improvements to (VM's) shutdown logic

- when shutting down corectld or halting multiple VMs those will
  always be halted in reverse boot order
- when hit by a signal corectld will now properly halt running VMs

Signed-off-by: António Meireles <[email protected]>

components/server/rpcservices.go

@@ -56,7 +56,7 @@ type (
 		Meta       *release.Info
 		VM         *VMInfo
 		Images     map[string]semver.Versions
-		Running    map[string]*VMInfo
+		Running    VMmap
 		WorkingNFS bool
 	}
 )
@@ -272,6 +272,11 @@ func (s *RPCservice) Run(r *http.Request,
 		Daemon.Jobs.Add(1)
 		defer Daemon.Jobs.Done()
 		Daemon.Lock()
+		vm.exec.SysProcAttr = &syscall.SysProcAttr{
+			Setpgid: true,
+			Setsid:  false,
+			Pgid:    0,
+		}
 		err := vm.exec.Start()
 		Daemon.Unlock()
 		if err != nil {
@@ -307,6 +312,11 @@ func (s *RPCservice) Stop(r *http.Request,
 	}
 
 	log.Info("Sky must be falling. Shutting down...")
+	Daemon.Lock()
+	Daemon.AcceptingRequests = false
+	Daemon.Unlock()
+
+	Daemon.Active.array().halt()
 	Daemon.Oops <- nil
 	return
 }
@@ -327,37 +337,24 @@ func (s *RPCservice) StopVMs(r *http.Request,
 	args *RPCquery, reply *RPCreply) (err error) {
 	log.Debug("vm:stop")
 
-	var toHalt []string
+	var targets VMs
 
 	if !Daemon.AcceptingRequests {
 		return ErrServerShuttingDown
 	}
 
 	if len(args.Input) == 0 {
-		for _, x := range Daemon.Active {
-			toHalt = append(toHalt, x.UUID)
-		}
+		targets = Daemon.Active.array()
 	} else {
 		for _, t := range args.Input {
 			for _, v := range Daemon.Active {
 				if v.Name == t || v.UUID == t {
-					toHalt = append(toHalt, v.UUID)
+					targets = append(targets, v)
 				}
 			}
 		}
 	}
-	for _, v := range toHalt {
-		Daemon.Active[v].halt()
-		for {
-			Daemon.Lock()
-			_, stillAlive := Daemon.Active[v]
-			Daemon.Unlock()
-			if !stillAlive {
-				break
-			}
-			time.Sleep(100 * time.Millisecond)
-		}
-	}
+	targets.halt()
 	return
 }
 

components/server/run.go

@@ -22,6 +22,7 @@ import (
 	"encoding/pem"
 	"fmt"
 	"os/exec"
+	"sort"
 	"syscall"
 
 	"net/http"
@@ -59,6 +60,10 @@ type (
 		isolationCheck, callBack sync.Once
 		cloudConfigContents      []byte
 	}
+	//
+	VMmap map[string]*VMInfo
+	// Config ...
+	VMs []*VMInfo
 	// NetworkInterface ...
 	NetworkInterface struct {
 		Type int
@@ -333,6 +338,25 @@ func (vm *VMInfo) assembleBootPayload() (xArgs []string, err error) {
 		err
 }
 
+func (list VMs) halt() {
+	sort.Sort(sort.Reverse(VMs(list)))
+
+	for _, v := range list {
+		log.Info("shutting down %v...", v.Name)
+		Daemon.Active[v.UUID].halt()
+		for {
+			Daemon.Lock()
+			_, stillAlive := Daemon.Active[v.UUID]
+			Daemon.Unlock()
+			if !stillAlive {
+				break
+			}
+			time.Sleep(100 * time.Millisecond)
+		}
+	}
+	return
+}
+
 func (vm *VMInfo) halt() {
 	// Try to gracefully terminate the process.
 	if err := vm.exec.Process.Signal(syscall.SIGTERM); err != nil {
@@ -455,3 +479,20 @@ func (volumes *StorageAssets) PrettyPrint(root int) {
 		}
 	}
 }
+
+func (run VMs) Len() int {
+	return len(run)
+}
+func (run VMs) Swap(i, j int) {
+	run[i], run[j] = run[j], run[i]
+}
+func (run VMs) Less(i, j int) bool {
+	return run[i].CreationTime.Before(run[j].CreationTime)
+}
+
+func (in VMmap) array() (out VMs) {
+	for _, r := range in {
+		out = append(out, r)
+	}
+	return
+}

components/server/server.go

@@ -41,7 +41,7 @@ type (
 	ServerContext struct {
 		Meta              *release.Info
 		Media             MediaAssets
-		Active            map[string]*VMInfo
+		Active            VMmap
 		APIserver         *manners.GracefulServer
 		EtcdServer        *EtcdServer
 		EtcdClient        client.KeysAPI
@@ -64,7 +64,7 @@ func New() (cfg *ServerContext) {
 		AcceptingRequests: true,
 		WorkingNFS:        false,
 		Oops:              make(chan error, 1),
-		Active:            make(map[string]*VMInfo),
+		Active:            make(VMmap),
 	}
 }
 
@@ -124,6 +124,7 @@ func Start() (err error) {
 		log.Info("Got '%v' signal, stopping server...", s)
 		signal.Stop(hades)
 		Daemon.Oops <- nil
+		Daemon.Active.array().halt()
 	}()
 
 	log.Info("server starting...")
@@ -154,11 +155,6 @@ func Start() (err error) {
 		}
 	}
 
-	Daemon.Lock()
-	for _, r := range Daemon.Active {
-		r.halt()
-	}
-	Daemon.Unlock()
 	Daemon.Jobs.Wait()
 	Daemon.APIserver.Close()
 	log.Info("gone!")