemonConfig()
// antoxa: need the fancy wrapper function to have testconf behave properly
// FIXME: testconf should check more stuff (below) and graceful restart also
initPidfileLogfile := func() error {
// FIXME(antoxa): this testconf thingy is everywhere! must... resist... full rewrite
if flags.Testconf {
err = PidfileTest(daemonConfig.GetPidFile())
} else {
pidfile, err = PidfileOpen(daemonConfig.GetPidFile())
}
if err != nil {
return fmt.Errorf("can't open pidfile: %s", err)
}
// FIXME: this is shit ugly
// need better integration between logger and daemon-config
// or 1-to-1 mapping
// or better log package :)
log_level := daemonConfig.GetLogLevel()
if log_level == 0 {
return fmt.Errorf("unknown log_level, supported: %v", badoo_config.ServiceConfigDaemonConfigTLogLevels_name)
}
err = reopenLogfile(daemonConfig.GetLogFile(), log.Level(log_level))
if err != nil {
return fmt.Errorf("can't open logfile: %s", err)
}
return nil
}
err = initPidfileLogfile()
if err != nil {
if flags.Testconf {
stderrLogger.Errorf("%v", err)
fmt.Printf("testconf failed\n")
} else {
stderrLogger.Errorf("%v", err) // always pidfile/logfile errors to stderr
}
os.Exit(1)
} else {
if flags.Testconf {
fmt.Printf("testconf successful\n")
os.Exit(0)
}
}
// log some version info like libangel does
versionString := func() string {
vi := &VersionInfo
version := func() string {
if vi.GetAutoBuildTag() != "" {
return fmt.Sprintf("%s-%s", vi.GetVersion(), vi.GetAutoBuildTag())
} else {
return vi.GetVersion()
}
}()
return fmt.Sprintf("%s version %s, git %s, built %s on %s",
vi.GetVcsBasename(), version, vi.GetVcsShortHash(), vi.GetBuildDate(), vi.GetBuildHost())
}()
log.Infof("%s", versionString)
// max cpus, 0 = all of them
numCPU := func() int {
maxCpus := int(daemonConfig.GetMaxCpus())
if maxCpus <= 0 || maxCpus > runtime.NumCPU() {
maxCpus = runtime.NumCPU()
}
return maxCpus
}()
runtime.GOMAXPROCS(numCPU)
// gc percent, <0 - disables GC
if daemonConfig.GcPercent != nil {
debug.SetGCPercent(int(daemonConfig.GetGcPercent()))
}
// process pinba configuration and related stuff
pinbaSender, err = func() (*PinbaSender, error) { // assigns a global
if daemonConfig.GetPinbaAddress() == "" {
return nil, nil // user doesn't want pinba configured
}
pi, err := PinbaInfoFromConfig(config)
if err != nil {
return nil, err
}
return NewPinbaSender(pi), nil
}()
if err != nil {
log.Fatalf("pinba config error: %v", err)
}
// graceful restart handling
// see restart.go and signals.go for more details
restartData, err = ParseRestartDataFromEnv()
if err != nil {
log.Fatalf("can't parse restart data: %v", err)
}
if restartData != nil {
log.Debugf("[CHILD] this is a restart, parent: %d, me: %d", restartData.PPid, os.Getpid())
}
// start http pprof server (possibly - inherit fd from parent if this is a restart)
err = func() (err error) {
HttpServer, err = newHttpServer(config, restartData) // assigning a global here
if err != nil {
return err
}
if HttpServer != nil { // nil here means it has not been configured
go HttpServer.Serve()
}
return nil
}()
if err != nil {
log.Fatalf("can't start http_pprof server: %v", err)
}
}
// 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()
}
}
