brick-mux : fix spurious socket connect failure issue during volume stop

glusterd2/brick/glusterfsd.go

@@ -38,7 +38,7 @@ type Glusterfsd struct {
 	// Externally consumable using methods of Glusterfsd interface
 	binarypath     string
 	args           []string
-	socketfilepath string
+	Socketfilepath string
 	pidfilepath    string
 
 	// For internal use
@@ -89,8 +89,8 @@ func (b *Glusterfsd) Args() []string {
 // SocketFile returns path to the brick socket file used for IPC.
 func (b *Glusterfsd) SocketFile() string {
 
-	if b.socketfilepath != "" {
-		return b.socketfilepath
+	if b.Socketfilepath != "" {
+		return b.Socketfilepath
 	}
 
 	// First we form a key
@@ -101,10 +101,10 @@ func (b *Glusterfsd) SocketFile() string {
 	// Example: /var/run/gluster/<xxhash-hash>.socket
 	glusterdSockDir := config.GetString("rundir")
 	hash := strconv.FormatUint(xxhash.Sum64String(key), 16)
-	b.socketfilepath = path.Join(glusterdSockDir, hash+".socket")
+	b.Socketfilepath = path.Join(glusterdSockDir, hash+".socket")
 	// FIXME: The brick can no longer clean this up on clean shut down
 
-	return b.socketfilepath
+	return b.Socketfilepath
 }
 
 // PidFile returns path to the pid file of the brick process

glusterd2/brickmux/demultiplex.go

@@ -3,7 +3,6 @@ package brickmux
 import (
 	"fmt"
 	"os"
-	"time"
 
 	"github.com/gluster/glusterd2/glusterd2/brick"
 	"github.com/gluster/glusterd2/glusterd2/daemon"
@@ -27,16 +26,32 @@ func IsLastBrickInProc(b brick.Brickinfo) bool {
 // Demultiplex sends a detach request to the brick process which the
 // specified brick is multiplexed onto.
 func Demultiplex(b brick.Brickinfo) error {
-
+	var pidOnFile int
 	log.WithField("brick", b.String()).Debug("get brick daemon for demultiplex process")
 	brickDaemon, err := brick.NewGlusterfsd(b)
 	if err != nil {
 		return err
 	}
+	if pidOnFile, err = daemon.ReadPidFromFile(brickDaemon.PidFile()); err != nil {
+		log.WithFields(log.Fields{"brick": b.String(),
+			"pidfile": brickDaemon.PidFile()}).Error("Failed to read the pidfile")
+		return err
+
+	}
+	if pidOnFile == -1 {
+		log.WithFields(log.Fields{"brick": b.String(),
+			"pidfile": brickDaemon.PidFile()}).Error("Pid is -1")
+		return err
 
+	}
+	brickDaemon.Socketfilepath, err = glusterdGetSockFromBrickPid(pidOnFile)
+	if err != nil {
+		log.WithFields(log.Fields{"brick": b.String(),
+			"pid": pidOnFile}).Error("Failed to get the socket file of the glusterfsd process")
+		return err
+	}
 	log.WithFields(log.Fields{"brick": b.String(),
 		"socketFile": brickDaemon.SocketFile()}).Debug("starting demultiplex process")
-
 	client, err := daemon.GetRPCClient(brickDaemon)
 	if err != nil {
 		return err
@@ -58,16 +73,8 @@ func Demultiplex(b brick.Brickinfo) error {
 	}
 	log.WithField("brick", b.String()).Debug("detach request succeeded with result")
 
-	// TODO: Find an alternative to substitute the sleep.
-	// There might be some changes on glusterfsd side related to socket
-	// files used while brick signout,
-	// make appropriate changes once glusterfsd side is fixed.
-	time.Sleep(time.Millisecond * 200)
-
 	log.WithField("brick", b.String()).Debug("deleting brick socket and pid file")
 	os.Remove(brickDaemon.PidFile())
-	os.Remove(brickDaemon.SocketFile())
-	log.WithField("brick", b.String()).Debug("deleted brick socket and pid file")
 
 	return nil
 }

glusterd2/brickmux/multiplex.go

@@ -3,7 +3,6 @@ package brickmux
 import (
 	"fmt"
 	"net/rpc"
-	"os"
 
 	"github.com/gluster/glusterd2/glusterd2/brick"
 	"github.com/gluster/glusterd2/glusterd2/daemon"
@@ -87,13 +86,6 @@ func Multiplex(b brick.Brickinfo, v *volume.Volinfo, volumes []*volume.Volinfo,
 		return err
 	}
 
-	// create Unix Domain Socket hardlink
-	os.Remove(brickProc.SocketFile())
-	if err := os.Link(targetBrickProc.SocketFile(), brickProc.SocketFile()); err != nil {
-		undoMultiplex(client, &b)
-		return err
-	}
-
 	// create duplicate pidfile for the multiplexed brick
 	ok, pid := daemon.IsRunning(targetBrickProc)
 	if !ok {

glusterd2/brickmux/utils.go

@@ -1,10 +1,12 @@
 package brickmux
 
 import (
+	"fmt"
+	config "github.com/spf13/viper"
+	"io/ioutil"
 	"os"
 	"path"
-
-	config "github.com/spf13/viper"
+	"strings"
 )
 
 // getBrickVolfilePath returns path correspponding to the volfileID. Since, brick volfiles are now stored on
@@ -20,3 +22,24 @@ func getBrickVolfilePath(volfileID string) (string, error) {
 
 	return volfilePath, nil
 }
+
+// glusterdGetSockFromBrickPid returns the socket file from the /proc/
+// filesystem for the concerned process running with the same pid
+func glusterdGetSockFromBrickPid(pid int) (string, error) {
+	content, err := ioutil.ReadFile(fmt.Sprintf("/proc/%d/cmdline", pid))
+	if err != nil {
+		return "", err
+	}
+
+	parts := strings.Split(string(content), "\x00")
+	prevPart := ""
+	socketFile := ""
+	for _, p := range parts {
+		if prevPart == "-S" {
+			socketFile = p
+			break
+		}
+		prevPart = p
+	}
+	return socketFile, nil
+}