333 lines
10 KiB
Go
333 lines
10 KiB
Go
/*
|
||
* Copyright 2020 The Dragonfly Authors
|
||
*
|
||
* Licensed under the Apache License, Version 2.0 (the "License");
|
||
* you may not use this file except in compliance with the License.
|
||
* You may obtain a copy of the License at
|
||
*
|
||
* http://www.apache.org/licenses/LICENSE-2.0
|
||
*
|
||
* Unless required by applicable law or agreed to in writing, software
|
||
* distributed under the License is distributed on an "AS IS" BASIS,
|
||
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||
* See the License for the specific language governing permissions and
|
||
* limitations under the License.
|
||
*/
|
||
|
||
package core
|
||
|
||
import (
|
||
"context"
|
||
"time"
|
||
|
||
"d7y.io/dragonfly/v2/internal/dfcodes"
|
||
logger "d7y.io/dragonfly/v2/internal/dflog"
|
||
"d7y.io/dragonfly/v2/pkg/rpc/base"
|
||
schedulerRPC "d7y.io/dragonfly/v2/pkg/rpc/scheduler"
|
||
"d7y.io/dragonfly/v2/pkg/structure/sortedlist"
|
||
"d7y.io/dragonfly/v2/pkg/synclock"
|
||
"d7y.io/dragonfly/v2/scheduler/core/scheduler"
|
||
"d7y.io/dragonfly/v2/scheduler/supervisor"
|
||
"go.opentelemetry.io/otel/trace"
|
||
"k8s.io/client-go/util/workqueue"
|
||
)
|
||
|
||
type event interface {
|
||
hashKey() string
|
||
apply(s *state)
|
||
}
|
||
|
||
type state struct {
|
||
sched scheduler.Scheduler
|
||
peerManager supervisor.PeerMgr
|
||
cdnManager supervisor.CDNMgr
|
||
waitScheduleParentPeerQueue workqueue.DelayingInterface
|
||
}
|
||
|
||
func newState(sched scheduler.Scheduler, peerManager supervisor.PeerMgr, cdnManager supervisor.CDNMgr, wsdq workqueue.DelayingInterface) *state {
|
||
return &state{
|
||
sched: sched,
|
||
peerManager: peerManager,
|
||
cdnManager: cdnManager,
|
||
waitScheduleParentPeerQueue: wsdq,
|
||
}
|
||
}
|
||
|
||
type reScheduleParentEvent struct {
|
||
peer *supervisor.Peer
|
||
}
|
||
|
||
var _ event = reScheduleParentEvent{}
|
||
|
||
func (e reScheduleParentEvent) apply(s *state) {
|
||
reScheduleParent(e.peer, s)
|
||
}
|
||
|
||
func (e reScheduleParentEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
type startReportPieceResultEvent struct {
|
||
ctx context.Context
|
||
peer *supervisor.Peer
|
||
}
|
||
|
||
var _ event = startReportPieceResultEvent{}
|
||
|
||
func (e startReportPieceResultEvent) apply(s *state) {
|
||
span := trace.SpanFromContext(e.ctx)
|
||
if e.peer.GetParent() != nil {
|
||
span.AddEvent("no parent")
|
||
logger.WithTaskAndPeerID(e.peer.Task.TaskID,
|
||
e.peer.PeerID).Warnf("startReportPieceResultEvent: no need schedule parent because peer already had parent %s", e.peer.GetParent().PeerID)
|
||
return
|
||
}
|
||
parent, candidates, hasParent := s.sched.ScheduleParent(e.peer)
|
||
span.AddEvent("parent")
|
||
if !hasParent {
|
||
logger.WithTaskAndPeerID(e.peer.Task.TaskID, e.peer.PeerID).Warnf("startReportPieceResultEvent: there is no available parent，reschedule it in one second")
|
||
s.waitScheduleParentPeerQueue.AddAfter(e.peer, time.Second)
|
||
return
|
||
}
|
||
e.peer.SendSchedulePacket(constructSuccessPeerPacket(e.peer, parent, candidates))
|
||
}
|
||
|
||
func (e startReportPieceResultEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
type peerDownloadPieceSuccessEvent struct {
|
||
ctx context.Context
|
||
peer *supervisor.Peer
|
||
pr *schedulerRPC.PieceResult
|
||
}
|
||
|
||
var _ event = peerDownloadPieceSuccessEvent{}
|
||
|
||
func (e peerDownloadPieceSuccessEvent) apply(s *state) {
|
||
span := trace.SpanFromContext(e.ctx)
|
||
span.AddEvent("piece success")
|
||
e.peer.AddPieceInfo(e.pr.FinishedCount, int(e.pr.EndTime-e.pr.BeginTime))
|
||
oldParent := e.peer.GetParent()
|
||
var candidates []*supervisor.Peer
|
||
parentPeer, ok := s.peerManager.Get(e.pr.DstPid)
|
||
if !ok || parentPeer.IsLeave() {
|
||
e.peer.ReplaceParent(nil)
|
||
var hasParent bool
|
||
parentPeer, candidates, hasParent = s.sched.ScheduleParent(e.peer)
|
||
if !hasParent {
|
||
logger.WithTaskAndPeerID(e.peer.Task.TaskID, e.peer.PeerID).Warnf("peerDownloadPieceSuccessEvent: there is no available parent, reschedule it in one second")
|
||
s.waitScheduleParentPeerQueue.AddAfter(e.peer, time.Second)
|
||
return
|
||
}
|
||
}
|
||
parentPeer.Touch()
|
||
if oldParent != nil {
|
||
candidates = append(candidates, oldParent)
|
||
}
|
||
// TODO if parentPeer is equal with oldParent, need schedule again ?
|
||
e.peer.SendSchedulePacket(constructSuccessPeerPacket(e.peer, parentPeer, candidates))
|
||
return
|
||
}
|
||
|
||
func (e peerDownloadPieceSuccessEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
type peerDownloadPieceFailEvent struct {
|
||
peer *supervisor.Peer
|
||
pr *schedulerRPC.PieceResult
|
||
}
|
||
|
||
var _ event = peerDownloadPieceFailEvent{}
|
||
|
||
func (e peerDownloadPieceFailEvent) apply(s *state) {
|
||
switch e.pr.Code {
|
||
case dfcodes.PeerTaskNotFound, dfcodes.ClientPieceRequestFail, dfcodes.ClientPieceDownloadFail:
|
||
// TODO PeerTaskNotFound remove dest peer task, ClientPieceDownloadFail add blank list
|
||
reScheduleParent(e.peer, s)
|
||
return
|
||
case dfcodes.CdnTaskNotFound, dfcodes.CdnError, dfcodes.CdnTaskRegistryFail, dfcodes.CdnTaskDownloadFail:
|
||
go func(task *supervisor.Task) {
|
||
// TODO
|
||
synclock.Lock(task.TaskID, false)
|
||
defer synclock.UnLock(task.TaskID, false)
|
||
task.SetStatus(supervisor.TaskStatusRunning)
|
||
if cdnPeer, err := s.cdnManager.StartSeedTask(context.Background(), task); err != nil {
|
||
logger.Errorf("start seed task fail: %v", err)
|
||
task.SetStatus(supervisor.TaskStatusFailed)
|
||
handleSeedTaskFail(task)
|
||
} else {
|
||
logger.Debugf("===== successfully obtain seeds from cdn, task: %+v =====", e.peer.Task)
|
||
children := s.sched.ScheduleChildren(cdnPeer)
|
||
for _, child := range children {
|
||
child.SendSchedulePacket(constructSuccessPeerPacket(child, cdnPeer, nil))
|
||
}
|
||
}
|
||
}(e.peer.Task)
|
||
default:
|
||
reScheduleParent(e.peer, s)
|
||
return
|
||
}
|
||
}
|
||
func (e peerDownloadPieceFailEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
type taskSeedFailEvent struct {
|
||
task *supervisor.Task
|
||
}
|
||
|
||
var _ event = taskSeedFailEvent{}
|
||
|
||
func (e taskSeedFailEvent) apply(s *state) {
|
||
handleSeedTaskFail(e.task)
|
||
}
|
||
|
||
func (e taskSeedFailEvent) hashKey() string {
|
||
return e.task.TaskID
|
||
}
|
||
|
||
type peerDownloadSuccessEvent struct {
|
||
peer *supervisor.Peer
|
||
peerResult *schedulerRPC.PeerResult
|
||
}
|
||
|
||
var _ event = peerDownloadSuccessEvent{}
|
||
|
||
func (e peerDownloadSuccessEvent) apply(s *state) {
|
||
e.peer.SetStatus(supervisor.PeerStatusSuccess)
|
||
removePeerFromCurrentTree(e.peer, s)
|
||
children := s.sched.ScheduleChildren(e.peer)
|
||
for _, child := range children {
|
||
child.SendSchedulePacket(constructSuccessPeerPacket(child, e.peer, nil))
|
||
}
|
||
e.peer.UnBindSendChannel()
|
||
}
|
||
|
||
func (e peerDownloadSuccessEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
type peerDownloadFailEvent struct {
|
||
peer *supervisor.Peer
|
||
peerResult *schedulerRPC.PeerResult
|
||
}
|
||
|
||
var _ event = peerDownloadFailEvent{}
|
||
|
||
func (e peerDownloadFailEvent) apply(s *state) {
|
||
e.peer.SetStatus(supervisor.PeerStatusFail)
|
||
removePeerFromCurrentTree(e.peer, s)
|
||
e.peer.GetChildren().Range(func(key, value interface{}) bool {
|
||
child := (value).(*supervisor.Peer)
|
||
parent, candidates, hasParent := s.sched.ScheduleParent(child)
|
||
if !hasParent {
|
||
logger.WithTaskAndPeerID(child.Task.TaskID, child.PeerID).Warnf("peerDownloadFailEvent: there is no available parent, reschedule it in one second")
|
||
s.waitScheduleParentPeerQueue.AddAfter(e.peer, time.Second)
|
||
return true
|
||
}
|
||
child.SendSchedulePacket(constructSuccessPeerPacket(child, parent, candidates))
|
||
return true
|
||
})
|
||
}
|
||
|
||
func (e peerDownloadFailEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
type peerLeaveEvent struct {
|
||
ctx context.Context
|
||
peer *supervisor.Peer
|
||
}
|
||
|
||
var _ event = peerLeaveEvent{}
|
||
|
||
func (e peerLeaveEvent) apply(s *state) {
|
||
e.peer.MarkLeave()
|
||
removePeerFromCurrentTree(e.peer, s)
|
||
e.peer.GetChildren().Range(func(key, value interface{}) bool {
|
||
child := value.(*supervisor.Peer)
|
||
parent, candidates, hasParent := s.sched.ScheduleParent(child)
|
||
if !hasParent {
|
||
logger.WithTaskAndPeerID(child.Task.TaskID, child.PeerID).Warnf("handlePeerLeave: there is no available parent，reschedule it in one second")
|
||
s.waitScheduleParentPeerQueue.AddAfter(child, time.Second)
|
||
return true
|
||
}
|
||
child.SendSchedulePacket(constructSuccessPeerPacket(child, parent, candidates))
|
||
return true
|
||
})
|
||
s.peerManager.Delete(e.peer.PeerID)
|
||
}
|
||
|
||
func (e peerLeaveEvent) hashKey() string {
|
||
return e.peer.Task.TaskID
|
||
}
|
||
|
||
func constructSuccessPeerPacket(peer *supervisor.Peer, parent *supervisor.Peer, candidates []*supervisor.Peer) *schedulerRPC.PeerPacket {
|
||
mainPeer := &schedulerRPC.PeerPacket_DestPeer{
|
||
Ip: parent.Host.IP,
|
||
RpcPort: parent.Host.RPCPort,
|
||
PeerId: parent.PeerID,
|
||
}
|
||
var stealPeers []*schedulerRPC.PeerPacket_DestPeer
|
||
for _, candidate := range candidates {
|
||
stealPeers = append(stealPeers, &schedulerRPC.PeerPacket_DestPeer{
|
||
Ip: candidate.Host.IP,
|
||
RpcPort: candidate.Host.RPCPort,
|
||
PeerId: candidate.PeerID,
|
||
})
|
||
}
|
||
peerPacket := &schedulerRPC.PeerPacket{
|
||
TaskId: peer.Task.TaskID,
|
||
SrcPid: peer.PeerID,
|
||
ParallelCount: 1,
|
||
MainPeer: mainPeer,
|
||
StealPeers: stealPeers,
|
||
Code: dfcodes.Success,
|
||
}
|
||
logger.Debugf("send peerPacket %+v to peer %s", peerPacket, peer.PeerID)
|
||
return peerPacket
|
||
}
|
||
|
||
func constructFailPeerPacket(peer *supervisor.Peer, errCode base.Code) *schedulerRPC.PeerPacket {
|
||
return &schedulerRPC.PeerPacket{
|
||
TaskId: peer.Task.TaskID,
|
||
SrcPid: peer.PeerID,
|
||
Code: errCode,
|
||
}
|
||
}
|
||
|
||
func reScheduleParent(peer *supervisor.Peer, s *state) {
|
||
parent, candidates, hasParent := s.sched.ScheduleParent(peer)
|
||
if !hasParent {
|
||
logger.Errorf("handleReplaceParent: failed to schedule parent to peer %s, reschedule it in one second", peer.PeerID)
|
||
//peer.PacketChan <- constructFailPeerPacket(peer, dfcodes.SchedWithoutParentPeer)
|
||
s.waitScheduleParentPeerQueue.AddAfter(peer, time.Second)
|
||
return
|
||
}
|
||
peer.SendSchedulePacket(constructSuccessPeerPacket(peer, parent, candidates))
|
||
}
|
||
|
||
func handleSeedTaskFail(task *supervisor.Task) {
|
||
if task.IsFail() {
|
||
task.ListPeers().Range(func(data sortedlist.Item) bool {
|
||
peer := data.(*supervisor.Peer)
|
||
peer.SendSchedulePacket(constructFailPeerPacket(peer, dfcodes.SchedNeedBackSource))
|
||
return true
|
||
})
|
||
}
|
||
}
|
||
|
||
func removePeerFromCurrentTree(peer *supervisor.Peer, s *state) {
|
||
parent := peer.GetParent()
|
||
peer.ReplaceParent(nil)
|
||
// parent frees up upload resources
|
||
if parent != nil {
|
||
children := s.sched.ScheduleChildren(parent)
|
||
for _, child := range children {
|
||
child.SendSchedulePacket(constructSuccessPeerPacket(child, peer, nil))
|
||
}
|
||
}
|
||
}
|