func (c *Controller) kickOffHealthChecks(healthExit chan struct{}) {
client, err := node.NewLBClient(c.options.ServicedEndpoint)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", c.options.ServicedEndpoint, err)
return
}
defer client.Close()
var healthChecks map[string]domain.HealthCheck
instanceID, err := strconv.Atoi(c.options.Service.InstanceID)
if err != nil {
glog.Errorf("Invalid instance from instanceID:%s", c.options.Service.InstanceID)
return
}
err = client.GetHealthCheck(node.HealthCheckRequest{
c.options.Service.ID, instanceID}, &healthChecks)
if err != nil {
glog.Errorf("Error getting health checks: %s", err)
return
}
for key, mapping := range healthChecks {
glog.Infof("Kicking off health check %s.", key)
glog.Infof("Setting up health check: %s", mapping.Script)
timeout := mapping.Timeout
if timeout == 0 {
timeout = time.Second * 30
}
go c.handleHealthCheck(key, mapping.Script, mapping.Interval, timeout, healthExit)
}
return
}
// sendLogMessage sends a log message to the host agent
func sendLogMessage(lbClientPort string, serviceLogInfo node.ServiceLogInfo) error {
client, err := node.NewLBClient(lbClientPort)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", lbClientPort, err)
return err
}
defer client.Close()
return client.SendLogMessage(serviceLogInfo, nil)
}
// getAgentHostID retrieves the agent's host id
func getAgentHostID(lbClientPort string) (string, error) {
client, err := node.NewLBClient(lbClientPort)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", lbClientPort, err)
return "", err
}
defer client.Close()
var hostID string
err = client.GetHostID(&hostID)
if err != nil {
glog.Errorf("Error getting host id, error: %s", err)
return "", err
}
glog.V(1).Infof("getAgentHostID: %s", hostID)
return hostID, nil
}
// getServiceTenantID retrieves a service's tenantID
func getServiceTenantID(lbClientPort string, serviceID string) (string, error) {
client, err := node.NewLBClient(lbClientPort)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", lbClientPort, err)
return "", err
}
defer client.Close()
var tenantID string
err = client.GetTenantId(serviceID, &tenantID)
if err != nil {
glog.Errorf("Error getting service %s's tenantID, error: %s", serviceID, err)
return "", err
}
glog.V(1).Infof("getServiceTenantID: service id=%s: %s", serviceID, tenantID)
return tenantID, nil
}
// getAgentZkInfo retrieves the agent's zookeeper dsn
func getAgentZkInfo(lbClientPort string) (node.ZkInfo, error) {
var zkInfo node.ZkInfo
client, err := node.NewLBClient(lbClientPort)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", lbClientPort, err)
return zkInfo, err
}
defer client.Close()
err = client.GetZkInfo(&zkInfo)
if err != nil {
glog.Errorf("Error getting zookeeper dsn/poolID, error: %s", err)
return zkInfo, err
}
glog.V(1).Infof("GetZkInfo: %+v", zkInfo)
return zkInfo, nil
}
// rpcHealthCheck returns a channel that will close when it not longer possible
// to ping the RPC server
func (c *Controller) rpcHealthCheck() (chan struct{}, error) {
gone := make(chan struct{})
client, err := node.NewLBClient(c.options.ServicedEndpoint)
if err != nil {
return nil, err
}
go func() {
var ts time.Time
for {
err := client.Ping(time.Minute, &ts)
if err != nil {
close(gone)
return
}
}
}()
return gone, nil
}
// getService retrieves a service
func getService(lbClientPort string, serviceID string, instanceID int) (*service.Service, error) {
client, err := node.NewLBClient(lbClientPort)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", lbClientPort, err)
return nil, err
}
defer client.Close()
var svc service.Service
err = client.GetServiceInstance(node.ServiceInstanceRequest{serviceID, instanceID}, &svc)
if err != nil {
glog.Errorf("Error getting service %s error: %s", serviceID, err)
return nil, err
}
glog.V(1).Infof("getService: service id=%s: %+v", serviceID, svc)
return &svc, nil
}
// getServiceBindMounts retrieves a service's bindmounts
func getServiceBindMounts(lbClientPort string, serviceID string) (map[string]string, error) {
client, err := node.NewLBClient(lbClientPort)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", lbClientPort, err)
return nil, err
}
defer client.Close()
var bindmounts map[string]string
err = client.GetServiceBindMounts(serviceID, &bindmounts)
if err != nil {
if strings.HasPrefix(err.Error(), "rpc: can't find service") {
glog.Errorf("`serviced service shell` is available only when running serviced in agent mode")
return nil, err
}
glog.Errorf("Error getting service %s's bindmounts, error: %s", serviceID, err)
return nil, err
}
glog.V(1).Infof("getServiceBindMounts: service id=%s: %s", serviceID, bindmounts)
return bindmounts, nil
}
func registerExportedEndpoints(c *Controller, closing chan struct{}) <-chan struct{} {
for {
err := c.registerExportedEndpoints()
if err == nil {
return c.watchregistry()
}
client, err2 := node.NewLBClient(c.options.ServicedEndpoint)
if err2 != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", c.options.ServicedEndpoint, err2)
} else {
client.SendLogMessage(node.ServiceLogInfo{ServiceID: c.options.Service.ID, Message: fmt.Sprintf("error registering exported endpoints: %s", err)}, nil)
client.Close()
}
select {
case <-time.After(time.Second):
case <-closing:
return nil
}
glog.Errorf("could not register exported expoints: %s", err)
}
}
func (c *Controller) handleControlCenterImports(rpcdead chan struct{}) error {
// this function is currently needed to handle special control center imports
// from GetServiceEndpoints() that does not exist in endpoints from getServiceState
// get service endpoints
client, err := node.NewLBClient(c.options.ServicedEndpoint)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", c.options.ServicedEndpoint, err)
return err
}
defer client.Close()
// TODO: instead of getting all endpoints, via GetServiceEndpoints(), create a new call
// that returns only special "controlplane" imported endpoints
// Note: GetServiceEndpoints has been modified to return only special controlplane endpoints.
// We should rename it and clean up the filtering code below.
epchan := make(chan map[string][]dao.ApplicationEndpoint)
timeout := make(chan struct{})
go func(c *node.LBClient, svcid string, epc chan map[string][]dao.ApplicationEndpoint, timeout chan struct{}) {
var endpoints map[string][]dao.ApplicationEndpoint
RetryGetServiceEndpoints:
for {
err = c.GetServiceEndpoints(svcid, &endpoints)
if err != nil {
select {
case <-time.After(1 * time.Second):
glog.V(3).Info("Couldn't retrieve service endpoints, trying again")
continue RetryGetServiceEndpoints
case <-timeout:
glog.V(3).Info("Timed out trying to retrieve service endpoints")
return
}
}
break
}
// deal with the race between the one minute timeout in handleControlCenterImports() and the
// call to GetServiceEndpoint() - the timeout may happen between GetServiceEndpoint() completing
// and sending the result via the epc channel.
select {
case _, ok := <-epc:
if ok {
panic("should never receive anything on the endpoints channel")
}
glog.V(3).Info("Endpoint channel closed, giving up")
return
default:
epc <- endpoints
}
}(client, c.options.Service.ID, epchan, timeout)
var endpoints map[string][]dao.ApplicationEndpoint
select {
case <-time.After(1 * time.Minute):
close(epchan)
timeout <- struct{}{}
client.SendLogMessage(node.ServiceLogInfo{ServiceID: c.options.Service.ID, Message: "unable to retrieve service endpoints"}, nil)
return ErrNoServiceEndpoints
case <-rpcdead:
close(epchan)
timeout <- struct{}{}
return fmt.Errorf("RPC Service has gone away")
case endpoints = <-epchan:
glog.Infof("Got service endpoints for %s: %+v", c.options.Service.ID, endpoints)
}
// convert keys set by GetServiceEndpoints to tenantID_endpointID
tmp := make(map[string][]dao.ApplicationEndpoint)
for key, endpointList := range endpoints {
if len(endpointList) <= 0 {
glog.Warningf("ignoring key: %s with empty endpointList", key)
continue
}
tenantEndpointID := registry.TenantEndpointKey(c.tenantID, endpointList[0].Application)
glog.Infof("changing key from %s to %s: %+v", key, tenantEndpointID, endpointList[0])
tmp[tenantEndpointID] = endpoints[key]
}
endpoints = tmp
cc_endpoint_purpose := "import" // Punting on control center dynamic imports for now
for key, endpointList := range endpoints {
// ignore endpoints that are not special controlplane imports
ignorePrefix := fmt.Sprintf("%s_controlplane", c.tenantID)
if !strings.HasPrefix(key, ignorePrefix) {
continue
}
// set proxy addresses
c.setProxyAddresses(key, endpointList, endpointList[0].VirtualAddress, cc_endpoint_purpose)
// add/replace entries in importedEndpoints
instanceIDStr := fmt.Sprintf("%d", endpointList[0].InstanceID)
setImportedEndpoint(&c.importedEndpoints, c.tenantID,
endpointList[0].Application, instanceIDStr,
endpointList[0].VirtualAddress, cc_endpoint_purpose,
endpointList[0].ContainerPort)
// TODO: agent needs to register controlplane and controlplane_consumer
// but don't do that here in the container code
}
return nil
}
func (c *Controller) handleHealthCheck(name string, script string, interval, timeout time.Duration, exitChannel chan struct{}) {
client, err := node.NewLBClient(c.options.ServicedEndpoint)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", c.options.ServicedEndpoint, err)
return
}
defer client.Close()
scriptFile, err := ioutil.TempFile("", name)
if err != nil {
glog.Errorf("Error creating temporary file for health check %s: %s", name, err)
return
}
defer scriptFile.Close()
defer os.Remove(scriptFile.Name())
err = ioutil.WriteFile(scriptFile.Name(), []byte(script), os.FileMode(0777))
if err != nil {
glog.Errorf("Error writing script for health check %s: %s", name, err)
return
}
scriptFile.Close()
err = os.Chmod(scriptFile.Name(), os.FileMode(0777))
if err != nil {
glog.Errorf("Error setting script executable for health check %s: %s", name, err)
return
}
var unused int
sigtermTimeout := time.Second * 10
for {
select {
case <-time.After(interval):
exited := make(chan error, 1)
sysProcAttr := &syscall.SysProcAttr{Setpgid: true, Pdeathsig: syscall.SIGTERM}
cmd := exec.Command("sh", "-c", scriptFile.Name())
cmd.SysProcAttr = sysProcAttr
if err := cmd.Start(); err != nil {
glog.Errorf("Could not run cmd %v: %s", cmd, err)
break
}
go func(c *exec.Cmd) { exited <- c.Wait() }(cmd)
select {
case err := <-exited:
if err == nil {
glog.V(4).Infof("Health check %s succeeded.", name)
client.LogHealthCheck(domain.HealthCheckResult{c.options.Service.ID, c.options.Service.InstanceID, name, time.Now().String(), "passed"}, &unused)
} else {
glog.Warningf("Health check %s failed.", name)
client.LogHealthCheck(domain.HealthCheckResult{c.options.Service.ID, c.options.Service.InstanceID, name, time.Now().String(), "failed"}, &unused)
}
case <-exitChannel:
proc.KillGroup(cmd.Process.Pid, sigtermTimeout)
return
case <-time.After(timeout):
proc.KillGroup(cmd.Process.Pid, sigtermTimeout)
glog.Warningf("Health check %s timeout.", name)
client.LogHealthCheck(domain.HealthCheckResult{c.options.Service.ID, c.options.Service.InstanceID, name, time.Now().String(), "failed"}, &unused)
}
case <-exitChannel:
return
}
}
}
// Run executes the controller's main loop and block until the service exits
// according to it's restart policy or Close() is called.
func (c *Controller) Run() (err error) {
defer c.shutdown()
sigc := make(chan os.Signal, 1)
signal.Notify(sigc,
syscall.SIGINT,
syscall.SIGTERM,
syscall.SIGQUIT)
env := os.Environ()
env = append(env, "CONTROLPLANE=1")
env = append(env, fmt.Sprintf("CONTROLPLANE_CONSUMER_URL=http://localhost%s/api/metrics/store", c.options.Metric.Address))
env = append(env, fmt.Sprintf("CONTROLPLANE_HOST_ID=%s", c.hostID))
env = append(env, fmt.Sprintf("CONTROLPLANE_TENANT_ID=%s", c.tenantID))
env = append(env, fmt.Sprintf("CONTROLPLANE_INSTANCE_ID=%s", c.options.Service.InstanceID))
env = append(env, fmt.Sprintf("CONTROLPLANE_SERVICED_ID=%s", c.options.Service.ID))
if err := writeEnvFile(env); err != nil {
return err
}
args := []string{"-c", "exec " + strings.Join(c.options.Service.Command, " ")}
startService := func() (*subprocess.Instance, chan error) {
service, serviceExited, _ := subprocess.New(time.Second*10, env, "/bin/sh", args...)
return service, serviceExited
}
sendSignal := func(service *subprocess.Instance, sig os.Signal) bool {
switch {
case c.PIDFile != "":
c.forwardSignal(sig)
case service != nil:
service.Notify(sig)
default:
return false
}
return true
}
rpcDead, err := c.rpcHealthCheck()
if err != nil {
glog.Error("Could not setup RPC ping check: %s", err)
return err
}
storageDead, err := c.storageHealthCheck()
if err != nil {
glog.Errorf("Could not set up storage check: %s", err)
return err
}
prereqsPassed := make(chan bool)
var startAfter <-chan time.Time
var exitAfter <-chan time.Time
var service *subprocess.Instance = nil
serviceExited := make(chan error, 1)
c.watchRemotePorts()
if err := c.handleControlCenterImports(rpcDead); err != nil {
glog.Error("Could not setup Control Center specific imports: ", err)
return err
}
go c.checkPrereqs(prereqsPassed, rpcDead)
go c.reapZombies(rpcDead)
healthExit := make(chan struct{})
defer close(healthExit)
c.kickOffHealthChecks(healthExit)
doRegisterEndpoints := true
exited := false
var shutdownService = func(service *subprocess.Instance, sig os.Signal) {
c.options.Service.Autorestart = false
if sendSignal(service, sig) {
sigc = nil
prereqsPassed = nil
startAfter = nil
rpcDead = nil
exitAfter = time.After(time.Second * 30)
close(healthExit)
} else {
c.exitStatus = 1
exited = true
}
}
var reregister <-chan struct{}
for !exited {
select {
case sig := <-sigc:
glog.Infof("Notifying subprocess of signal %v", sig)
shutdownService(service, sig)
case <-exitAfter:
glog.Infof("Killing unresponsive subprocess")
sendSignal(service, syscall.SIGKILL)
c.exitStatus = 1
exited = true
case <-prereqsPassed:
startAfter = time.After(time.Millisecond * 1)
prereqsPassed = nil
case exitError := <-serviceExited:
if !c.options.Service.Autorestart {
exitStatus, _ := utils.GetExitStatus(exitError)
if c.options.Logforwarder.Enabled {
time.Sleep(c.options.Logforwarder.SettleTime)
}
glog.Infof("Service Exited with status:%d due to %+v", exitStatus, exitError)
//set loop to end
exited = true
//exit with exit code, defer so that other cleanup can happen
c.exitStatus = exitStatus
} else {
glog.Infof("Restarting service process in 10 seconds.")
service = nil
startAfter = time.After(time.Second * 10)
}
case <-startAfter:
glog.Infof("Starting service process.")
service, serviceExited = startService()
if doRegisterEndpoints {
reregister = registerExportedEndpoints(c, rpcDead)
doRegisterEndpoints = false
}
startAfter = nil
case <-reregister:
reregister = registerExportedEndpoints(c, rpcDead)
case <-rpcDead:
glog.Infof("RPC Server has gone away, cleaning up")
shutdownService(service, syscall.SIGTERM)
case <-storageDead:
glog.Infof("Distributed storage for service %s has gone away; shutting down", c.options.Service.ID)
shutdownService(service, syscall.SIGTERM)
}
}
// Signal to health check registry that this instance is giving up the ghost.
client, err := node.NewLBClient(c.options.ServicedEndpoint)
if err != nil {
glog.Errorf("Could not create a client to endpoint: %s, %s", c.options.ServicedEndpoint, err)
return nil
}
defer client.Close()
c.Close()
var unused int
client.LogHealthCheck(domain.HealthCheckResult{c.options.Service.ID, c.options.Service.InstanceID, "__instance_shutdown", time.Now().String(), "passed"}, &unused)
return nil
}
