func (s *storage) Set(key, value string) error {
s.Log.WithTags(spec.Tags{C: nil, L: "D", O: s, V: 13}, "call Set")
action := func() error {
conn := s.Pool.Get()
defer conn.Close()
reply, err := redis.String(conn.Do("SET", s.withPrefix(key), value))
if err != nil {
return maskAny(err)
}
if reply != "OK" {
return maskAnyf(queryExecutionFailedError, "SET not executed correctly")
}
return nil
}
err := backoff.Retry(s.Instrumentation.WrapFunc("Set", action), s.BackoffFactory())
if err != nil {
return maskAny(err)
}
return nil
}
// listenForNodeChanges listens for changes to node status using change feeds.
// This function will block until the query fails
func (c *Cluster) listenForNodeChanges() error {
// Start listening to changes from a random active node
node, hpr, err := c.GetNextNode()
if err != nil {
return err
}
q, err := newQuery(
DB("rethinkdb").Table("server_status").Changes(),
map[string]interface{}{},
c.opts,
)
if err != nil {
return fmt.Errorf("Error building query: %s", err)
}
cursor, err := node.Query(q)
if err != nil {
hpr.Mark(err)
return err
}
// Keep reading node status updates from changefeed
var result struct {
NewVal nodeStatus `gorethink:"new_val"`
OldVal nodeStatus `gorethink:"old_val"`
}
for cursor.Next(&result) {
addr := fmt.Sprintf("%s:%d", result.NewVal.Network.Hostname, result.NewVal.Network.ReqlPort)
addr = strings.ToLower(addr)
switch result.NewVal.Status {
case "connected":
// Connect to node using exponential backoff (give up after waiting 5s)
// to give the node time to start-up.
b := backoff.NewExponentialBackOff()
b.MaxElapsedTime = time.Second * 5
backoff.Retry(func() error {
node, err := c.connectNodeWithStatus(result.NewVal)
if err == nil {
if !c.nodeExists(node) {
c.addNode(node)
Log.WithFields(logrus.Fields{
"id": node.ID,
"host": node.Host.String(),
}).Debug("Connected to node")
}
}
return err
}, b)
}
}
err = cursor.Err()
hpr.Mark(err)
return err
}
func (c *Client) executeRequest(req *http.Request) (*Response, error) {
var parsedError error
var parsedResponse *Response
// Setup our retryable operation.
retry := func() error {
// Execute the actual request.
response, err := c.goClient.Do(req)
if err != nil {
parsedError = err
return nil
}
// Defer closing response body if it exists.
if response != nil {
defer response.Body.Close()
}
// Parse our response into a gohttp.Response object.
parsedResponse, parsedError = NewResponse(response)
if parsedError != nil {
return nil
}
// Retry status code checks.
//
// If the client encounters a retryable status code, we return
// an error to signal a retry should occur
for _, code := range c.RetryableStatusCodes {
if code == parsedResponse.Code {
return errors.New("Encountered retryable status code.")
}
}
return nil
}
// Rate Limiter - If we are using a rate limiter, we enter here.
//
// This will block until the rate limiter is satisfied.
if c.rateLimiter != nil {
err := c.rateLimiter.Enter()
if err != nil {
return nil, err
}
}
// Execute the retryable operation
err := backoff.Retry(retry, c.Backoff)
if err != nil {
return nil, err
}
return parsedResponse, parsedError
}
func connectSlackRTM(h *helios.Engine, s *SlackService) {
err := backoff.Retry(func() error {
h.Debug("Connecting to slack rtm")
err := runSlackRTM(h, s)
if err != nil {
h.Warn("Failed to start slack rtm. Retrying.", "error", err.Error())
}
return err
}, backoff.NewExponentialBackOff())
if err != nil {
s.Messages <- helios.NewError("Slack service error: %v", err)
}
}
// TryRequest try operation timeout, and retry backoff
func TryRequest(url string, timeout time.Duration, condition Condition) error {
exponentialBackOff := backoff.NewExponentialBackOff()
exponentialBackOff.MaxElapsedTime = timeout
var res *http.Response
err := backoff.Retry(func() error {
var err error
res, err = http.Get(url)
if err != nil {
return err
}
return condition(res)
}, exponentialBackOff)
return err
}
func SupervisorCommand() cli.Command {
return cli.Command{
Name: "supervisor",
SkipFlagParsing: true,
Action: func(c *cli.Context) error {
var cmd *exec.Cmd = nil
go func() {
backoff.Retry(func() error {
log.Println(color.GreenString("supervisor: Starting %v", c.Args()))
cmd = exec.Command(c.Args().First(), c.Args().Tail()...)
cmd.Stdout = os.Stdout
cmd.Stderr = os.Stderr
if err := cmd.Start(); err != nil {
log.Println(color.RedString("supervisor: Error on process staring %v", err))
return err
}
err := cmd.Wait()
if err != nil {
log.Println(color.RedString("supervisor: Command exit with error: %v", err))
} else {
log.Println(color.YellowString("supervisor: Command completed."))
os.Exit(0)
}
return err
}, backoff.NewConstantBackOff(time.Second*3))
}()
// Handle shutdown signals and kill child process
ch := make(chan os.Signal)
signal.Notify(ch, syscall.SIGINT, syscall.SIGTERM, syscall.SIGKILL)
log.Println("supervisor: receive signal", <-ch)
if cmd != nil {
log.Println("supervisor: send SIGTERM to subprocess")
cmd.Process.Signal(syscall.SIGTERM)
res, err := cmd.Process.Wait()
log.Println("supervisor: subprocess stopped with", res, err)
}
log.Println(color.YellowString("supervisor: Stopped."))
os.Exit(143)
return nil
},
}
}
func (s *storage) Remove(key string) error {
s.Log.WithTags(spec.Tags{C: nil, L: "D", O: s, V: 13}, "call Remove")
action := func() error {
conn := s.Pool.Get()
defer conn.Close()
_, err := redis.Int64(conn.Do("DEL", s.withPrefix(key)))
if err != nil {
return maskAny(err)
}
return nil
}
err := backoff.Retry(s.Instrumentation.WrapFunc("Remove", action), s.BackoffFactory())
if err != nil {
return maskAny(err)
}
return nil
}
func (s *service) CreateMax(max int) (int, error) {
// Define the action.
var result int
action := func() error {
done := make(chan struct{}, 1)
fail := make(chan error, 1)
go func() {
m := big.NewInt(int64(max))
j, err := s.RandFactory(s.RandReader, m)
if err != nil {
fail <- maskAny(err)
return
}
result = int(j.Int64())
done <- struct{}{}
}()
select {
case <-time.After(s.Timeout):
return maskAnyf(timeoutError, "after %s", s.Timeout)
case err := <-fail:
return maskAny(err)
case <-done:
return nil
}
}
// Execute the action wrapped with a retrier.
err := backoff.Retry(action, s.BackoffFactory())
if err != nil {
return 0, maskAny(err)
}
return result, nil
}
// Try try operation timeout, and retry backoff
func Try(timeout time.Duration, operation func() error) error {
exponentialBackOff := backoff.NewExponentialBackOff()
exponentialBackOff.MaxElapsedTime = timeout
err := backoff.Retry(operation, exponentialBackOff)
return err
}
func (p DeployMarathon) Install(data manifest.Manifest) error {
marathonApi, err := MarathonClient(data.GetString("marathon-address"))
if err != nil {
return err
}
fullName := data.GetString("app-name")
bs, bf, bmax, grace := 5.0, 2.0, 120.0, 30.0
app := &marathon.Application{
BackoffSeconds: &bs,
BackoffFactor: &bf,
MaxLaunchDelaySeconds: &bmax,
TaskKillGracePeriodSeconds: &grace,
UpgradeStrategy: &marathon.UpgradeStrategy{
MinimumHealthCapacity: 0.5, // rolling update
MaximumOverCapacity: 0.0,
},
}
portArgs := ""
if port := data.GetString("listen-port"); port != "" {
portArgs = "--port " + port
}
app.Name(fullName)
app.Command(fmt.Sprintf("exec serve-tools consul supervisor --service '%s' %s start %s", fullName, portArgs, data.GetString("cmd")))
app.Count(data.GetInt("instances"))
app.Memory(float64(data.GetInt("mem")))
if cpu, err := strconv.ParseFloat(data.GetString("cpu"), 64); err == nil {
app.CPU(cpu)
}
if cluster := data.GetString("cluster"); cluster != "" {
cs := strings.SplitN(cluster, ":", 2)
app.AddConstraint(cs[0], "CLUSTER", cs[1])
app.AddLabel(cs[0], cs[1])
}
for _, cons := range data.GetArray("constraints") {
if consArr, ok := cons.Unwrap().([]interface{}); ok {
consStrings := make([]string, len(consArr))
for i, c := range consArr {
consStrings[i] = fmt.Sprintf("%s", c)
}
app.AddConstraint(consStrings...)
}
}
for _, port := range data.GetArray("ports") {
app.AddPortDefinition(marathon.PortDefinition{Name: port.GetStringOr("name", "")}.SetPort(port.GetIntOr("port", 0)))
}
app.AddEnv("SERVICE_DEPLOY_TIME", time.Now().Format(time.RFC3339)) // force redeploy app
for k, v := range data.GetMap("environment") {
app.AddEnv(k, fmt.Sprintf("%s", v.Unwrap()))
}
app.AddUris(data.GetString("package-uri"))
// todo: в манифесте задавать массив healthchecks, их использовтаь в марафоне и консул-супервизоре
// todo: открыть сетевой доступ от марафона до мезос-агентов, чтобы марафон мог хелсчеки посылать
//if portArgs != "" {
// health := marathon.NewDefaultHealthCheck()
// health.Protocol = "TCP"
// health.IntervalSeconds = 5
// *health.PortIndex = 0
// app.AddHealthCheck(*health)
//}
if _, err := marathonApi.UpdateApplication(app, false); err != nil {
color.Yellow("marathon <- %s", app)
return err
}
color.Green("marathon <- %s", app)
consulApi, err := utils.ConsulClient(data.GetString("consul-address"))
if err != nil {
return err
}
if err := utils.RegisterPluginData("deploy.marathon", data.GetString("app-name"), data.String(), data.GetString("consul-address")); err != nil {
return err
}
return backoff.Retry(func() error {
services, _, err := consulApi.Health().Service(fullName, "", true, nil)
if err != nil {
log.Println(color.RedString("Error in check health in consul: %v", err))
return err
}
if len(services) == 0 {
log.Printf("Service `%s` not started yet! Retry...", fullName)
return fmt.Errorf("Service `%s` not started!", fullName)
}
log.Println(color.GreenString("Service `%s` successfully started!", fullName))
return nil
}, backoff.NewExponentialBackOff())
}
func (p ReleaseHttp) Run(data manifest.Manifest) error {
if !data.Has("routes") {
log.Println("No routes configured for release.")
return nil
}
consul, err := utils.ConsulClient(data.GetString("consul-address"))
if err != nil {
return err
}
fullName := data.GetString("full-name")
// check current service is alive and healthy
if err := backoff.Retry(func() error {
services, _, err := consul.Health().Service(fullName, "", true, nil)
if err != nil {
log.Println(color.RedString("Error in check health in consul: %v", err))
return err
}
if len(services) == 0 {
log.Printf("Service `%s` not started yet! Retry...", fullName)
return fmt.Errorf("Service `%s` not started!", fullName)
} else {
log.Printf("Service `%s` started with %v instances.", fullName, len(services))
return nil
}
}, backoff.NewExponentialBackOff()); err != nil {
return err
}
routeVars := make(map[string]string, 0)
if data.Has("route") {
if err := json.Unmarshal([]byte(data.GetString("route")), &routeVars); err != nil {
log.Println(color.RedString("Error parse route json: %v, %s", err, data.GetString("route")))
return err
}
}
// collect routes
routes := make([]map[string]string, 0)
for _, route := range data.GetArray("routes") {
if !route.Has("host") {
log.Printf("Not found 'host': %s, skip...", route.String())
continue
}
fields := make(map[string]string)
for k, v := range route.Unwrap().(map[string]interface{}) {
fields[k] = fmt.Sprintf("%v", v)
}
routes = append(routes, utils.MergeMaps(fields, routeVars))
}
if len(routes) == 0 {
log.Println("No routes configured for release.")
return nil
}
routesJson, err := json.MarshalIndent(routes, "", " ")
if err != nil {
return err
}
// write routes to consul kv
if err := utils.PutConsulKv(consul, "services/routes/"+fullName, string(routesJson)); err != nil {
return err
}
log.Println(color.GreenString("Service `%s` released with routes: %s", fullName, string(routesJson)))
// find old services with the same routes
existsRoutes, err := utils.ListConsulKv(consul, "services/routes/"+data.GetString("name-prefix"), nil)
if err != nil {
return err
}
for _, existsRoute := range existsRoutes {
if existsRoute.Key != fmt.Sprintf("services/routes/%s", fullName) { // skip current service
oldRoutes := make([]map[string]string, 0)
if err := json.Unmarshal(existsRoute.Value, &oldRoutes); err != nil {
return err
}
OuterLoop:
for _, route := range routes {
for _, oldRoute := range oldRoutes {
if utils.MapsEqual(route, oldRoute) {
outdated := strings.TrimPrefix(existsRoute.Key, "services/routes/")
log.Println(color.GreenString("Found %s with the same routes %v. Remove it!", outdated, string(existsRoute.Value)))
if err := utils.DelConsulKv(consul, existsRoute.Key); err != nil {
return err
}
if err := utils.MarkAsOutdated(consul, outdated, 10*time.Minute); err != nil {
return err
}
break OuterLoop
}
}
}
}
}
return nil
}