1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
|
package topology
import (
"fmt"
"math/rand"
"sync"
"github.com/chrislusf/seaweedfs/weed/storage/needle"
"github.com/chrislusf/seaweedfs/weed/storage/super_block"
"github.com/chrislusf/seaweedfs/weed/util"
"google.golang.org/grpc"
"github.com/chrislusf/seaweedfs/weed/glog"
"github.com/chrislusf/seaweedfs/weed/storage"
)
/*
This package is created to resolve these replica placement issues:
1. growth factor for each replica level, e.g., add 10 volumes for 1 copy, 20 volumes for 2 copies, 30 volumes for 3 copies
2. in time of tight storage, how to reduce replica level
3. optimizing for hot data on faster disk, cold data on cheaper storage,
4. volume allocation for each bucket
*/
type VolumeGrowOption struct {
Collection string
ReplicaPlacement *super_block.ReplicaPlacement
Ttl *needle.TTL
Prealloacte int64
DataCenter string
Rack string
DataNode string
MemoryMapMaxSizeMb uint32
}
type VolumeGrowth struct {
accessLock sync.Mutex
}
func (o *VolumeGrowOption) String() string {
return fmt.Sprintf("Collection:%s, ReplicaPlacement:%v, Ttl:%v, DataCenter:%s, Rack:%s, DataNode:%s", o.Collection, o.ReplicaPlacement, o.Ttl, o.DataCenter, o.Rack, o.DataNode)
}
func NewDefaultVolumeGrowth() *VolumeGrowth {
return &VolumeGrowth{}
}
// one replication type may need rp.GetCopyCount() actual volumes
// given copyCount, how many logical volumes to create
func (vg *VolumeGrowth) findVolumeCount(copyCount int) (count int) {
v := util.GetViper()
v.SetDefault("master.volume_growth.copy_1", 7)
v.SetDefault("master.volume_growth.copy_2", 6)
v.SetDefault("master.volume_growth.copy_3", 3)
v.SetDefault("master.volume_growth.copy_other", 1)
switch copyCount {
case 1:
count = v.GetInt("master.volume_growth.copy_1")
case 2:
count = v.GetInt("master.volume_growth.copy_2")
case 3:
count = v.GetInt("master.volume_growth.copy_3")
default:
count = v.GetInt("master.volume_growth.copy_other")
}
return
}
func (vg *VolumeGrowth) AutomaticGrowByType(option *VolumeGrowOption, grpcDialOption grpc.DialOption, topo *Topology, targetCount int) (count int, err error) {
if targetCount == 0 {
targetCount = vg.findVolumeCount(option.ReplicaPlacement.GetCopyCount())
}
count, err = vg.GrowByCountAndType(grpcDialOption, targetCount, option, topo)
if count > 0 && count%option.ReplicaPlacement.GetCopyCount() == 0 {
return count, nil
}
return count, err
}
func (vg *VolumeGrowth) GrowByCountAndType(grpcDialOption grpc.DialOption, targetCount int, option *VolumeGrowOption, topo *Topology) (counter int, err error) {
vg.accessLock.Lock()
defer vg.accessLock.Unlock()
for i := 0; i < targetCount; i++ {
if c, e := vg.findAndGrow(grpcDialOption, topo, option); e == nil {
counter += c
} else {
glog.V(0).Infof("create %d volume, created %d: %v", targetCount, counter, e)
return counter, e
}
}
return
}
func (vg *VolumeGrowth) findAndGrow(grpcDialOption grpc.DialOption, topo *Topology, option *VolumeGrowOption) (int, error) {
servers, e := vg.findEmptySlotsForOneVolume(topo, option)
if e != nil {
return 0, e
}
vid, raftErr := topo.NextVolumeId()
if raftErr != nil {
return 0, raftErr
}
err := vg.grow(grpcDialOption, topo, vid, option, servers...)
return len(servers), err
}
// 1. find the main data node
// 1.1 collect all data nodes that have 1 slots
// 2.2 collect all racks that have rp.SameRackCount+1
// 2.2 collect all data centers that have DiffRackCount+rp.SameRackCount+1
// 2. find rest data nodes
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.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)
}
if len(node.Children()) < rp.DiffRackCount+1 {
return fmt.Errorf("Only has %d racks, not enough for %d.", len(node.Children()), rp.DiffRackCount+1)
}
if node.FreeSpace() < int64(rp.DiffRackCount+rp.SameRackCount+1) {
return fmt.Errorf("Free:%d < Expected:%d", node.FreeSpace(), rp.DiffRackCount+rp.SameRackCount+1)
}
possibleRacksCount := 0
for _, rack := range node.Children() {
possibleDataNodesCount := 0
for _, n := range rack.Children() {
if n.FreeSpace() >= 1 {
possibleDataNodesCount++
}
}
if possibleDataNodesCount >= rp.SameRackCount+1 {
possibleRacksCount++
}
}
if possibleRacksCount < rp.DiffRackCount+1 {
return fmt.Errorf("Only has %d racks with more than %d free data nodes, not enough for %d.", possibleRacksCount, rp.SameRackCount+1, rp.DiffRackCount+1)
}
return nil
})
if dc_err != nil {
return nil, dc_err
}
//find main rack and other racks
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)
}
if node.FreeSpace() < int64(rp.SameRackCount+1) {
return fmt.Errorf("Free:%d < Expected:%d", node.FreeSpace(), rp.SameRackCount+1)
}
if len(node.Children()) < rp.SameRackCount+1 {
// a bit faster way to test free racks
return fmt.Errorf("Only has %d data nodes, not enough for %d.", len(node.Children()), rp.SameRackCount+1)
}
possibleDataNodesCount := 0
for _, n := range node.Children() {
if n.FreeSpace() >= 1 {
possibleDataNodesCount++
}
}
if possibleDataNodesCount < rp.SameRackCount+1 {
return fmt.Errorf("Only has %d data nodes with a slot, not enough for %d.", possibleDataNodesCount, rp.SameRackCount+1)
}
return nil
})
if rackErr != nil {
return nil, rackErr
}
//find main rack and other racks
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)
}
if node.FreeSpace() < 1 {
return fmt.Errorf("Free:%d < Expected:%d", node.FreeSpace(), 1)
}
return nil
})
if serverErr != nil {
return nil, serverErr
}
servers = append(servers, mainServer.(*DataNode))
for _, server := range otherServers {
servers = append(servers, server.(*DataNode))
}
for _, rack := range otherRacks {
r := rand.Int63n(rack.FreeSpace())
if server, e := rack.ReserveOneVolume(r); e == nil {
servers = append(servers, server)
} else {
return servers, e
}
}
for _, datacenter := range otherDataCenters {
r := rand.Int63n(datacenter.FreeSpace())
if server, e := datacenter.ReserveOneVolume(r); e == nil {
servers = append(servers, server)
} else {
return servers, e
}
}
return
}
func (vg *VolumeGrowth) grow(grpcDialOption grpc.DialOption, topo *Topology, vid needle.VolumeId, option *VolumeGrowOption, servers ...*DataNode) error {
for _, server := range servers {
if err := AllocateVolume(server, grpcDialOption, vid, option); err == nil {
vi := storage.VolumeInfo{
Id: vid,
Size: 0,
Collection: option.Collection,
ReplicaPlacement: option.ReplicaPlacement,
Ttl: option.Ttl,
Version: needle.CurrentVersion,
}
server.AddOrUpdateVolume(vi)
topo.RegisterVolumeLayout(vi, server)
glog.V(0).Infoln("Created Volume", vid, "on", server.NodeImpl.String())
} else {
glog.V(0).Infoln("Failed to assign volume", vid, "to", servers, "error", err)
return fmt.Errorf("Failed to assign %d: %v", vid, err)
}
}
return nil
}
|
