Merge pull request #1219 from song-zhang/master

schedule new volume by free volume number of nodes

weed/topology/node.go

@@ -62,56 +62,64 @@ type NodeImpl struct {
 }
 
 // the first node must satisfy filterFirstNodeFn(), the rest nodes must have one free slot
-func (n *NodeImpl) RandomlyPickNodes(numberOfNodes int, filterFirstNodeFn func(dn Node) error) (firstNode Node, restNodes []Node, err error) {
-	candidates := make([]Node, 0, len(n.children))
+func (n *NodeImpl) PickNodesByWeight(numberOfNodes int, filterFirstNodeFn func(dn Node) error) (firstNode Node, restNodes []Node, err error) {
+	var totalWeights int64
 	var errs []string
 	n.RLock()
+	candidates := make([]Node, 0, len(n.children))
+	candidatesWeights := make([]int64, 0, len(n.children))
+	//pick nodes which has enough free volumes as candidates, and use free volumes number as node weight.
 	for _, node := range n.children {
-		if err := filterFirstNodeFn(node); err == nil {
-			candidates = append(candidates, node)
-		} else {
-			errs = append(errs, string(node.Id())+":"+err.Error())
+		if node.FreeSpace() <= 0 {
+			continue
 		}
+		totalWeights += node.FreeSpace()
+		candidates = append(candidates, node)
+		candidatesWeights = append(candidatesWeights, node.FreeSpace())
 	}
 	n.RUnlock()
-	if len(candidates) == 0 {
-		return nil, nil, errors.New("No matching data node found! \n" + strings.Join(errs, "\n"))
+	if len(candidates) < numberOfNodes {
+		glog.V(2).Infoln(n.Id(), "failed to pick", numberOfNodes, "from ", len(candidates), "node candidates")
+		return nil, nil, errors.New("No enough data node found!")
 	}
-	firstNode = candidates[rand.Intn(len(candidates))]
-	glog.V(2).Infoln(n.Id(), "picked main node:", firstNode.Id())
 
-	restNodes = make([]Node, numberOfNodes-1)
-	candidates = candidates[:0]
-	n.RLock()
-	for _, node := range n.children {
-		if node.Id() == firstNode.Id() {
-			continue
-		}
-		if node.FreeSpace() <= 0 {
-			continue
+	//pick nodes randomly by weights, the node picked earlier has higher final weights
+	sortedCandidates := make([]Node, 0, len(candidates))
+	for i:=0; i<len(candidates); i++  {
+		weightsInterval := rand.Int63n(totalWeights)
+		lastWeights := int64(0)
+		for k, weights := range candidatesWeights {
+			if (weightsInterval>=lastWeights) && (weightsInterval<lastWeights + weights) {
+				sortedCandidates = append(sortedCandidates, candidates[k])
+				candidatesWeights[k] = 0
+				totalWeights -= weights
+				break
+			}
+			lastWeights += weights
 		}
-		glog.V(2).Infoln("select rest node candidate:", node.Id())
-		candidates = append(candidates, node)
 	}
-	n.RUnlock()
-	glog.V(2).Infoln(n.Id(), "picking", numberOfNodes-1, "from rest", len(candidates), "node candidates")
-	ret := len(restNodes) == 0
-	for k, node := range candidates {
-		if k < len(restNodes) {
-			restNodes[k] = node
-			if k == len(restNodes)-1 {
-				ret = true
+
+	restNodes = make([]Node, 0, numberOfNodes-1)
+	ret := false
+	n.RLock()
+	for k, node := range sortedCandidates {
+		if err := filterFirstNodeFn(node); err == nil {
+			firstNode = node
+			if k >= numberOfNodes-1 {
+				restNodes = sortedCandidates[:numberOfNodes-1]
+			} else {
+				restNodes = append(restNodes, sortedCandidates[:k]...)
+				restNodes = append(restNodes, sortedCandidates[k+1:numberOfNodes]...)
 			}
+			ret = true
+			break
 		} else {
-			r := rand.Intn(k + 1)
-			if r < len(restNodes) {
-				restNodes[r] = node
-			}
+			errs = append(errs, string(node.Id())+":"+err.Error())
 		}
 	}
+	n.RUnlock()
 	if !ret {
-		glog.V(2).Infoln(n.Id(), "failed to pick", numberOfNodes-1, "from rest", len(candidates), "node candidates")
-		err = errors.New("No enough data node found!")
+		return nil, nil, errors.New("No matching data node found! \n" + strings.Join(errs, "\n"))
 	}
 	return
 }

weed/topology/volume_growth.go

@@ -112,7 +112,7 @@ func (vg *VolumeGrowth) findAndGrow(grpcDialOption grpc.DialOption, topo *Topolo
 func (vg *VolumeGrowth) findEmptySlotsForOneVolume(topo *Topology, option *VolumeGrowOption) (servers []*DataNode, err error) {
 	//find main datacenter and other data centers
 	rp := option.ReplicaPlacement
-	mainDataCenter, otherDataCenters, dc_err := topo.RandomlyPickNodes(rp.DiffDataCenterCount+1, func(node Node) error {
+	mainDataCenter, otherDataCenters, dc_err := topo.PickNodesByWeight(rp.DiffDataCenterCount+1, func(node Node) error {
 		if option.DataCenter != "" && node.IsDataCenter() && node.Id() != NodeId(option.DataCenter) {
 			return fmt.Errorf("Not matching preferred data center:%s", option.DataCenter)
 		}
@@ -144,7 +144,7 @@ func (vg *VolumeGrowth) findEmptySlotsForOneVolume(topo *Topology, option *Volum
 	}
 
 	//find main rack and other racks
-	mainRack, otherRacks, rackErr := mainDataCenter.(*DataCenter).RandomlyPickNodes(rp.DiffRackCount+1, func(node Node) error {
+	mainRack, otherRacks, rackErr := mainDataCenter.(*DataCenter).PickNodesByWeight(rp.DiffRackCount+1, func(node Node) error {
 		if option.Rack != "" && node.IsRack() && node.Id() != NodeId(option.Rack) {
 			return fmt.Errorf("Not matching preferred rack:%s", option.Rack)
 		}
@@ -171,7 +171,7 @@ func (vg *VolumeGrowth) findEmptySlotsForOneVolume(topo *Topology, option *Volum
 	}
 
 	//find main rack and other racks
-	mainServer, otherServers, serverErr := mainRack.(*Rack).RandomlyPickNodes(rp.SameRackCount+1, func(node Node) error {
+	mainServer, otherServers, serverErr := mainRack.(*Rack).PickNodesByWeight(rp.SameRackCount+1, func(node Node) error {
 		if option.DataNode != "" && node.IsDataNode() && node.Id() != NodeId(option.DataNode) {
 			return fmt.Errorf("Not matching preferred data node:%s", option.DataNode)
 		}

weed/topology/volume_growth_test.go

@@ -253,3 +253,90 @@ func TestReplication011(t *testing.T) {
 		fmt.Println("assigned node :", server.Id())
 	}
 }
+
+var topologyLayout3 = `
+{
+  "dc1":{
+    "rack1":{
+      "server111":{
+        "volumes":[],
+        "limit":2000
+      }
+    }
+  },
+  "dc2":{
+    "rack2":{
+      "server222":{
+        "volumes":[],
+        "limit":2000
+      }
+    }
+  },
+  "dc3":{
+    "rack3":{
+      "server333":{
+        "volumes":[],
+        "limit":1000
+      }
+    }
+  },
+  "dc4":{
+    "rack4":{
+      "server444":{
+        "volumes":[],
+        "limit":1000
+      }
+    }
+  },
+  "dc5":{
+    "rack5":{
+      "server555":{
+        "volumes":[],
+        "limit":500
+      }
+    }
+  },
+  "dc6":{
+    "rack6":{
+      "server666":{
+        "volumes":[],
+        "limit":500
+      }
+    }
+  }
+}
+`
+
+func TestFindEmptySlotsForOneVolumeScheduleByWeight(t *testing.T) {
+	topo := setup(topologyLayout3)
+	vg := NewDefaultVolumeGrowth()
+	rp, _ := super_block.NewReplicaPlacementFromString("100")
+	volumeGrowOption := &VolumeGrowOption{
+		Collection:       "Weight",
+		ReplicaPlacement: rp,
+		DataCenter:       "",
+		Rack:             "",
+		DataNode:         "",
+	}
+
+	distribution := map[NodeId]int{}
+	// assign 1000 volumes
+	for i:=0;i<1000 ;i++ {
+		servers, err := vg.findEmptySlotsForOneVolume(topo, volumeGrowOption)
+		if err != nil {
+			fmt.Println("finding empty slots error :", err)
+			t.Fail()
+		}
+		for _, server := range servers {
+			fmt.Println("assigned node :", server.Id())
+			if _, ok := distribution[server.id]; !ok {
+				distribution[server.id] = 0
+			}
+			distribution[server.id] += 1
+		}
+	}
+
+	for k, v := range distribution {
+		fmt.Println(k, "%s : %d", k, v)
+	}
+}