Merge pull request #10850 from yousong/automated-cherry-pick-of-#10849-upstream-release-3.7

Automated cherry pick of #10849: Feature/yousong zombie

cmd/ansibleserver/main.go

@@ -15,9 +15,14 @@
 package main
 
 import (
+	"context"
+
 	"yunion.io/x/onecloud/pkg/ansibleserver/service"
+	"yunion.io/x/onecloud/pkg/util/procutils"
 )
 
 func main() {
+	go procutils.WaitZombieLoop(context.TODO())
+
 	service.StartService()
 }

cmd/baremetal-agent/main.go

@@ -15,12 +15,17 @@
 package main
 
 import (
+	"context"
+
 	"yunion.io/x/onecloud/pkg/baremetal/service"
 	"yunion.io/x/onecloud/pkg/util/atexit"
+	"yunion.io/x/onecloud/pkg/util/procutils"
 )
 
 func main() {
 	defer atexit.Handle()
 
+	go procutils.WaitZombieLoop(context.TODO())
+
 	service.New().StartService()
 }

cmd/esxi-agent/main.go

@@ -15,12 +15,17 @@
 package main
 
 import (
+	"context"
+
 	"yunion.io/x/onecloud/pkg/esxi/service"
 	"yunion.io/x/onecloud/pkg/util/atexit"
+	"yunion.io/x/onecloud/pkg/util/procutils"
 )
 
 func main() {
 	defer atexit.Handle()
 
+	go procutils.WaitZombieLoop(context.TODO())
+
 	service.New().StartService()
 }

cmd/glance/main.go

@@ -15,12 +15,17 @@
 package main
 
 import (
+	"context"
+
 	"yunion.io/x/onecloud/pkg/image/service"
 	"yunion.io/x/onecloud/pkg/util/atexit"
+	"yunion.io/x/onecloud/pkg/util/procutils"
 )
 
 func main() {
 	defer atexit.Handle()
 
+	go procutils.WaitZombieLoop(context.TODO())
+
 	service.StartService()
 }

cmd/host-deployer/main.go

@@ -14,9 +14,16 @@
 
 package main
 
-import "yunion.io/x/onecloud/pkg/hostman/hostdeployer/deployserver"
+import (
+	"context"
+
+	"yunion.io/x/onecloud/pkg/hostman/hostdeployer/deployserver"
+	"yunion.io/x/onecloud/pkg/util/procutils"
+)
 
 func main() {
+	go procutils.WaitZombieLoop(context.TODO())
+
 	deployer := deployserver.NewDeployService()
 	deployer.StartService()
 }

cmd/host/main.go

@@ -15,14 +15,19 @@
 package main
 
 import (
+	"context"
+
 	"yunion.io/x/onecloud/pkg/cloudcommon/service"
 	"yunion.io/x/onecloud/pkg/hostman"
 	"yunion.io/x/onecloud/pkg/util/atexit"
+	"yunion.io/x/onecloud/pkg/util/procutils"
 )
 
 func main() {
 	defer atexit.Handle()
 
+	go procutils.WaitZombieLoop(context.TODO())
+
 	var srv = &hostman.SHostService{}
 	srv.SServiceBase = &service.SServiceBase{
 		Service: srv,

cmd/vpcagent/main.go

@@ -26,6 +26,7 @@ import (
 	app_common "yunion.io/x/onecloud/pkg/cloudcommon/app"
 	common_options "yunion.io/x/onecloud/pkg/cloudcommon/options"
 	"yunion.io/x/onecloud/pkg/util/atexit"
+	"yunion.io/x/onecloud/pkg/util/procutils"
 	"yunion.io/x/onecloud/pkg/vpcagent/options"
 	_ "yunion.io/x/onecloud/pkg/vpcagent/ovn"
 	"yunion.io/x/onecloud/pkg/vpcagent/worker"
@@ -52,8 +53,12 @@ func main() {
 	}
 
 	{
+
+		ctx := context.Background()
+		ctx, cancelFunc := context.WithCancel(ctx)
+		go procutils.WaitZombieLoop(ctx)
+
 		wg := &sync.WaitGroup{}
-		ctx, cancelFunc := context.WithCancel(context.Background())
 		ctx = context.WithValue(ctx, "wg", wg)
 		wg.Add(1)
 		go w.Start(ctx)

pkg/ansibleserver/service/service.go

@@ -15,7 +15,6 @@
 package service
 
 import (
-	"context"
 	"os"
 
 	_ "github.com/go-sql-driver/mysql"
@@ -49,7 +48,5 @@ func StartService() {
 	db.EnsureAppInitSyncDB(app, dbOpts, models.InitDB)
 	defer cloudcommon.CloseDB()
 
-	go ReapZomebieLoop(context.TODO())
-
 	common_app.ServeForever(app, baseOpts)
 }

pkg/util/procutils/zombie.go

@@ -12,11 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-package service
+package procutils
 
 import (
 	"context"
-	"fmt"
 	"io/ioutil"
 	"os"
 	"path/filepath"
@@ -28,13 +27,13 @@ import (
 	"yunion.io/x/log"
 )
 
-func ReapZomebieLoop(ctx context.Context) {
+func WaitZombieLoop(ctx context.Context) {
 	myPid := os.Getpid()
 	if myPid != 1 {
 		log.Infof("My pid is not 1 and no need to wait zombies")
 		return
 	}
-	myPidStr := fmt.Sprintf("%d", myPid)
+	const myPidStr = "1"
 
 	tick := time.NewTicker(31 * time.Second)
 	for {
@@ -69,11 +68,11 @@ func ReapZomebieLoop(ctx context.Context) {
 			dataStr := string(data)
 			items := strings.Split(dataStr, " ")
 			const (
-				idxPid    = iota
-				idxName   = iota
-				idxState  = iota
-				idxPpid   = iota
-				idxMinLen = iota
+				idxPid = iota
+				idxName
+				idxState
+				idxPpid
+				idxMinLen
 			)
 			if len(items) < idxMinLen {
 				log.Errorf("%s contains less than %d items: %s", statPath, idxMinLen, dataStr)
@@ -97,11 +96,7 @@ func ReapZomebieLoop(ctx context.Context) {
 				log.Errorf("%s: %s has invalid pid number %q: %v", pname, statPath, pidStr, err)
 				continue
 			}
-			var (
-				status syscall.WaitStatus
-				rusage syscall.Rusage
-			)
-			pid1, err := syscall.Wait4(pid, &status, 0, &rusage)
+			pid1, err := syscall.Wait4(pid, nil, 0, nil)
 			if err != nil {
 				log.Errorf("%s: %s: wait error: %v", pname, statPath, err)
 				continue