seaweedfs/weed/topology/volume_layout.go

297 lines
7.8 KiB
Go
Raw Normal View History

package topology
import (
2012-09-11 00:08:52 +00:00
"errors"
"fmt"
"math/rand"
"sync"
"time"
"github.com/chrislusf/seaweedfs/weed/glog"
"github.com/chrislusf/seaweedfs/weed/storage"
)
// mapping from volume to its locations, inverted from server to volume
type VolumeLayout struct {
rp *storage.ReplicaPlacement
ttl *storage.TTL
vid2location map[storage.VolumeId]*VolumeLocationList
writables []storage.VolumeId // transient array of writable volume id
readonlyVolumes map[storage.VolumeId]bool // transient set of readonly volumes
oversizedVolumes map[storage.VolumeId]bool // set of oversized volumes
volumeSizeLimit uint64
accessLock sync.RWMutex
}
type VolumeLayoutStats struct {
TotalSize uint64
UsedSize uint64
FileCount uint64
}
func NewVolumeLayout(rp *storage.ReplicaPlacement, ttl *storage.TTL, volumeSizeLimit uint64) *VolumeLayout {
return &VolumeLayout{
rp: rp,
ttl: ttl,
vid2location: make(map[storage.VolumeId]*VolumeLocationList),
writables: *new([]storage.VolumeId),
readonlyVolumes: make(map[storage.VolumeId]bool),
oversizedVolumes: make(map[storage.VolumeId]bool),
volumeSizeLimit: volumeSizeLimit,
}
}
func (vl *VolumeLayout) String() string {
return fmt.Sprintf("rp:%v, ttl:%v, vid2location:%v, writables:%v, volumeSizeLimit:%v", vl.rp, vl.ttl, vl.vid2location, vl.writables, vl.volumeSizeLimit)
}
func (vl *VolumeLayout) RegisterVolume(v *storage.VolumeInfo, dn *DataNode) {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
if _, ok := vl.vid2location[v.Id]; !ok {
2012-09-19 23:48:04 +00:00
vl.vid2location[v.Id] = NewVolumeLocationList()
}
vl.vid2location[v.Id].Set(dn)
2018-11-22 22:31:10 +00:00
glog.V(4).Infof("volume %d added to %s len %s copy %d", v.Id, dn.Id(), vl.vid2location[v.Id].Length(), v.ReplicaPlacement.GetCopyCount())
for _, dn := range vl.vid2location[v.Id].list {
if v_info, err := dn.GetVolumesById(v.Id); err == nil {
if v_info.ReadOnly {
glog.V(3).Infof("vid %d removed from writable", v.Id)
vl.removeFromWritable(v.Id)
vl.readonlyVolumes[v.Id] = true
return
} else {
delete(vl.readonlyVolumes, v.Id)
}
} else {
glog.V(3).Infof("vid %d removed from writable", v.Id)
vl.removeFromWritable(v.Id)
delete(vl.readonlyVolumes, v.Id)
return
}
}
if vl.vid2location[v.Id].Length() == vl.rp.GetCopyCount() && vl.isWritable(v) {
if _, ok := vl.oversizedVolumes[v.Id]; !ok {
vl.addToWritable(v.Id)
}
} else {
vl.rememberOversizedVolumne(v)
vl.removeFromWritable(v.Id)
}
}
func (vl *VolumeLayout) rememberOversizedVolumne(v *storage.VolumeInfo) {
if vl.isOversized(v) {
vl.oversizedVolumes[v.Id] = true
}
}
func (vl *VolumeLayout) UnRegisterVolume(v *storage.VolumeInfo, dn *DataNode) {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
vl.removeFromWritable(v.Id)
delete(vl.vid2location, v.Id)
}
2016-04-14 08:28:40 +00:00
func (vl *VolumeLayout) addToWritable(vid storage.VolumeId) {
for _, id := range vl.writables {
if vid == id {
return
2012-09-11 00:08:52 +00:00
}
}
vl.writables = append(vl.writables, vid)
}
func (vl *VolumeLayout) isOversized(v *storage.VolumeInfo) bool {
return uint64(v.Size) >= vl.volumeSizeLimit
}
func (vl *VolumeLayout) isWritable(v *storage.VolumeInfo) bool {
return !vl.isOversized(v) &&
v.Version == storage.CurrentVersion &&
!v.ReadOnly
}
func (vl *VolumeLayout) isEmpty() bool {
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
return len(vl.vid2location) == 0
}
func (vl *VolumeLayout) Lookup(vid storage.VolumeId) []*DataNode {
2016-04-14 08:28:40 +00:00
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
if location := vl.vid2location[vid]; location != nil {
return location.list
}
return nil
}
func (vl *VolumeLayout) ListVolumeServers() (nodes []*DataNode) {
2016-04-14 08:28:40 +00:00
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
for _, location := range vl.vid2location {
nodes = append(nodes, location.list...)
}
return
}
func (vl *VolumeLayout) PickForWrite(count uint64, option *VolumeGrowOption) (*storage.VolumeId, uint64, *VolumeLocationList, error) {
2016-04-14 08:28:40 +00:00
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
2012-09-11 00:08:52 +00:00
len_writers := len(vl.writables)
if len_writers <= 0 {
glog.V(0).Infoln("No more writable volumes!")
return nil, 0, nil, errors.New("No more writable volumes!")
2012-09-11 00:08:52 +00:00
}
if option.DataCenter == "" {
vid := vl.writables[rand.Intn(len_writers)]
locationList := vl.vid2location[vid]
if locationList != nil {
return &vid, count, locationList, nil
}
return nil, 0, nil, errors.New("Strangely vid " + vid.String() + " is on no machine!")
2015-03-10 07:20:31 +00:00
}
var vid storage.VolumeId
var locationList *VolumeLocationList
counter := 0
for _, v := range vl.writables {
volumeLocationList := vl.vid2location[v]
for _, dn := range volumeLocationList.list {
if dn.GetDataCenter().Id() == NodeId(option.DataCenter) {
if option.Rack != "" && dn.GetRack().Id() != NodeId(option.Rack) {
continue
}
if option.DataNode != "" && dn.Id() != NodeId(option.DataNode) {
continue
}
counter++
if rand.Intn(counter) < 1 {
vid, locationList = v, volumeLocationList
}
}
}
2012-09-11 00:08:52 +00:00
}
2015-03-10 07:20:31 +00:00
return &vid, count, locationList, nil
}
func (vl *VolumeLayout) GetActiveVolumeCount(option *VolumeGrowOption) int {
2016-04-14 08:28:40 +00:00
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
if option.DataCenter == "" {
return len(vl.writables)
}
counter := 0
for _, v := range vl.writables {
for _, dn := range vl.vid2location[v].list {
if dn.GetDataCenter().Id() == NodeId(option.DataCenter) {
if option.Rack != "" && dn.GetRack().Id() != NodeId(option.Rack) {
continue
}
if option.DataNode != "" && dn.Id() != NodeId(option.DataNode) {
continue
}
counter++
}
}
}
return counter
}
2012-09-19 23:48:04 +00:00
func (vl *VolumeLayout) removeFromWritable(vid storage.VolumeId) bool {
toDeleteIndex := -1
for k, id := range vl.writables {
if id == vid {
toDeleteIndex = k
break
2012-09-19 23:48:04 +00:00
}
}
if toDeleteIndex >= 0 {
glog.V(0).Infoln("Volume", vid, "becomes unwritable")
vl.writables = append(vl.writables[0:toDeleteIndex], vl.writables[toDeleteIndex+1:]...)
return true
}
2012-09-19 23:48:04 +00:00
return false
}
func (vl *VolumeLayout) setVolumeWritable(vid storage.VolumeId) bool {
for _, v := range vl.writables {
if v == vid {
return false
}
}
glog.V(0).Infoln("Volume", vid, "becomes writable")
2012-09-19 23:48:04 +00:00
vl.writables = append(vl.writables, vid)
return true
}
func (vl *VolumeLayout) SetVolumeUnavailable(dn *DataNode, vid storage.VolumeId) bool {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
if location, ok := vl.vid2location[vid]; ok {
if location.Remove(dn) {
if location.Length() < vl.rp.GetCopyCount() {
glog.V(0).Infoln("Volume", vid, "has", location.Length(), "replica, less than required", vl.rp.GetCopyCount())
return vl.removeFromWritable(vid)
}
}
}
return false
}
2012-09-19 23:48:04 +00:00
func (vl *VolumeLayout) SetVolumeAvailable(dn *DataNode, vid storage.VolumeId) bool {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
vl.vid2location[vid].Set(dn)
if vl.vid2location[vid].Length() >= vl.rp.GetCopyCount() {
return vl.setVolumeWritable(vid)
2012-09-19 23:48:04 +00:00
}
return false
}
2012-09-19 23:48:04 +00:00
func (vl *VolumeLayout) SetVolumeCapacityFull(vid storage.VolumeId) bool {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
// glog.V(0).Infoln("Volume", vid, "reaches full capacity.")
2012-09-19 23:48:04 +00:00
return vl.removeFromWritable(vid)
}
2013-11-12 10:19:06 +00:00
func (vl *VolumeLayout) ToMap() map[string]interface{} {
m := make(map[string]interface{})
m["replication"] = vl.rp.String()
m["ttl"] = vl.ttl.String()
m["writables"] = vl.writables
//m["locations"] = vl.vid2location
return m
}
func (vl *VolumeLayout) Stats() *VolumeLayoutStats {
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
ret := &VolumeLayoutStats{}
freshThreshold := time.Now().Unix() - 60
for vid, vll := range vl.vid2location {
size, fileCount := vll.Stats(vid, freshThreshold)
ret.FileCount += uint64(fileCount)
ret.UsedSize += size
if vl.readonlyVolumes[vid] {
ret.TotalSize += size
} else {
ret.TotalSize += vl.volumeSizeLimit
}
}
return ret
}