seaweedfs/weed/topology/volume_layout.go

437 lines
11 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"
2019-04-19 04:43:36 +00:00
"github.com/chrislusf/seaweedfs/weed/storage/needle"
2019-12-23 20:48:20 +00:00
"github.com/chrislusf/seaweedfs/weed/storage/super_block"
)
type copyState int
const (
noCopies copyState = 0 + iota
insufficientCopies
enoughCopies
)
type volumeState string
const (
readOnlyState volumeState = "ReadOnly"
oversizedState = "Oversized"
)
type stateIndicator func(copyState) bool
func ExistCopies() stateIndicator {
return func(state copyState) bool { return state != noCopies }
}
func NoCopies() stateIndicator {
return func(state copyState) bool { return state == noCopies }
}
type volumesBinaryState struct {
rp *super_block.ReplicaPlacement
name volumeState // the name for volume state (eg. "Readonly", "Oversized")
indicator stateIndicator // indicate whether the volumes should be marked as `name`
copyMap map[needle.VolumeId]*VolumeLocationList
}
func NewVolumesBinaryState(name volumeState, rp *super_block.ReplicaPlacement, indicator stateIndicator) *volumesBinaryState {
return &volumesBinaryState{
rp: rp,
name: name,
indicator: indicator,
copyMap: make(map[needle.VolumeId]*VolumeLocationList),
}
}
func (v *volumesBinaryState) Dump() (res []uint32) {
for vid, list := range v.copyMap {
if v.indicator(v.copyState(list)) {
res = append(res, uint32(vid))
}
}
return
}
func (v *volumesBinaryState) IsTrue(vid needle.VolumeId) bool {
list, _ := v.copyMap[vid]
return v.indicator(v.copyState(list))
}
func (v *volumesBinaryState) Add(vid needle.VolumeId, dn *DataNode) {
list, _ := v.copyMap[vid]
if list != nil {
list.Set(dn)
return
}
list = NewVolumeLocationList()
list.Set(dn)
v.copyMap[vid] = list
}
func (v *volumesBinaryState) Remove(vid needle.VolumeId, dn *DataNode) {
list, _ := v.copyMap[vid]
if list != nil {
list.Remove(dn)
if list.Length() == 0 {
delete(v.copyMap, vid)
}
}
}
func (v *volumesBinaryState) copyState(list *VolumeLocationList) copyState {
if list == nil {
return noCopies
}
if list.Length() < v.rp.GetCopyCount() {
return insufficientCopies
}
return enoughCopies
}
// mapping from volume to its locations, inverted from server to volume
type VolumeLayout struct {
2019-12-23 20:48:20 +00:00
rp *super_block.ReplicaPlacement
2019-04-19 04:43:36 +00:00
ttl *needle.TTL
vid2location map[needle.VolumeId]*VolumeLocationList
writables []needle.VolumeId // transient array of writable volume id
readonlyVolumes *volumesBinaryState // readonly volumes
oversizedVolumes *volumesBinaryState // oversized volumes
volumeSizeLimit uint64
replicationAsMin bool
accessLock sync.RWMutex
}
type VolumeLayoutStats struct {
TotalSize uint64
UsedSize uint64
FileCount uint64
}
func NewVolumeLayout(rp *super_block.ReplicaPlacement, ttl *needle.TTL, volumeSizeLimit uint64, replicationAsMin bool) *VolumeLayout {
return &VolumeLayout{
rp: rp,
ttl: ttl,
2019-04-19 04:43:36 +00:00
vid2location: make(map[needle.VolumeId]*VolumeLocationList),
writables: *new([]needle.VolumeId),
readonlyVolumes: NewVolumesBinaryState(readOnlyState, rp, ExistCopies()),
oversizedVolumes: NewVolumesBinaryState(oversizedState, rp, ExistCopies()),
volumeSizeLimit: volumeSizeLimit,
replicationAsMin: replicationAsMin,
}
}
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()
defer vl.ensureCorrectWritables(v.Id)
defer vl.rememberOversizedVolume(v, dn)
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)
2019-01-02 20:58:06 +00:00
// glog.V(4).Infof("volume %d added to %s len %d copy %d", v.Id, dn.Id(), vl.vid2location[v.Id].Length(), v.ReplicaPlacement.GetCopyCount())
for _, dn := range vl.vid2location[v.Id].list {
2019-01-17 01:17:19 +00:00
if vInfo, err := dn.GetVolumesById(v.Id); err == nil {
if vInfo.ReadOnly {
2019-04-21 06:53:37 +00:00
glog.V(1).Infof("vid %d removed from writable", v.Id)
vl.removeFromWritable(v.Id)
vl.readonlyVolumes.Add(v.Id, dn)
return
} else {
vl.readonlyVolumes.Remove(v.Id, dn)
}
} else {
2019-04-21 06:53:37 +00:00
glog.V(1).Infof("vid %d removed from writable", v.Id)
vl.removeFromWritable(v.Id)
vl.readonlyVolumes.Remove(v.Id, dn)
return
}
}
}
func (vl *VolumeLayout) rememberOversizedVolume(v *storage.VolumeInfo, dn *DataNode) {
if vl.isOversized(v) {
vl.oversizedVolumes.Add(v.Id, dn)
} else {
vl.oversizedVolumes.Remove(v.Id, dn)
}
}
func (vl *VolumeLayout) UnRegisterVolume(v *storage.VolumeInfo, dn *DataNode) {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
// remove from vid2location map
location, ok := vl.vid2location[v.Id]
if !ok {
return
}
if location.Remove(dn) {
vl.readonlyVolumes.Remove(v.Id, dn)
vl.oversizedVolumes.Remove(v.Id, dn)
vl.ensureCorrectWritables(v.Id)
if location.Length() == 0 {
delete(vl.vid2location, v.Id)
}
}
}
adding locking to avoid nil VolumeLocationList fix panic: runtime error: invalid memory address or nil pointer dereference Oct 22 00:53:44 bedb-master1 weed[8055]: [signal SIGSEGV: segmentation violation code=0x1 addr=0x8 pc=0x17658da] Oct 22 00:53:44 bedb-master1 weed[8055]: goroutine 310 [running]: Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*VolumeLocationList).Length(...) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/volume_location_list.go:35 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*VolumeLayout).enoughCopies(...) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/volume_layout.go:376 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*VolumeLayout).ensureCorrectWritables(0xc000111d50, 0xc000b55438) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/volume_layout.go:202 +0x5a Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*Topology).SyncDataNodeRegistration(0xc00042ac60, 0xc001454d30, 0x1, 0x1, 0xc0005fc000, 0xc00135de40, 0x4, 0xc00135de50, 0x10, 0x10d, ...) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/topology.go:224 +0x616 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/server.(*MasterServer).SendHeartbeat(0xc000162700, 0x23b97c0, 0xc000ae2c90, 0x0, 0x0) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/server/master_grpc_server.go:106 +0x325 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/pb/master_pb._Seaweed_SendHeartbeat_Handler(0x1f8e7c0, 0xc000162700, 0x23b0a60, 0xc00024b440, 0x3172c38, 0xc000ab7100) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/pb/master_pb/master.pb.go:4250 +0xad Oct 22 00:53:44 bedb-master1 weed[8055]: google.golang.org/grpc.(*Server).processStreamingRPC(0xc0001f31e0, 0x23bb800, 0xc000ac5500, 0xc000ab7100, 0xc0001fea80, 0x311fec0, 0x0, 0x0, 0x0) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:1329 +0xcd8 Oct 22 00:53:44 bedb-master1 weed[8055]: google.golang.org/grpc.(*Server).handleStream(0xc0001f31e0, 0x23bb800, 0xc000ac5500, 0xc000ab7100, 0x0) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:1409 +0xc5c Oct 22 00:53:44 bedb-master1 weed[8055]: google.golang.org/grpc.(*Server).serveStreams.func1.1(0xc0001ce8b0, 0xc0001f31e0, 0x23bb800, 0xc000ac5500, 0xc000ab7100) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:746 +0xa5 Oct 22 00:53:44 bedb-master1 weed[8055]: created by google.golang.org/grpc.(*Server).serveStreams.func1 Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:744 +0xa5 Oct 22 00:53:44 bedb-master1 systemd[1]: weedmaster.service: Main process exited, code=exited, status=2/INVALIDARGUMENT Oct 22 00:53:44 bedb-master1 systemd[1]: weedmaster.service: Failed with result 'exit-code'.
2020-10-22 06:15:48 +00:00
func (vl *VolumeLayout) EnsureCorrectWritables(v *storage.VolumeInfo) {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
vl.ensureCorrectWritables(v.Id)
adding locking to avoid nil VolumeLocationList fix panic: runtime error: invalid memory address or nil pointer dereference Oct 22 00:53:44 bedb-master1 weed[8055]: [signal SIGSEGV: segmentation violation code=0x1 addr=0x8 pc=0x17658da] Oct 22 00:53:44 bedb-master1 weed[8055]: goroutine 310 [running]: Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*VolumeLocationList).Length(...) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/volume_location_list.go:35 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*VolumeLayout).enoughCopies(...) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/volume_layout.go:376 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*VolumeLayout).ensureCorrectWritables(0xc000111d50, 0xc000b55438) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/volume_layout.go:202 +0x5a Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/topology.(*Topology).SyncDataNodeRegistration(0xc00042ac60, 0xc001454d30, 0x1, 0x1, 0xc0005fc000, 0xc00135de40, 0x4, 0xc00135de50, 0x10, 0x10d, ...) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/topology/topology.go:224 +0x616 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/server.(*MasterServer).SendHeartbeat(0xc000162700, 0x23b97c0, 0xc000ae2c90, 0x0, 0x0) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/server/master_grpc_server.go:106 +0x325 Oct 22 00:53:44 bedb-master1 weed[8055]: github.com/chrislusf/seaweedfs/weed/pb/master_pb._Seaweed_SendHeartbeat_Handler(0x1f8e7c0, 0xc000162700, 0x23b0a60, 0xc00024b440, 0x3172c38, 0xc000ab7100) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/seaweedfs/weed/pb/master_pb/master.pb.go:4250 +0xad Oct 22 00:53:44 bedb-master1 weed[8055]: google.golang.org/grpc.(*Server).processStreamingRPC(0xc0001f31e0, 0x23bb800, 0xc000ac5500, 0xc000ab7100, 0xc0001fea80, 0x311fec0, 0x0, 0x0, 0x0) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:1329 +0xcd8 Oct 22 00:53:44 bedb-master1 weed[8055]: google.golang.org/grpc.(*Server).handleStream(0xc0001f31e0, 0x23bb800, 0xc000ac5500, 0xc000ab7100, 0x0) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:1409 +0xc5c Oct 22 00:53:44 bedb-master1 weed[8055]: google.golang.org/grpc.(*Server).serveStreams.func1.1(0xc0001ce8b0, 0xc0001f31e0, 0x23bb800, 0xc000ac5500, 0xc000ab7100) Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:746 +0xa5 Oct 22 00:53:44 bedb-master1 weed[8055]: created by google.golang.org/grpc.(*Server).serveStreams.func1 Oct 22 00:53:44 bedb-master1 weed[8055]: #011/root/go/pkg/mod/google.golang.org/grpc@v1.29.1/server.go:744 +0xa5 Oct 22 00:53:44 bedb-master1 systemd[1]: weedmaster.service: Main process exited, code=exited, status=2/INVALIDARGUMENT Oct 22 00:53:44 bedb-master1 systemd[1]: weedmaster.service: Failed with result 'exit-code'.
2020-10-22 06:15:48 +00:00
}
func (vl *VolumeLayout) ensureCorrectWritables(vid needle.VolumeId) {
if vl.enoughCopies(vid) && vl.isAllWritable(vid) {
if !vl.oversizedVolumes.IsTrue(vid) {
vl.setVolumeWritable(vid)
}
} else {
vl.removeFromWritable(vid)
}
}
func (vl *VolumeLayout) isAllWritable(vid needle.VolumeId) bool {
for _, dn := range vl.vid2location[vid].list {
if v, found := dn.volumes[vid]; found {
if v.ReadOnly {
return false
}
}
}
return true
}
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) &&
2019-04-19 04:43:36 +00:00
v.Version == needle.CurrentVersion &&
!v.ReadOnly
}
func (vl *VolumeLayout) isEmpty() bool {
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
return len(vl.vid2location) == 0
}
2019-04-19 04:43:36 +00:00
func (vl *VolumeLayout) Lookup(vid needle.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
}
2019-04-19 04:43:36 +00:00
func (vl *VolumeLayout) PickForWrite(count uint64, option *VolumeGrowOption) (*needle.VolumeId, uint64, *VolumeLocationList, error) {
2016-04-14 08:28:40 +00:00
vl.accessLock.RLock()
defer vl.accessLock.RUnlock()
2019-01-17 01:17:19 +00:00
lenWriters := len(vl.writables)
if lenWriters <= 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 == "" {
2019-01-17 01:17:19 +00:00
vid := vl.writables[rand.Intn(lenWriters)]
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
}
2019-04-19 04:43:36 +00:00
var vid needle.VolumeId
2015-03-10 07:20:31 +00:00
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
}
2019-04-19 04:43:36 +00:00
func (vl *VolumeLayout) removeFromWritable(vid needle.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
}
2019-04-19 04:43:36 +00:00
func (vl *VolumeLayout) setVolumeWritable(vid needle.VolumeId) bool {
2012-09-19 23:48:04 +00:00
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
}
2019-04-19 04:43:36 +00:00
func (vl *VolumeLayout) SetVolumeUnavailable(dn *DataNode, vid needle.VolumeId) bool {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
if location, ok := vl.vid2location[vid]; ok {
if location.Remove(dn) {
vl.readonlyVolumes.Remove(vid, dn)
vl.oversizedVolumes.Remove(vid, 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
}
func (vl *VolumeLayout) SetVolumeAvailable(dn *DataNode, vid needle.VolumeId, isReadOnly bool) bool {
vl.accessLock.Lock()
defer vl.accessLock.Unlock()
2020-03-13 22:51:38 +00:00
vInfo, err := dn.GetVolumesById(vid)
if err != nil {
return false
}
vl.vid2location[vid].Set(dn)
if vInfo.ReadOnly || isReadOnly {
return false
}
if vl.enoughCopies(vid) {
return vl.setVolumeWritable(vid)
2012-09-19 23:48:04 +00:00
}
return false
}
func (vl *VolumeLayout) enoughCopies(vid needle.VolumeId) bool {
locations := vl.vid2location[vid].Length()
desired := vl.rp.GetCopyCount()
return locations == desired || (vl.replicationAsMin && locations > desired)
}
2019-04-19 04:43:36 +00:00
func (vl *VolumeLayout) SetVolumeCapacityFull(vid needle.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.IsTrue(vid) {
ret.TotalSize += size
} else {
ret.TotalSize += vl.volumeSizeLimit
}
}
return ret
}