[master] refactor func ShouldGrowVolumes (#5884)

weed/server/master_grpc_server_volume.go

@@ -27,7 +27,7 @@ func (ms *MasterServer) DoAutomaticVolumeGrow(req *topology.VolumeGrowRequest) {
 	newVidLocations, err := ms.vg.AutomaticGrowByType(req.Option, ms.grpcDialOption, ms.Topo, req.Count)
 	glog.V(1).Infoln("finished automatic volume grow, cost ", time.Now().Sub(start))
 	if err != nil {
-		glog.Warningf("automatic volume grow %s: %+v", req.Option, err)
+		glog.V(1).Infof("automatic volume grow failed: %+v", err)
 		return
 	}
 	for _, newVidLocation := range newVidLocations {
@@ -38,19 +38,37 @@ func (ms *MasterServer) DoAutomaticVolumeGrow(req *topology.VolumeGrowRequest) {
 func (ms *MasterServer) ProcessGrowRequest() {
 	go func() {
 		for {
-			time.Sleep(14*time.Minute + time.Duration(120*rand.Float32())*time.Second)
 			if !ms.Topo.IsLeader() {
 				continue
 			}
-			for _, vl := range ms.Topo.ListVolumeLayouts() {
+			dcs := ms.Topo.ListDataCenters()
-				if !vl.HasGrowRequest() && vl.ShouldGrowVolumes(&topology.VolumeGrowOption{}) {
+			for _, vlc := range ms.Topo.ListVolumeLayoutCollections() {
+				vl := vlc.VolumeLayout
+				if vl.HasGrowRequest() {
+					continue
+				}
+				if vl.ShouldGrowVolumes(vlc.Collection) {
 					vl.AddGrowRequest()
 					ms.volumeGrowthRequestChan <- &topology.VolumeGrowRequest{
-						Option: vl.ToGrowOption(),
+						Option: vlc.ToGrowOption(),
 						Count:  vl.GetLastGrowCount(),
 					}
+				} else {
+					for _, dc := range dcs {
+						if vl.ShouldGrowVolumesByDataNode("DataCenter", dc) {
+							vl.AddGrowRequest()
+							volumeGrowOption := vlc.ToGrowOption()
+							volumeGrowOption.DataCenter = dc
+							ms.volumeGrowthRequestChan <- &topology.VolumeGrowRequest{
+								Option: volumeGrowOption,
+								Count:  vl.GetLastGrowCount(),
+								Force:  true,
+							}
+						}
+					}
 				}
 			}
+			time.Sleep(14*time.Minute + time.Duration(120*rand.Float32())*time.Second)
 		}
 	}()
 	go func() {
@@ -81,19 +99,20 @@ func (ms *MasterServer) ProcessGrowRequest() {
 			})
 
 			// not atomic but it's okay
-			if !found && vl.ShouldGrowVolumes(option) {
+			if found || (!req.Force && !vl.ShouldGrowVolumes(req.Option.Collection)) {
-				filter.Store(req, nil)
-				// we have lock called inside vg
-				go func(req *topology.VolumeGrowRequest, vl *topology.VolumeLayout) {
-					ms.DoAutomaticVolumeGrow(req)
-					vl.DoneGrowRequest()
-					filter.Delete(req)
-				}(req, vl)
-			} else {
 				glog.V(4).Infoln("discard volume grow request")
 				time.Sleep(time.Millisecond * 211)
 				vl.DoneGrowRequest()
+				continue
 			}
+
+			filter.Store(req, nil)
+			// we have lock called inside vg
+			go func(req *topology.VolumeGrowRequest, vl *topology.VolumeLayout) {
+				ms.DoAutomaticVolumeGrow(req)
+				vl.DoneGrowRequest()
+				filter.Delete(req)
+			}(req, vl)
 		}
 	}()
 }

weed/stats/metrics.go

@@ -70,13 +70,21 @@ var (
 			Help:      "replica placement mismatch",
 		}, []string{"collection", "id"})
 
-	MasterVolumeLayout = prometheus.NewGaugeVec(
+	MasterVolumeLayoutWritable = prometheus.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Namespace: Namespace,
 			Subsystem: "master",
-			Name:      "volume_layout_total",
+			Name:      "volume_layout_writable",
-			Help:      "Number of volumes in volume layouts",
+			Help:      "Number of writable volumes in volume layouts",
-		}, []string{"collection", "dataCenter", "type"})
+		}, []string{"collection", "disk", "rp", "ttl"})
+
+	MasterVolumeLayoutCrowded = prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Namespace: Namespace,
+			Subsystem: "master",
+			Name:      "volume_layout_crowded",
+			Help:      "Number of crowded volumes in volume layouts",
+		}, []string{"collection", "disk", "rp", "ttl"})
 
 	MasterPickForWriteErrorCounter = prometheus.NewCounter(
 		prometheus.CounterOpts{
@@ -281,7 +289,8 @@ func init() {
 	Gather.MustRegister(MasterReceivedHeartbeatCounter)
 	Gather.MustRegister(MasterLeaderChangeCounter)
 	Gather.MustRegister(MasterReplicaPlacementMismatch)
-	Gather.MustRegister(MasterVolumeLayout)
+	Gather.MustRegister(MasterVolumeLayoutWritable)
+	Gather.MustRegister(MasterVolumeLayoutCrowded)
 
 	Gather.MustRegister(FilerRequestCounter)
 	Gather.MustRegister(FilerHandlerCounter)

weed/topology/topology.go

@@ -329,7 +329,7 @@ func (t *Topology) UnRegisterVolumeLayout(v storage.VolumeInfo, dn *DataNode) {
 }
 
 func (t *Topology) DataCenterExists(dcName string) bool {
-	return dcName == "" || t.GetOrCreateDataCenter(dcName) != nil
+	return dcName == "" || t.GetDataCenter(dcName) != nil
 }
 
 func (t *Topology) GetDataCenter(dcName string) (dc *DataCenter) {
@@ -358,6 +358,15 @@ func (t *Topology) GetOrCreateDataCenter(dcName string) *DataCenter {
 	return dc
 }
 
+func (t *Topology) ListDataCenters() (dcs []string) {
+	t.RLock()
+	defer t.RUnlock()
+	for _, c := range t.children {
+		dcs = append(dcs, string(c.(*DataCenter).Id()))
+	}
+	return dcs
+}
+
 func (t *Topology) SyncDataNodeRegistration(volumes []*master_pb.VolumeInformationMessage, dn *DataNode) (newVolumes, deletedVolumes []storage.VolumeInfo) {
 	// convert into in memory struct storage.VolumeInfo
 	var volumeInfos []storage.VolumeInfo

weed/topology/topology_info.go

@@ -13,6 +13,11 @@ type TopologyInfo struct {
 	Layouts     []VolumeLayoutInfo `json:"Layouts"`
 }
 
+type VolumeLayoutCollection struct {
+	Collection   string
+	VolumeLayout *VolumeLayout
+}
+
 func (t *Topology) ToInfo() (info TopologyInfo) {
 	info.Max = t.diskUsages.GetMaxVolumeCount()
 	info.Free = t.diskUsages.FreeSpace()
@@ -42,10 +47,12 @@ func (t *Topology) ToInfo() (info TopologyInfo) {
 	return
 }
 
-func (t *Topology) ListVolumeLayouts() (volumeLayouts []*VolumeLayout) {
+func (t *Topology) ListVolumeLayoutCollections() (volumeLayouts []*VolumeLayoutCollection) {
 	for _, col := range t.collectionMap.Items() {
 		for _, volumeLayout := range col.(*Collection).storageType2VolumeLayout.Items() {
-			volumeLayouts = append(volumeLayouts, volumeLayout.(*VolumeLayout))
+			volumeLayouts = append(volumeLayouts,
+				&VolumeLayoutCollection{col.(*Collection).Name, volumeLayout.(*VolumeLayout)},
+			)
 		}
 	}
 	return volumeLayouts

weed/topology/volume_growth.go

@@ -28,6 +28,7 @@ This package is created to resolve these replica placement issues:
 type VolumeGrowRequest struct {
 	Option *VolumeGrowOption
 	Count  uint32
+	Force  bool
 }
 
 type volumeGrowthStrategy struct {

weed/topology/volume_layout.go

@@ -3,7 +3,7 @@ package topology
 import (
 	"fmt"
 	"github.com/seaweedfs/seaweedfs/weed/stats"
-	"math/rand/v2"
+	"math/rand"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -218,17 +218,21 @@ func (vl *VolumeLayout) EnsureCorrectWritables(v *storage.VolumeInfo) {
 }
 
 func (vl *VolumeLayout) ensureCorrectWritables(vid needle.VolumeId) {
-	if vl.enoughCopies(vid) && vl.isAllWritable(vid) {
+	isEnoughCopies := vl.enoughCopies(vid)
-		if !vl.oversizedVolumes.IsTrue(vid) {
+	isAllWritable := vl.isAllWritable(vid)
-			vl.setVolumeWritable(vid)
+	isOversizedVolume := vl.oversizedVolumes.IsTrue(vid)
-		}
+	if isEnoughCopies && isAllWritable && !isOversizedVolume {
+		vl.setVolumeWritable(vid)
 	} else {
-		if !vl.enoughCopies(vid) {
+		if !isEnoughCopies {
 			glog.V(0).Infof("volume %d does not have enough copies", vid)
 		}
-		if !vl.isAllWritable(vid) {
+		if !isAllWritable {
 			glog.V(0).Infof("volume %d are not all writable", vid)
 		}
+		if isOversizedVolume {
+			glog.V(1).Infof("volume %d are oversized", vid)
+		}
 		glog.V(0).Infof("volume %d remove from writable", vid)
 		vl.removeFromWritable(vid)
 	}
@@ -254,6 +258,10 @@ func (vl *VolumeLayout) isOversized(v *storage.VolumeInfo) bool {
 	return uint64(v.Size) >= vl.volumeSizeLimit
 }
 
+func (vl *VolumeLayout) isCrowdedVolume(v *storage.VolumeInfo) bool {
+	return float64(v.Size) > float64(vl.volumeSizeLimit)*VolumeGrowStrategy.Threshold
+}
+
 func (vl *VolumeLayout) isWritable(v *storage.VolumeInfo) bool {
 	return !vl.isOversized(v) &&
 		v.Version == needle.CurrentVersion &&
@@ -296,7 +304,7 @@ func (vl *VolumeLayout) PickForWrite(count uint64, option *VolumeGrowOption) (vi
 		return 0, 0, nil, true, fmt.Errorf("%s in volume layout", noWritableVolumes)
 	}
 	if option.DataCenter == "" && option.Rack == "" && option.DataNode == "" {
-		vid := vl.writables[rand.IntN(lenWriters)]
+		vid := vl.writables[rand.Intn(lenWriters)]
 		locationList = vl.vid2location[vid]
 		if locationList == nil || len(locationList.list) == 0 {
 			return 0, 0, nil, false, fmt.Errorf("Strangely vid %s is on no machine!", vid.String())
@@ -351,40 +359,45 @@ func (vl *VolumeLayout) GetLastGrowCount() uint32 {
 	return vl.lastGrowCount.Load()
 }
 
-func (vl *VolumeLayout) ShouldGrowVolumes(option *VolumeGrowOption) bool {
+func (vl *VolumeLayout) ShouldGrowVolumes(collection string) bool {
-	total, active, crowded := vl.GetActiveVolumeCount(option)
+	writable, crowded := vl.GetWritableVolumeCount()
-	stats.MasterVolumeLayout.WithLabelValues(option.Collection, option.DataCenter, "total").Set(float64(total))
+	stats.MasterVolumeLayoutWritable.WithLabelValues(collection, vl.diskType.String(), vl.rp.String(), vl.ttl.String()).Set(float64(writable))
-	stats.MasterVolumeLayout.WithLabelValues(option.Collection, option.DataCenter, "active").Set(float64(active))
+	stats.MasterVolumeLayoutCrowded.WithLabelValues(collection, vl.diskType.String(), vl.rp.String(), vl.ttl.String()).Set(float64(crowded))
-	stats.MasterVolumeLayout.WithLabelValues(option.Collection, option.DataCenter, "crowded").Set(float64(crowded))
+	return writable <= crowded
-	//glog.V(0).Infof("active volume: %d, high usage volume: %d\n", active, high)
-	return active <= crowded
 }
 
-func (vl *VolumeLayout) GetActiveVolumeCount(option *VolumeGrowOption) (total, active, crowded int) {
+func (vl *VolumeLayout) ShouldGrowVolumesByDataNode(nodeType string, dataNode string) bool {
 	vl.accessLock.RLock()
-	defer vl.accessLock.RUnlock()
+	writables := make([]needle.VolumeId, len(vl.writables))
-	if option.DataCenter == "" {
+	copy(writables, vl.writables)
-		return len(vl.writables), len(vl.writables), len(vl.crowded)
+	vl.accessLock.RUnlock()
-	}
+
-	total = len(vl.writables)
+	dataNodeId := NodeId(dataNode)
-	for _, v := range vl.writables {
+	for _, v := range writables {
 		for _, dn := range vl.vid2location[v].list {
-			if dn.GetDataCenter().Id() == NodeId(option.DataCenter) {
+			dataNodeFound := false
-				if option.Rack != "" && dn.GetRack().Id() != NodeId(option.Rack) {
+			switch nodeType {
-					continue
+			case "DataCenter":
-				}
+				dataNodeFound = dn.GetDataCenter().Id() == dataNodeId
-				if option.DataNode != "" && dn.Id() != NodeId(option.DataNode) {
+			case "Rack":
-					continue
+				dataNodeFound = dn.GetRack().Id() == dataNodeId
-				}
+			case "DataNode":
-				active++
+				dataNodeFound = dn.Id() == dataNodeId
-				info, _ := dn.GetVolumesById(v)
+			}
-				if float64(info.Size) > float64(vl.volumeSizeLimit)*VolumeGrowStrategy.Threshold {
+			if dataNodeFound {
-					crowded++
+				if info, err := dn.GetVolumesById(v); err == nil && !vl.isCrowdedVolume(&info) {
+					return false
 				}
 			}
 		}
 	}
-	return
+	return true
+}
+
+func (vl *VolumeLayout) GetWritableVolumeCount() (active, crowded int) {
+	vl.accessLock.RLock()
+	defer vl.accessLock.RUnlock()
+	return len(vl.writables), len(vl.crowded)
 }
 
 func (vl *VolumeLayout) removeFromWritable(vid needle.VolumeId) bool {
@@ -531,12 +544,13 @@ func (vl *VolumeLayout) ToInfo() (info VolumeLayoutInfo) {
 	return
 }
 
-func (vl *VolumeLayout) ToGrowOption() (option *VolumeGrowOption) {
+func (vlc *VolumeLayoutCollection) ToGrowOption() (option *VolumeGrowOption) {
-	option = &VolumeGrowOption{}
+	return &VolumeGrowOption{
-	option.ReplicaPlacement = vl.rp
+		Collection:       vlc.Collection,
-	option.Ttl = vl.ttl
+		ReplicaPlacement: vlc.VolumeLayout.rp,
-	option.DiskType = vl.diskType
+		Ttl:              vlc.VolumeLayout.ttl,
-	return
+		DiskType:         vlc.VolumeLayout.diskType,
+	}
 }
 
 func (vl *VolumeLayout) Stats() *VolumeLayoutStats {