func runService(dc *deferer.Deferer, serviceDone chan struct{}, id int, ttl uint64, target, cmd, base, arg string) {
d := deferer.NewDeferer(dc)
defer d.Run()
conn, err := dbus.New()
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "dbus.New",
}, "error creating new dbus connection")
}
f, err := os.Create("/run/systemd/system/" + target)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "os.Create",
}, "error creating service file")
}
d.Defer(func() { _ = f.Close() })
arg = base64.StdEncoding.EncodeToString([]byte(arg))
dotService := fmt.Sprintf(service, base, cmd, arg, ttl)
_, err = f.WriteString(dotService)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "f.WriteString",
}, "error writing service file")
}
if err := f.Sync(); err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "f.Sync",
}, "error syncing service file")
}
log.WithFields(log.Fields{
"locker": fmt.Sprintf("locker-%s-%d", base, id),
"locked": fmt.Sprintf("locked-%s-%d", base, id),
}).Info("created service names")
done := make(chan string)
_, err = conn.StartUnit(target, "fail", done)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "conn.StartUnit",
"name": target,
}, "error starting service")
}
status := <-done
if status != "done" {
d.FatalWithFields(log.Fields{
"status": status,
"func": "StartUnit",
"name": target,
}, "StartUnit returned a bad status")
}
serviceDone <- struct{}{}
}
func runService(dc *deferer.Deferer, serviceDone chan struct{}, id int, name string, args []string) {
d := deferer.NewDeferer(dc)
defer d.Run()
conn, err := dbus.New()
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "dbus.New",
}, "error creating new dbus connection")
}
f, err := os.Create("/run/systemd/system/" + name)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "os.Create",
}, "error creating service file")
}
d.Defer(func() { _ = f.Close() })
base := filepath.Base(args[0])
// For args with spaces, quote 'em
for i, v := range args {
if strings.Contains(v, " ") {
args[i] = "'" + v + "'"
}
}
dotService := fmt.Sprintf(service, base, strings.Join(args, " "))
_, err = f.WriteString(dotService)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "f.WriteString",
"name": name,
}, "error writing service file")
}
if err := f.Sync(); err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "f.Sync",
"name": name,
}, "error syncing service file")
}
done := make(chan string)
_, err = conn.StartUnit(name, "fail", done)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "conn.StartUnit",
"name": name,
}, "error starting service")
}
status := <-done
if status != "done" {
d.FatalWithFields(log.Fields{
"status": status,
"func": "StartUnit",
}, "StartUnit returned a bad status")
}
serviceDone <- struct{}{}
}
func main() {
d := deferer.NewDeferer(nil)
defer d.Run()
rand.Seed(time.Now().UnixNano())
id := rand.Int()
if ID := os.Getenv("ID"); ID != "" {
fmt.Sscanf(ID, "%d", &id)
}
params := params{ID: id}
flag.Uint64VarP(¶ms.Interval, "interval", "i", 30, "Interval in seconds to refresh lock")
flag.Uint64VarP(¶ms.TTL, "ttl", "t", 0, "TTL for key in seconds, leave 0 for (2 * interval)")
flag.StringVarP(¶ms.Key, "key", "k", "/lock", "Key to use as lock")
flag.BoolVarP(¶ms.Blocking, "block", "b", false, "Block if we failed to acquire the lock")
flag.StringVarP(¶ms.Addr, "etcd", "e", defaultAddr, "address of etcd machine")
flag.Usage = func() {
fmt.Fprintf(os.Stderr, "Usage of %s: [options] -- command args\n", os.Args[0])
flag.PrintDefaults()
fmt.Fprintf(os.Stderr, "\ncommand will be run with args via fork/exec not a shell\n")
}
flag.Parse()
if params.TTL == 0 {
params.TTL = params.Interval * 2
}
params.Args = flag.Args()
if len(params.Args) < 1 {
d.Fatal("command is required")
}
cmd := resolveCommand(params.Args[0])
if cmd == "" {
d.Fatal("")
}
params.Args[0] = cmd
hostname, err := os.Hostname()
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "os.Hostname",
}, "failed to get hostname")
}
c := etcd.NewClient([]string{params.Addr})
l, err := lock.Acquire(c, params.Key, hostname, params.TTL, params.Blocking)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "lock.Acquire",
"lock": params.Key,
"ttl": params.TTL,
"blocking": params.Blocking,
}, "failed to get lock")
}
d.Defer(func() {
if err := l.Release(); err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "l.Release",
}, "failed to release lock")
}
})
params.Lock = l
args, err := json.Marshal(¶ms)
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "json.Marshal",
}, "failed to serialize params")
}
serviceDone := make(chan struct{})
base := filepath.Base(params.Args[0])
target := fmt.Sprintf("locker-%s-%d.service", base, id)
locker := resolveCommand("locker")
if locker == "" {
d.Fatal("")
}
go runService(d, serviceDone, params.ID, params.TTL, target, locker, base, string(args))
sigs := make(chan os.Signal)
signal.Notify(sigs, os.Interrupt, syscall.SIGTERM)
select {
case <-serviceDone:
log.WithField("service_state", "done").Info("service is done")
case s := <-sigs:
log.WithField("signal", s).Info("signal received")
if err := killService(target, int32(s.(syscall.Signal))); err != nil {
log.WithField("error", err).Info("failed to kill service")
}
}
}
func main() {
d := deferer.NewDeferer(nil)
defer d.Run()
params := params{}
arg, err := base64.StdEncoding.DecodeString(os.Args[1])
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "base64.DecodeString",
"arg": os.Args[1],
}, "error decoding arg string")
}
if err := json.Unmarshal(arg, ¶ms); err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "json.Unmarshal",
"json": arg,
}, "error unmarshaling json")
}
l := params.Lock
if err := l.Refresh(); err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "lock.Refresh",
}, "failed to refresh lock")
}
d.Defer(func() {
if err := l.Release(); err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "lock.Release",
}, "failed to release lock")
}
})
locker := refresh(l, params.Interval)
sdttl, err := sd.WatchdogEnabled()
if err != nil {
d.FatalWithFields(log.Fields{
"error": err,
"func": "sd.WatchdogEnabled",
}, "failed to check watchdog configuration")
}
if uint64(sdttl.Seconds()) != params.TTL {
d.FatalWithFields(log.Fields{
"serviceTTL": sdttl,
"paramTTL": params.TTL,
}, "params and systemd ttls do not match")
}
tickler := tickle(params.Interval)
serviceDone := make(chan struct{})
base := filepath.Base(params.Args[0])
target := fmt.Sprintf("locked-%s-%d.service", base, params.ID)
go runService(d, serviceDone, params.ID, target, params.Args)
sigs := make(chan os.Signal)
signal.Notify(sigs, os.Interrupt, syscall.SIGTERM)
var killErr error
select {
case <-locker:
// TODO: should we never expect this?
killErr = killService(target, int32(syscall.SIGINT))
case <-serviceDone:
close(locker)
case s := <-sigs:
log.WithField("signal", s).Info("signal received")
killErr = killService(target, int32(s.(syscall.Signal)))
close(locker)
case <-tickler:
// watchdog tickler stopped, uh oh we are going down pretty soon
killErr = killService(target, int32(syscall.SIGINT))
close(locker)
}
if killErr != nil {
log.WithField("error", killErr).Fatal("failed to kill service")
}
}