1
Fork 0
mirror of https://github.com/jech/galene.git synced 2024-11-09 18:25:58 +01:00
galene/rtpconn/rtpwriter.go
Juliusz Chroboczek 9093339b62 Rework handling of buffered keyframes.
Instead of buffering the last keyframe, we merely keep track of its
seqno, and use the main cache for recovering.  We also send the whole
sequence of packets rather than just the keyframe itself.
2021-08-04 02:58:24 +02:00

370 lines
7.1 KiB
Go

package rtpconn
import (
"errors"
"log"
"sort"
"time"
"github.com/jech/galene/conn"
"github.com/jech/galene/packetcache"
"github.com/jech/galene/rtptime"
)
// packetIndex is a request to send a packet from the cache.
type packetIndex struct {
// the packet's seqno
seqno uint16
// the index in the cache
index uint16
}
// An rtpWriterPool is a set of rtpWriters
type rtpWriterPool struct {
track *rtpUpTrack
writers []*rtpWriter
count int
}
// sqrt computes the integer square root
func sqrt(n int) int {
if n < 2 {
return n
}
s := sqrt(n/2) * 2
l := s + 1
if l*l > n {
return s
} else {
return l
}
}
// add adds or removes a track from a writer pool
func (wp *rtpWriterPool) add(track conn.DownTrack, add bool) error {
n := 4
if wp.count > 16 {
n = sqrt(wp.count)
}
i := 0
for i < len(wp.writers) {
w := wp.writers[i]
err := w.add(track, add, n)
if err == nil {
if add {
wp.count++
} else {
if wp.count > 0 {
wp.count--
} else {
log.Printf("Negative writer count!")
}
}
return nil
} else if err == ErrWriterDead {
wp.del(wp.writers[i])
} else {
i++
}
}
if add {
writer := newRtpWriter(wp.track)
wp.writers = append(wp.writers, writer)
err := writer.add(track, true, n)
if err == nil {
wp.count++
}
return err
} else {
return errors.New("deleting unknown track")
}
}
// del deletes a writer.
func (wp *rtpWriterPool) del(w *rtpWriter) bool {
for i, ww := range wp.writers {
if ww == w {
close(w.ch)
wp.writers = append(wp.writers[:i], wp.writers[i+1:]...)
return true
}
}
return false
}
// close deletes all writers.
func (wp *rtpWriterPool) close() {
for _, w := range wp.writers {
close(w.ch)
}
wp.writers = nil
wp.count = 0
}
// write writes a packet stored in the packet cache to all local tracks
func (wp *rtpWriterPool) write(seqno uint16, index uint16, delay uint32, isvideo bool, marker bool) {
pi := packetIndex{seqno, index}
var dead []*rtpWriter
for _, w := range wp.writers {
if w.drop > 0 {
// currently dropping
if marker {
// last packet in frame
w.drop = 0
} else {
w.drop--
}
continue
}
select {
case w.ch <- pi:
// all is well
case <-w.done:
// the writer is dead.
dead = append(dead, w)
default:
// the writer is congested
if isvideo {
// drop until the end of the frame
if !marker {
w.drop = 7
}
continue
}
// audio, try again with a delay
d := delay / uint32(2*len(wp.writers))
timer := time.NewTimer(rtptime.ToDuration(
uint64(d), rtptime.JiffiesPerSec,
))
select {
case w.ch <- pi:
timer.Stop()
case <-w.done:
dead = append(dead, w)
case <-timer.C:
}
}
}
if dead != nil {
for _, d := range dead {
wp.del(d)
}
dead = nil
}
}
var ErrWriterDead = errors.New("writer is dead")
var ErrWriterBusy = errors.New("writer is busy")
var ErrUnknownTrack = errors.New("unknown track")
type writerAction struct {
add bool
track conn.DownTrack
maxTracks int
ch chan error
}
// an rtpWriter is a thread writing to a set of tracks.
type rtpWriter struct {
ch chan packetIndex
done chan struct{}
action chan writerAction
// this is not touched by the writer loop, used by the caller
drop int
}
func newRtpWriter(track *rtpUpTrack) *rtpWriter {
writer := &rtpWriter{
ch: make(chan packetIndex, 32),
done: make(chan struct{}),
action: make(chan writerAction, 1),
}
go rtpWriterLoop(writer, track)
return writer
}
// add adds or removes a track from a writer.
func (writer *rtpWriter) add(track conn.DownTrack, add bool, max int) error {
ch := make(chan error, 1)
select {
case writer.action <- writerAction{add, track, max, ch}:
select {
case err := <-ch:
return err
case <-writer.done:
return ErrWriterDead
}
case <-writer.done:
return ErrWriterDead
}
}
func sendSequence(kf, last uint16, track conn.DownTrack, cache *packetcache.Cache) {
buf := make([]byte, packetcache.BufSize)
seqno := kf
for ((last - seqno) & 0x8000) == 0 {
bytes := cache.Get(seqno, buf)
if bytes == 0 {
return
}
_, err := track.Write(buf[:bytes])
if err != nil {
return
}
seqno++
}
}
// rtpWriterLoop is the main loop of an rtpWriter.
func rtpWriterLoop(writer *rtpWriter, track *rtpUpTrack) {
defer close(writer.done)
buf := make([]byte, packetcache.BufSize)
local := make([]conn.DownTrack, 0)
for {
select {
case action := <-writer.action:
if action.add {
if len(local) >= action.maxTracks {
action.ch <- ErrWriterBusy
close(action.ch)
continue
}
local = append(local, action.track)
action.ch <- nil
close(action.ch)
track.mu.Lock()
ntp := track.srNTPTime
rtp := track.srRTPTime
track.mu.Unlock()
if ntp != 0 {
action.track.SetTimeOffset(ntp, rtp)
}
cname, ok := track.cname.Load().(string)
if ok && cname != "" {
action.track.SetCname(cname)
}
last, foundLast := track.cache.Last()
kf, foundKf := track.cache.Keyframe()
if foundLast && foundKf {
if last-kf < 40 { // modulo 2^16
go sendSequence(
kf, last,
action.track,
track.cache,
)
} else {
track.RequestKeyframe()
}
} else {
// no keyframe yet, one should
// arrive soon. Do nothing.
}
} else {
found := false
for i, t := range local {
if t == action.track {
local = append(local[:i],
local[i+1:]...)
found = true
break
}
}
if !found {
action.ch <- ErrUnknownTrack
} else {
action.ch <- nil
}
close(action.ch)
if len(local) == 0 {
return
}
}
case pi, ok := <-writer.ch:
if !ok {
return
}
bytes := track.cache.GetAt(pi.seqno, pi.index, buf)
if bytes == 0 {
continue
}
for _, l := range local {
_, err := l.Write(buf[:bytes])
if err != nil {
continue
}
}
}
}
}
// nackWriter is called when bufferedNACKs becomes non-empty. It decides
// which nacks to ship out.
func nackWriter(track *rtpUpTrack) {
// a client might send us a NACK for a packet that has already
// been nacked by the reader loop. Give recovery a chance.
time.Sleep(50 * time.Millisecond)
track.mu.Lock()
nacks := track.bufferedNACKs
track.bufferedNACKs = nil
track.mu.Unlock()
if len(nacks) == 0 || !track.hasRtcpFb("nack", "") {
return
}
time.Sleep(50 * time.Millisecond)
// drop any nacks before the last keyframe
var cutoff uint16
seqno, found := track.cache.Keyframe()
if found {
cutoff = seqno
} else {
lastSeqno, last := track.cache.Last()
if !last {
// NACK on a fresh track? Give up.
return
}
// no keyframe, use an arbitrary cutoff
cutoff = lastSeqno - 256
}
i := 0
for i < len(nacks) {
if ((nacks[i] - cutoff) & 0x8000) != 0 {
// earlier than the cutoff, drop
nacks = append(nacks[:i], nacks[i+1:]...)
continue
}
l := track.cache.Get(nacks[i], nil)
if l > 0 {
// the packet arrived in the meantime
nacks = append(nacks[:i], nacks[i+1:]...)
continue
}
i++
}
sort.Slice(nacks, func(i, j int) bool {
return nacks[i]-cutoff < nacks[j]-cutoff
})
if len(nacks) > 0 {
track.sendNACKs(nacks)
}
}