fix(cloudcommon): avoid service exit when etcd unreachable

pkg/cloudcommon/database.go

@@ -19,9 +19,11 @@ import (
 	"database/sql"
 	"fmt"
 	"net/http"
+	"time"
 
 	"yunion.io/x/jsonutils"
 	"yunion.io/x/log"
+	"yunion.io/x/pkg/errors"
 	"yunion.io/x/sqlchemy"
 
 	"yunion.io/x/onecloud/pkg/appsrv"
@@ -86,25 +88,45 @@ func InitDB(options *common_options.DBOptions) {
 	}
 	// lm := lockman.NewNoopLockManager()
 
-	if len(options.EtcdEndpoints) != 0 {
-		log.Infof("using etcd as resource informer backend")
-		tlsCfg, err := options.GetEtcdTLSConfig()
-		if err != nil {
-			log.Fatalf("get etcd informer backend tls config err: %v", err)
+	startInitInformer(options)
+}
+
+// startInitInformer starts goroutine init informer backend
+func startInitInformer(options *common_options.DBOptions) {
+	go func() {
+		if len(options.EtcdEndpoints) == 0 {
+			return
 		}
-		informerBackend, err := informer.NewEtcdBackend(&etcd.SEtcdOptions{
-			EtcdEndpoint:              options.EtcdEndpoints,
-			EtcdTimeoutSeconds:        5,
-			EtcdRequestTimeoutSeconds: 2,
-			EtcdLeaseExpireSeconds:    5,
-			EtcdEnabldSsl:             options.EtcdUseTLS,
-			TLSConfig:                 tlsCfg,
-		}, nil)
-		if err != nil {
-			log.Fatalf("new etcd informer backend error: %v", err)
+		for {
+			log.Infof("using etcd as resource informer backend")
+			if err := initInformer(options); err != nil {
+				log.Errorf("Init informer error: %v", err)
+				time.Sleep(10 * time.Second)
+			} else {
+				break
+			}
 		}
-		informer.Init(informerBackend)
+	}()
+}
+
+func initInformer(options *common_options.DBOptions) error {
+	tlsCfg, err := options.GetEtcdTLSConfig()
+	if err != nil {
+		return errors.Wrap(err, "get etcd informer backend tls config")
 	}
+	informerBackend, err := informer.NewEtcdBackend(&etcd.SEtcdOptions{
+		EtcdEndpoint:              options.EtcdEndpoints,
+		EtcdTimeoutSeconds:        5,
+		EtcdRequestTimeoutSeconds: 2,
+		EtcdLeaseExpireSeconds:    5,
+		EtcdEnabldSsl:             options.EtcdUseTLS,
+		TLSConfig:                 tlsCfg,
+	}, nil)
+	if err != nil {
+		return errors.Wrap(err, "new etcd informer backend")
+	}
+	informer.Init(informerBackend)
+	return nil
 }
 
 func CloseDB() {

pkg/cloudcommon/syncman/watcher/watcher.go

@@ -82,12 +82,11 @@ func (manager *SInformerSyncManager) startWatcher() error {
 	log.Infof("[%s] Start resource informer watcher for %s", manager.Name(), manager.resourceManager.GetKeyword())
 	ctx := context.Background()
 	s := auth.GetAdminSession(ctx, consts.GetRegion(), "v1")
-	watchMan, err := informer.NewWatchManagerBySession(s)
-	if err != nil {
-		return errors.Wrap(err, "NewWatchManagerBySession")
-	}
-	if err := watchMan.For(manager.resourceManager).AddEventHandler(ctx, manager); err != nil {
-		return errors.Wrapf(err, "watch resource %s", manager.resourceManager.GetKeyword())
-	}
+	informer.NewWatchManagerBySessionBg(s, func(watchMan *informer.SWatchManager) error {
+		if err := watchMan.For(manager.resourceManager).AddEventHandler(ctx, manager); err != nil {
+			return errors.Wrapf(err, "watch resource %s", manager.resourceManager.GetKeyword())
+		}
+		return nil
+	})
 	return nil
 }