Merge pull request #18641 from wanyaoqi/automated-cherry-pick-of-#186…

…39-upstream-master Automated cherry pick of #18639: fix(region,host): sriov nic hotplug and metrics fix
yunionio · Nov 13, 2023 · ee80ff4 · ee80ff4
2 parents 4116a06 + 0c98109
commit ee80ff4
Show file tree

Hide file tree

Showing 4 changed files with 35 additions and 25 deletions.
diff --git a/pkg/apis/compute/guestnetwork.go b/pkg/apis/compute/guestnetwork.go
@@ -141,4 +141,6 @@ type GuestnetworkJsonDesc struct {
 type SNicTrafficRecord struct {
 	RxTraffic int64
 	TxTraffic int64
+
+	HasBeenSetDown bool
 }
diff --git a/pkg/compute/models/guest_actions.go b/pkg/compute/models/guest_actions.go
@@ -2571,14 +2571,11 @@ func (self *SGuest) PerformAttachnetwork(ctx context.Context, userCred mcclient.
 			quotas.CancelPendingUsage(ctx, userCred, pendingUsage, pendingUsage, false)
 			return nil, httperrors.NewBadRequestError("%v", err)
 		}
-		net := gns[0].GetNetwork()
 		if input.Nets[i].SriovDevice != nil {
-			input.Nets[i].SriovDevice.NetworkIndex = &gns[0].Index
-			input.Nets[i].SriovDevice.WireId = net.WireId
-			err = self.createIsolatedDeviceOnHost(ctx, userCred, host, input.Nets[i].SriovDevice, pendingUsageHost)
+			err = self.allocSriovNicDevice(ctx, userCred, host, &gns[0], input.Nets[i], pendingUsageHost)
 			if err != nil {
 				quotas.CancelPendingUsage(ctx, userCred, pendingUsageHost, pendingUsageHost, false)
-				return nil, errors.Wrap(err, "self.createIsolatedDeviceOnHost")
+				return nil, errors.Wrap(err, "self.allocSriovNicDevice")
 			}
 		}
 	}

diff --git a/pkg/hostman/guestman/guesttasks.go b/pkg/hostman/guestman/guesttasks.go
@@ -712,18 +712,18 @@ func (n *SGuestNetworkSyncTask) delNicDevice(nic *desc.SGuestNetwork) {
 }
 
 func (n *SGuestNetworkSyncTask) addNic(nic *desc.SGuestNetwork) {
-	if nic.Driver == "vfio-pci" {
-		// vfio device will add on isolated devices sync task
-		n.onDeviceAdd(nic)
-		return
-	}
-
 	if err := n.guest.generateNicScripts(nic); err != nil {
 		log.Errorln(err)
 		n.errors = append(n.errors, err)
 		n.syncNetworkConf()
 		return
 	}
+
+	if nic.Driver == "vfio-pci" {
+		// vfio device will add on isolated devices sync task
+		n.onDeviceAdd(nic)
+		return
+	}
 	var (
 		netType    = "tap"
 		upscript   = n.guest.getNicUpScriptPath(nic)

diff --git a/pkg/hostman/hostmetrics/hostmetrics.go b/pkg/hostman/hostmetrics/hostmetrics.go
@@ -255,20 +255,18 @@ func (s *SGuestMonitorCollector) saveNicTraffics(reportData map[string]*GuestMet
 				log.Warningf("failed found report data for nic %s", gm.Nics[i].Ifname)
 				continue
 			}
+			nicHasBeenSetDown := false
 			nicTraffic := compute.SNicTrafficRecord{}
 			for index, record := range guestTrafficRecord {
 				if index == strconv.Itoa(nicIo.Meta.Index) {
 					nicTraffic.RxTraffic += record.RxTraffic
 					nicTraffic.TxTraffic += record.TxTraffic
+					nicHasBeenSetDown = record.HasBeenSetDown
 				}
 			}
 
 			if gm.Nics[i].RxTrafficLimit > 0 || gm.Nics[i].TxTrafficLimit > 0 {
-				var nicDown, nicHasBeenSetDown = false, false
-				if nicTraffic.RxTraffic >= gm.Nics[i].RxTrafficLimit || nicTraffic.RxTraffic >= gm.Nics[i].TxTrafficLimit {
-					// record traffic excced, nic must has been set down
-					nicHasBeenSetDown = true
-				}
+				var nicDown = false
 				if gm.Nics[i].RxTrafficLimit > 0 {
 					nicTraffic.RxTraffic += int64(nicIo.TimeDiff * nicIo.BPSRecv / 8)
 					if nicTraffic.RxTraffic >= gm.Nics[i].RxTrafficLimit {
@@ -283,11 +281,12 @@ func (s *SGuestMonitorCollector) saveNicTraffics(reportData map[string]*GuestMet
 						nicDown = true
 					}
 				}
-				guestTraffics[strconv.Itoa(nicIo.Meta.Index)] = nicTraffic
 				if !nicHasBeenSetDown && nicDown {
 					log.Infof("guest %s nic %d traffic exceed, set nic down", gm.Id, nicIo.Meta.Index)
 					gm.SetNicDown(nicIo.Meta.Index)
+					nicTraffic.HasBeenSetDown = true
 				}
+				guestTraffics[strconv.Itoa(nicIo.Meta.Index)] = nicTraffic
 			}
 		}
 		if len(guestTraffics) == 0 {
@@ -621,14 +620,26 @@ func (m *SGuestMonitor) Netio() []*NetIOMetric {
 		data.Meta.NetId = nic.NetId
 		data.Meta.Uptime, _ = host.Uptime()
 
-		data.BytesSent = nicStat.BytesRecv
-		data.BytesRecv = nicStat.BytesSent
-		data.PacketsRecv = nicStat.PacketsSent
-		data.PacketsSent = nicStat.PacketsRecv
-		data.ErrIn = nicStat.Errout
-		data.ErrOut = nicStat.Errin
-		data.DropIn = nicStat.Dropout
-		data.DropOut = nicStat.Dropin
+		if nic.Driver == "vfio-pci" {
+			data.BytesSent = nicStat.BytesSent
+			data.BytesRecv = nicStat.BytesRecv
+			data.PacketsRecv = nicStat.PacketsRecv
+			data.PacketsSent = nicStat.PacketsSent
+			data.ErrIn = nicStat.Errin
+			data.ErrOut = nicStat.Errout
+			data.DropIn = nicStat.Dropin
+			data.DropOut = nicStat.Dropout
+		} else {
+			data.BytesSent = nicStat.BytesRecv
+			data.BytesRecv = nicStat.BytesSent
+			data.PacketsRecv = nicStat.PacketsSent
+			data.PacketsSent = nicStat.PacketsRecv
+			data.ErrIn = nicStat.Errout
+			data.ErrOut = nicStat.Errin
+			data.DropIn = nicStat.Dropout
+			data.DropOut = nicStat.Dropin
+		}
+
 		res = append(res, data)
 	}
 	return res