c2003ed63b
This change - implements concurrent LES request serving even for a single peer. - replaces the request cost estimation method with a cost table based on benchmarks which gives much more consistent results. Until now the allowed number of light peers was just a guess which probably contributed a lot to the fluctuating quality of available service. Everything related to request cost is implemented in a single object, the 'cost tracker'. It uses a fixed cost table with a global 'correction factor'. Benchmark code is included and can be run at any time to adapt costs to low-level implementation changes. - reimplements flowcontrol.ClientManager in a cleaner and more efficient way, with added capabilities: There is now control over bandwidth, which allows using the flow control parameters for client prioritization. Target utilization over 100 percent is now supported to model concurrent request processing. Total serving bandwidth is reduced during block processing to prevent database contention. - implements an RPC API for the LES servers allowing server operators to assign priority bandwidth to certain clients and change prioritized status even while the client is connected. The new API is meant for cases where server operators charge for LES using an off-protocol mechanism. - adds a unit test for the new client manager. - adds an end-to-end test using the network simulator that tests bandwidth control functions through the new API.
413 lines
11 KiB
Go
413 lines
11 KiB
Go
// Copyright 2017 The go-ethereum Authors
|
|
// This file is part of the go-ethereum library.
|
|
//
|
|
// The go-ethereum library is free software: you can redistribute it and/or modify
|
|
// it under the terms of the GNU Lesser General Public License as published by
|
|
// the Free Software Foundation, either version 3 of the License, or
|
|
// (at your option) any later version.
|
|
//
|
|
// The go-ethereum library is distributed in the hope that it will be useful,
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
// GNU Lesser General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU Lesser General Public License
|
|
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
package les
|
|
|
|
import (
|
|
"context"
|
|
"crypto/rand"
|
|
"encoding/binary"
|
|
"fmt"
|
|
"sync"
|
|
"time"
|
|
|
|
"github.com/ethereum/go-ethereum/common/mclock"
|
|
"github.com/ethereum/go-ethereum/light"
|
|
)
|
|
|
|
var (
|
|
retryQueue = time.Millisecond * 100
|
|
softRequestTimeout = time.Millisecond * 500
|
|
hardRequestTimeout = time.Second * 10
|
|
)
|
|
|
|
// retrieveManager is a layer on top of requestDistributor which takes care of
|
|
// matching replies by request ID and handles timeouts and resends if necessary.
|
|
type retrieveManager struct {
|
|
dist *requestDistributor
|
|
peers *peerSet
|
|
serverPool peerSelector
|
|
|
|
lock sync.RWMutex
|
|
sentReqs map[uint64]*sentReq
|
|
}
|
|
|
|
// validatorFunc is a function that processes a reply message
|
|
type validatorFunc func(distPeer, *Msg) error
|
|
|
|
// peerSelector receives feedback info about response times and timeouts
|
|
type peerSelector interface {
|
|
adjustResponseTime(*poolEntry, time.Duration, bool)
|
|
}
|
|
|
|
// sentReq represents a request sent and tracked by retrieveManager
|
|
type sentReq struct {
|
|
rm *retrieveManager
|
|
req *distReq
|
|
id uint64
|
|
validate validatorFunc
|
|
|
|
eventsCh chan reqPeerEvent
|
|
stopCh chan struct{}
|
|
stopped bool
|
|
err error
|
|
|
|
lock sync.RWMutex // protect access to sentTo map
|
|
sentTo map[distPeer]sentReqToPeer
|
|
|
|
lastReqQueued bool // last request has been queued but not sent
|
|
lastReqSentTo distPeer // if not nil then last request has been sent to given peer but not timed out
|
|
reqSrtoCount int // number of requests that reached soft (but not hard) timeout
|
|
}
|
|
|
|
// sentReqToPeer notifies the request-from-peer goroutine (tryRequest) about a response
|
|
// delivered by the given peer. Only one delivery is allowed per request per peer,
|
|
// after which delivered is set to true, the validity of the response is sent on the
|
|
// valid channel and no more responses are accepted.
|
|
type sentReqToPeer struct {
|
|
delivered bool
|
|
valid chan bool
|
|
}
|
|
|
|
// reqPeerEvent is sent by the request-from-peer goroutine (tryRequest) to the
|
|
// request state machine (retrieveLoop) through the eventsCh channel.
|
|
type reqPeerEvent struct {
|
|
event int
|
|
peer distPeer
|
|
}
|
|
|
|
const (
|
|
rpSent = iota // if peer == nil, not sent (no suitable peers)
|
|
rpSoftTimeout
|
|
rpHardTimeout
|
|
rpDeliveredValid
|
|
rpDeliveredInvalid
|
|
)
|
|
|
|
// newRetrieveManager creates the retrieve manager
|
|
func newRetrieveManager(peers *peerSet, dist *requestDistributor, serverPool peerSelector) *retrieveManager {
|
|
return &retrieveManager{
|
|
peers: peers,
|
|
dist: dist,
|
|
serverPool: serverPool,
|
|
sentReqs: make(map[uint64]*sentReq),
|
|
}
|
|
}
|
|
|
|
// retrieve sends a request (to multiple peers if necessary) and waits for an answer
|
|
// that is delivered through the deliver function and successfully validated by the
|
|
// validator callback. It returns when a valid answer is delivered or the context is
|
|
// cancelled.
|
|
func (rm *retrieveManager) retrieve(ctx context.Context, reqID uint64, req *distReq, val validatorFunc, shutdown chan struct{}) error {
|
|
sentReq := rm.sendReq(reqID, req, val)
|
|
select {
|
|
case <-sentReq.stopCh:
|
|
case <-ctx.Done():
|
|
sentReq.stop(ctx.Err())
|
|
case <-shutdown:
|
|
sentReq.stop(fmt.Errorf("Client is shutting down"))
|
|
}
|
|
return sentReq.getError()
|
|
}
|
|
|
|
// sendReq starts a process that keeps trying to retrieve a valid answer for a
|
|
// request from any suitable peers until stopped or succeeded.
|
|
func (rm *retrieveManager) sendReq(reqID uint64, req *distReq, val validatorFunc) *sentReq {
|
|
r := &sentReq{
|
|
rm: rm,
|
|
req: req,
|
|
id: reqID,
|
|
sentTo: make(map[distPeer]sentReqToPeer),
|
|
stopCh: make(chan struct{}),
|
|
eventsCh: make(chan reqPeerEvent, 10),
|
|
validate: val,
|
|
}
|
|
|
|
canSend := req.canSend
|
|
req.canSend = func(p distPeer) bool {
|
|
// add an extra check to canSend: the request has not been sent to the same peer before
|
|
r.lock.RLock()
|
|
_, sent := r.sentTo[p]
|
|
r.lock.RUnlock()
|
|
return !sent && canSend(p)
|
|
}
|
|
|
|
request := req.request
|
|
req.request = func(p distPeer) func() {
|
|
// before actually sending the request, put an entry into the sentTo map
|
|
r.lock.Lock()
|
|
r.sentTo[p] = sentReqToPeer{false, make(chan bool, 1)}
|
|
r.lock.Unlock()
|
|
return request(p)
|
|
}
|
|
rm.lock.Lock()
|
|
rm.sentReqs[reqID] = r
|
|
rm.lock.Unlock()
|
|
|
|
go r.retrieveLoop()
|
|
return r
|
|
}
|
|
|
|
// deliver is called by the LES protocol manager to deliver reply messages to waiting requests
|
|
func (rm *retrieveManager) deliver(peer distPeer, msg *Msg) error {
|
|
rm.lock.RLock()
|
|
req, ok := rm.sentReqs[msg.ReqID]
|
|
rm.lock.RUnlock()
|
|
|
|
if ok {
|
|
return req.deliver(peer, msg)
|
|
}
|
|
return errResp(ErrUnexpectedResponse, "reqID = %v", msg.ReqID)
|
|
}
|
|
|
|
// reqStateFn represents a state of the retrieve loop state machine
|
|
type reqStateFn func() reqStateFn
|
|
|
|
// retrieveLoop is the retrieval state machine event loop
|
|
func (r *sentReq) retrieveLoop() {
|
|
go r.tryRequest()
|
|
r.lastReqQueued = true
|
|
state := r.stateRequesting
|
|
|
|
for state != nil {
|
|
state = state()
|
|
}
|
|
|
|
r.rm.lock.Lock()
|
|
delete(r.rm.sentReqs, r.id)
|
|
r.rm.lock.Unlock()
|
|
}
|
|
|
|
// stateRequesting: a request has been queued or sent recently; when it reaches soft timeout,
|
|
// a new request is sent to a new peer
|
|
func (r *sentReq) stateRequesting() reqStateFn {
|
|
select {
|
|
case ev := <-r.eventsCh:
|
|
r.update(ev)
|
|
switch ev.event {
|
|
case rpSent:
|
|
if ev.peer == nil {
|
|
// request send failed, no more suitable peers
|
|
if r.waiting() {
|
|
// we are already waiting for sent requests which may succeed so keep waiting
|
|
return r.stateNoMorePeers
|
|
}
|
|
// nothing to wait for, no more peers to ask, return with error
|
|
r.stop(light.ErrNoPeers)
|
|
// no need to go to stopped state because waiting() already returned false
|
|
return nil
|
|
}
|
|
case rpSoftTimeout:
|
|
// last request timed out, try asking a new peer
|
|
go r.tryRequest()
|
|
r.lastReqQueued = true
|
|
return r.stateRequesting
|
|
case rpDeliveredInvalid:
|
|
// if it was the last sent request (set to nil by update) then start a new one
|
|
if !r.lastReqQueued && r.lastReqSentTo == nil {
|
|
go r.tryRequest()
|
|
r.lastReqQueued = true
|
|
}
|
|
return r.stateRequesting
|
|
case rpDeliveredValid:
|
|
r.stop(nil)
|
|
return r.stateStopped
|
|
}
|
|
return r.stateRequesting
|
|
case <-r.stopCh:
|
|
return r.stateStopped
|
|
}
|
|
}
|
|
|
|
// stateNoMorePeers: could not send more requests because no suitable peers are available.
|
|
// Peers may become suitable for a certain request later or new peers may appear so we
|
|
// keep trying.
|
|
func (r *sentReq) stateNoMorePeers() reqStateFn {
|
|
select {
|
|
case <-time.After(retryQueue):
|
|
go r.tryRequest()
|
|
r.lastReqQueued = true
|
|
return r.stateRequesting
|
|
case ev := <-r.eventsCh:
|
|
r.update(ev)
|
|
if ev.event == rpDeliveredValid {
|
|
r.stop(nil)
|
|
return r.stateStopped
|
|
}
|
|
if r.waiting() {
|
|
return r.stateNoMorePeers
|
|
}
|
|
r.stop(light.ErrNoPeers)
|
|
return nil
|
|
case <-r.stopCh:
|
|
return r.stateStopped
|
|
}
|
|
}
|
|
|
|
// stateStopped: request succeeded or cancelled, just waiting for some peers
|
|
// to either answer or time out hard
|
|
func (r *sentReq) stateStopped() reqStateFn {
|
|
for r.waiting() {
|
|
r.update(<-r.eventsCh)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// update updates the queued/sent flags and timed out peers counter according to the event
|
|
func (r *sentReq) update(ev reqPeerEvent) {
|
|
switch ev.event {
|
|
case rpSent:
|
|
r.lastReqQueued = false
|
|
r.lastReqSentTo = ev.peer
|
|
case rpSoftTimeout:
|
|
r.lastReqSentTo = nil
|
|
r.reqSrtoCount++
|
|
case rpHardTimeout:
|
|
r.reqSrtoCount--
|
|
case rpDeliveredValid, rpDeliveredInvalid:
|
|
if ev.peer == r.lastReqSentTo {
|
|
r.lastReqSentTo = nil
|
|
} else {
|
|
r.reqSrtoCount--
|
|
}
|
|
}
|
|
}
|
|
|
|
// waiting returns true if the retrieval mechanism is waiting for an answer from
|
|
// any peer
|
|
func (r *sentReq) waiting() bool {
|
|
return r.lastReqQueued || r.lastReqSentTo != nil || r.reqSrtoCount > 0
|
|
}
|
|
|
|
// tryRequest tries to send the request to a new peer and waits for it to either
|
|
// succeed or time out if it has been sent. It also sends the appropriate reqPeerEvent
|
|
// messages to the request's event channel.
|
|
func (r *sentReq) tryRequest() {
|
|
sent := r.rm.dist.queue(r.req)
|
|
var p distPeer
|
|
select {
|
|
case p = <-sent:
|
|
case <-r.stopCh:
|
|
if r.rm.dist.cancel(r.req) {
|
|
p = nil
|
|
} else {
|
|
p = <-sent
|
|
}
|
|
}
|
|
|
|
r.eventsCh <- reqPeerEvent{rpSent, p}
|
|
if p == nil {
|
|
return
|
|
}
|
|
|
|
reqSent := mclock.Now()
|
|
srto, hrto := false, false
|
|
|
|
r.lock.RLock()
|
|
s, ok := r.sentTo[p]
|
|
r.lock.RUnlock()
|
|
if !ok {
|
|
panic(nil)
|
|
}
|
|
|
|
defer func() {
|
|
// send feedback to server pool and remove peer if hard timeout happened
|
|
pp, ok := p.(*peer)
|
|
if ok && r.rm.serverPool != nil {
|
|
respTime := time.Duration(mclock.Now() - reqSent)
|
|
r.rm.serverPool.adjustResponseTime(pp.poolEntry, respTime, srto)
|
|
}
|
|
if hrto {
|
|
pp.Log().Debug("Request timed out hard")
|
|
if r.rm.peers != nil {
|
|
r.rm.peers.Unregister(pp.id)
|
|
}
|
|
}
|
|
|
|
r.lock.Lock()
|
|
delete(r.sentTo, p)
|
|
r.lock.Unlock()
|
|
}()
|
|
|
|
select {
|
|
case ok := <-s.valid:
|
|
if ok {
|
|
r.eventsCh <- reqPeerEvent{rpDeliveredValid, p}
|
|
} else {
|
|
r.eventsCh <- reqPeerEvent{rpDeliveredInvalid, p}
|
|
}
|
|
return
|
|
case <-time.After(softRequestTimeout):
|
|
srto = true
|
|
r.eventsCh <- reqPeerEvent{rpSoftTimeout, p}
|
|
}
|
|
|
|
select {
|
|
case ok := <-s.valid:
|
|
if ok {
|
|
r.eventsCh <- reqPeerEvent{rpDeliveredValid, p}
|
|
} else {
|
|
r.eventsCh <- reqPeerEvent{rpDeliveredInvalid, p}
|
|
}
|
|
case <-time.After(hardRequestTimeout):
|
|
hrto = true
|
|
r.eventsCh <- reqPeerEvent{rpHardTimeout, p}
|
|
}
|
|
}
|
|
|
|
// deliver a reply belonging to this request
|
|
func (r *sentReq) deliver(peer distPeer, msg *Msg) error {
|
|
r.lock.Lock()
|
|
defer r.lock.Unlock()
|
|
|
|
s, ok := r.sentTo[peer]
|
|
if !ok || s.delivered {
|
|
return errResp(ErrUnexpectedResponse, "reqID = %v", msg.ReqID)
|
|
}
|
|
valid := r.validate(peer, msg) == nil
|
|
r.sentTo[peer] = sentReqToPeer{true, s.valid}
|
|
s.valid <- valid
|
|
if !valid {
|
|
return errResp(ErrInvalidResponse, "reqID = %v", msg.ReqID)
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// stop stops the retrieval process and sets an error code that will be returned
|
|
// by getError
|
|
func (r *sentReq) stop(err error) {
|
|
r.lock.Lock()
|
|
if !r.stopped {
|
|
r.stopped = true
|
|
r.err = err
|
|
close(r.stopCh)
|
|
}
|
|
r.lock.Unlock()
|
|
}
|
|
|
|
// getError returns any retrieval error (either internally generated or set by the
|
|
// stop function) after stopCh has been closed
|
|
func (r *sentReq) getError() error {
|
|
return r.err
|
|
}
|
|
|
|
// genReqID generates a new random request ID
|
|
func genReqID() uint64 {
|
|
var rnd [8]byte
|
|
rand.Read(rnd[:])
|
|
return binary.BigEndian.Uint64(rnd[:])
|
|
}
|