fix: cleanup tap flows and services for deleted guest and host (#14503)

Co-authored-by: Qiu Jian <qiujian@yunionyun.com>

pkg/compute/models/guests.go

@@ -4071,19 +4071,22 @@ func (self *SGuest) Delete(ctx context.Context, userCred mcclient.TokenCredentia
 	return nil
 }
 
-func (self *SGuest) RealDelete(ctx context.Context, userCred mcclient.TokenCredential) error {
+func (self *SGuest) CleanTapRecords(ctx context.Context, userCred mcclient.TokenCredential) error {
 	// delete tap devices
-	if srvs, err := NetTapServiceManager.getTapServicesByGuestId(self.Id, false); err != nil {
+	if err := NetTapServiceManager.removeTapServicesByGuestId(ctx, userCred, self.Id); err != nil {
 		return errors.Wrap(err, "NetTapServiceManager.getTapServicesByGuestId")
-	} else {
-		for _, srv := range srvs {
-			err := srv.cleanup(ctx, userCred)
-			if err != nil {
-				return errors.Wrap(err, "srv.Delete")
-			}
-		}
 	}
+	if err := NetTapFlowManager.removeTapFlowsByGuestId(ctx, userCred, self.Id); err != nil {
+		return errors.Wrap(err, "NetTapFlowManager.getTapServicesByGuestId")
+	}
+	return nil
+}
 
+func (self *SGuest) RealDelete(ctx context.Context, userCred mcclient.TokenCredential) error {
+	err := self.CleanTapRecords(ctx, userCred)
+	if err != nil {
+		return errors.Wrap(err, "CleanTapRecords")
+	}
 	return self.SVirtualResourceBase.Delete(ctx, userCred)
 }
 

pkg/compute/models/host_taps.go

@@ -34,7 +34,11 @@ func (h *SHost) GetDetailsTapConfig(ctx context.Context, userCred mcclient.Token
 	for _, flow := range flows {
 		mirror, err := flow.getMirrorConfig(true)
 		if err != nil {
-			return conf, errors.Wrap(err, "flow.getMirrorConfig")
+			if errors.Cause(err) == errors.ErrNotFound {
+				continue
+			} else {
+				return conf, errors.Wrap(err, "flow.getMirrorConfig")
+			}
 		}
 		mirrors = append(mirrors, mirror)
 	}

pkg/compute/models/hosts.go

@@ -782,15 +782,11 @@ func (self *SHost) RealDelete(ctx context.Context, userCred mcclient.TokenCreden
 	DeleteResourceJointSchedtags(self, ctx, userCred)
 
 	// delete tap devices
-	if srvs, err := NetTapServiceManager.getTapServicesByHostId(self.Id, false); err != nil {
+	if err := NetTapServiceManager.removeTapServicesByHostId(ctx, userCred, self.Id); err != nil {
 		return errors.Wrap(err, "NetTapServiceManager.getTapServicesByHostId")
-	} else {
-		for _, srv := range srvs {
-			err := srv.cleanup(ctx, userCred)
-			if err != nil {
-				return errors.Wrap(err, "srv.Delete")
-			}
-		}
+	}
+	if err := NetTapFlowManager.removeTapFlowsByHostId(ctx, userCred, self.Id); err != nil {
+		return errors.Wrap(err, "NetTapFlowManager.getTapFlowsByHostId")
 	}
 
 	IsolatedDeviceManager.DeleteDevicesByHost(ctx, userCred, self)

pkg/compute/models/net_tap_flows.go

@@ -210,17 +210,25 @@ func (flow *SNetTapFlow) getMoreDetails(ctx context.Context, details api.NetTapF
 	switch flow.Type {
 	case api.TapFlowVSwitch:
 		host := HostManager.FetchHostById(flow.SourceId)
-		details.Source = host.Name
-		details.SourceIps = host.AccessIp
+		if host != nil {
+			details.Source = host.Name
+			details.SourceIps = host.AccessIp
+		}
 		wire := WireManager.FetchWireById(flow.NetId)
-		details.Net = wire.Name
+		if wire != nil {
+			details.Net = wire.Name
+		}
 	case api.TapFlowGuestNic:
 		guest := GuestManager.FetchGuestById(flow.SourceId)
-		details.Source = guest.Name
-		ret := fetchGuestIPs([]string{flow.SourceId}, tristate.False)
-		details.SourceIps = strings.Join(ret[flow.SourceId], ",")
+		if guest != nil {
+			details.Source = guest.Name
+			ret := fetchGuestIPs([]string{flow.SourceId}, tristate.False)
+			details.SourceIps = strings.Join(ret[flow.SourceId], ",")
+		}
 		netObj, _ := NetworkManager.FetchById(flow.NetId)
-		details.Net = netObj.GetName()
+		if netObj != nil {
+			details.Net = netObj.GetName()
+		}
 	}
 	return details
 }
@@ -432,6 +440,10 @@ func (flow *SNetTapFlow) getMirrorConfig(needTapHostIp bool) (api.SMirrorConfig,
 		wireId = flow.NetId
 	case api.TapFlowGuestNic:
 		guest := GuestManager.FetchGuestById(flow.SourceId)
+		if guest == nil {
+			// guest has been deleted?
+			return ret, errors.Wrap(errors.ErrNotFound, "source not found")
+		}
 		gn, err := guest.GetGuestnetworkByMac(flow.MacAddr)
 		if err != nil {
 			return ret, errors.Wrap(err, "GetGuestnetworkByMac")
@@ -463,3 +475,37 @@ func (flow *SNetTapFlow) getMirrorConfig(needTapHostIp bool) (api.SMirrorConfig,
 	ret.Direction = flow.Direction
 	return ret, nil
 }
+
+func (manager *SNetTapFlowManager) removeTapFlowsByGuestId(ctx context.Context, userCred mcclient.TokenCredential, sourceId string) error {
+	return manager.removeTapFlows(ctx, userCred, api.TapFlowGuestNic, sourceId)
+}
+
+func (manager *SNetTapFlowManager) removeTapFlowsByHostId(ctx context.Context, userCred mcclient.TokenCredential, sourceId string) error {
+	return manager.removeTapFlows(ctx, userCred, api.TapFlowVSwitch, sourceId)
+}
+
+func (manager *SNetTapFlowManager) removeTapFlows(ctx context.Context, userCred mcclient.TokenCredential, srvType string, targetId string) error {
+	srvs, err := manager.getTapFlows(srvType, targetId)
+	if err != nil {
+		return errors.Wrap(err, "getTapServicesByHostId")
+	}
+	for i := range srvs {
+		err := srvs[i].Delete(ctx, userCred)
+		if err != nil {
+			return errors.Wrap(err, "Delete")
+		}
+	}
+	return nil
+}
+
+func (manager *SNetTapFlowManager) getTapFlows(srvType string, sourceId string) ([]SNetTapFlow, error) {
+	q := manager.Query()
+	q = q.Equals("type", srvType)
+	q = q.Equals("source_id", sourceId)
+	ret := make([]SNetTapFlow, 0)
+	err := db.FetchModelObjects(manager, q, &ret)
+	if err != nil {
+		return nil, errors.Wrap(err, "FetchModelObjects")
+	}
+	return ret, nil
+}

pkg/compute/models/net_tap_services.go

@@ -548,3 +548,25 @@ func (srv *SNetTapService) cleanup(ctx context.Context, userCred mcclient.TokenC
 	}
 	return nil
 }
+
+func (manager *SNetTapServiceManager) removeTapServicesByGuestId(ctx context.Context, userCred mcclient.TokenCredential, targetId string) error {
+	return manager.removeTapServices(ctx, userCred, api.TapServiceGuest, targetId)
+}
+
+func (manager *SNetTapServiceManager) removeTapServicesByHostId(ctx context.Context, userCred mcclient.TokenCredential, targetId string) error {
+	return manager.removeTapServices(ctx, userCred, api.TapServiceHost, targetId)
+}
+
+func (manager *SNetTapServiceManager) removeTapServices(ctx context.Context, userCred mcclient.TokenCredential, srvType string, targetId string) error {
+	srvs, err := manager.getTapServices(srvType, targetId, false)
+	if err != nil {
+		return errors.Wrap(err, "getTapServicesByHostId")
+	}
+	for i := range srvs {
+		err := srvs[i].cleanup(ctx, userCred)
+		if err != nil {
+			return errors.Wrap(err, "cleanup")
+		}
+	}
+	return nil
+}

pkg/compute/tasks/guest_delete_task.go

@@ -203,6 +203,9 @@ func (self *GuestDeleteTask) OnSyncConfigComplete(ctx context.Context, obj db.IS
 
 	// try to leave all groups
 	guest.LeaveAllGroups(ctx, self.UserCred)
+	// cleanup tap services and flows
+	guest.CleanTapRecords(ctx, self.UserCred)
+
 	isPurge := jsonutils.QueryBoolean(self.Params, "purge", false)
 	overridePendingDelete := jsonutils.QueryBoolean(self.Params, "override_pending_delete", false)