// restoreContainers loads all running containers and inserts
// them into skydns when skydock starts
func restoreContainers() error {
containers, err := dockerClient.FetchAllContainers()
if err != nil {
return err
}
var container *docker.Container
for _, cnt := range containers {
uuid := utils.Truncate(cnt.Id)
if container, err = dockerClient.FetchContainer(uuid, cnt.Image); err != nil {
if err != docker.ErrImageNotTagged {
log.Logf(log.ERROR, "failed to fetch %s on restore: %s", cnt.Id, err)
}
continue
}
service, err := plugins.createService(container)
if err != nil {
// doing a fatal here because we cannot do much if the plugins
// return an invalid service or error
fatal(err)
}
if err := sendService(uuid, service); err != nil {
log.Logf(log.ERROR, "failed to send %s to skydns on restore: %s", uuid, err)
}
}
return nil
}
// sendService sends the uuid and service data to skydns
func sendService(uuid string, service *msg.Service) error {
log.Logf(log.INFO, "adding %s (%s) to skydns", uuid, service.Name)
if err := skydns.Add(uuid, service); err != nil {
// ignore erros for conflicting uuids and start the heartbeat again
if err != client.ErrConflictingUUID {
return err
}
log.Logf(log.INFO, "service already exists for %s. Resetting ttl.", uuid)
updateService(uuid, ttl)
}
go heartbeat(uuid)
return nil
}
func main() {
validateSettings()
if err := setupLogger(); err != nil {
fatal(err)
}
var (
err error
group = &sync.WaitGroup{}
)
plugins, err = newRuntime(pluginFile)
if err != nil {
fatal(err)
}
if dockerClient, err = docker.NewClient(pathToSocket); err != nil {
log.Logf(log.FATAL, "error connecting to docker: %s", err)
fatal(err)
}
if skydnsContainerName != "" {
container, err := dockerClient.FetchContainer(skydnsContainerName, "")
if err != nil {
log.Logf(log.FATAL, "error retrieving skydns container '%s': %s", skydnsContainerName, err)
fatal(err)
}
skydnsUrl = "http://" + container.NetworkSettings.IpAddress + ":8080"
}
log.Logf(log.INFO, "skydns URL: %s", skydnsUrl)
if skydns, err = client.NewClient(skydnsUrl, secret, domain, "172.17.42.1:53"); err != nil {
log.Logf(log.FATAL, "error connecting to skydns: %s", err)
fatal(err)
}
log.Logf(log.DEBUG, "starting restore of containers")
if err := restoreContainers(); err != nil {
log.Logf(log.FATAL, "error restoring containers: %s", err)
fatal(err)
}
events := dockerClient.GetEvents()
group.Add(numberOfHandlers)
// Start event handlers
for i := 0; i < numberOfHandlers; i++ {
go eventHandler(events, group)
}
log.Logf(log.DEBUG, "starting main process")
group.Wait()
log.Logf(log.DEBUG, "stopping cleanly via EOF")
}
func eventHandler(c chan *docker.Event, group *sync.WaitGroup) {
defer group.Done()
for event := range c {
log.Logf(log.DEBUG, "received event (%s) %s %s", event.Status, event.ContainerId, event.Image)
uuid := utils.Truncate(event.ContainerId)
switch event.Status {
case "die", "stop", "kill":
if err := removeService(uuid); err != nil {
log.Logf(log.ERROR, "error removing %s from skydns: %s", uuid, err)
}
case "start", "restart":
if err := addService(uuid, event.Image); err != nil {
log.Logf(log.ERROR, "error adding %s to skydns: %s", uuid, err)
}
}
}
}
func heartbeat(uuid string) {
runningLock.Lock()
if _, exists := running[uuid]; exists {
runningLock.Unlock()
return
}
running[uuid] = struct{}{}
runningLock.Unlock()
defer func() {
runningLock.Lock()
delete(running, uuid)
runningLock.Unlock()
}()
var errorCount int
for _ = range time.Tick(time.Duration(beat) * time.Second) {
if errorCount > 10 {
// if we encountered more than sequential 10 errors just quit
log.Logf(log.ERROR, "aborting heartbeat for %s after sequential 10 errors", uuid)
return
}
// don't fill logs if we have a low beat
// may need to do something better here
if beat >= 30 {
log.Logf(log.INFO, "updating ttl for %s", uuid)
}
if err := updateService(uuid, ttl); err != nil {
errorCount++
log.Logf(log.ERROR, "updateService(): %s", err)
} else {
if errorCount--; errorCount < 0 {
errorCount = 0
}
}
}
}
func (d *dockerClient) GetEvents() chan *Event {
eventChan := make(chan *Event, 100) // 100 event buffer
go func() {
defer close(eventChan)
c, err := d.newConn()
if err != nil {
log.Logf(log.FATAL, "cannot connect to docker: %s", err)
return
}
defer c.Close()
req, err := http.NewRequest("GET", "/events", nil)
if err != nil {
log.Logf(log.ERROR, "bad request for events: %s", err)
return
}
resp, err := c.Do(req)
if err != nil {
log.Logf(log.FATAL, "cannot connect to events endpoint: %s", err)
return
}
defer resp.Body.Close()
// handle signals to stop the socket
sigChan := make(chan os.Signal, 1)
signal.Notify(sigChan, os.Interrupt, syscall.SIGTERM, syscall.SIGQUIT)
go func() {
for sig := range sigChan {
log.Logf(log.INFO, "received signal '%v', exiting", sig)
c.Close()
close(eventChan)
os.Exit(0)
}
}()
dec := json.NewDecoder(resp.Body)
for {
var event *Event
if err := dec.Decode(&event); err != nil {
if err == io.EOF {
break
}
log.Logf(log.ERROR, "cannot decode json: %s", err)
continue
}
eventChan <- event
}
log.Logf(log.DEBUG, "closing event channel")
}()
return eventChan
}
func newRuntime(file string) (*pluginRuntime, error) {
runtime := otto.New()
log.Logf(log.INFO, "loading plugins from %s", file)
content, err := ioutil.ReadFile(file)
if err != nil {
return nil, err
}
if _, err := runtime.Run(string(content)); err != nil {
return nil, err
}
if err := loadDefaults(runtime); err != nil {
return nil, err
}
return &pluginRuntime{runtime}, nil
}
func removeService(uuid string) error {
log.Logf(log.INFO, "removing %s from skydns", uuid)
return skydns.Delete(uuid)
}