fix(gateway): bound Apollo video ingress
Verify Data Plane / gateway (push) Successful in 4m14s

This commit is contained in:
sechmachine
2026-08-09 19:07:22 +07:00
parent a0ca194691
commit 55afea72a1
8 changed files with 467 additions and 39 deletions
+126 -36
View File
@@ -33,6 +33,9 @@ const (
nativeApolloVideoQueueLatency = 250 * time.Millisecond
nativeApolloAudioQueuePackets = 16
nativeApolloEventQueuePackets = 16
nativeApolloVideoIngressSlots = 2048
nativeApolloVideoPacketBytes = apolloVideoHeaderSize + apolloVideoRawPacketSize
nativeApolloVideoReadBuffer = nativeApolloVideoIngressSlots * nativeApolloVideoPacketBytes
)
// NativeApolloBackend keeps provider sockets inside the gateway process. The
@@ -41,8 +44,9 @@ const (
type NativeApolloBackend struct {
Dialer *net.Dialer
mu sync.Mutex
pending map[string]*apolloRTSPSetup
mu sync.Mutex
pending map[string]*apolloRTSPSetup
configureMedia func(*apolloMediaCodec)
}
func NewNativeApolloBackend() *NativeApolloBackend {
@@ -268,7 +272,7 @@ func (b *NativeApolloBackend) Open(ctx context.Context, request LaunchRequest, _
if !ok || setup == nil {
return nil, ErrProviderDisconnected
}
return newNativeApolloProviderSession(ctx, setup)
return newNativeApolloProviderSession(ctx, setup, b.configureMedia)
}
func readBounded(reader io.Reader, max int) ([]byte, error) {
@@ -336,7 +340,7 @@ func newNativeApolloSession(sessionID string) *nativeApolloSession {
}
}
func newNativeApolloProviderSession(ctx context.Context, setup *apolloRTSPSetup) (*nativeApolloSession, error) {
func newNativeApolloProviderSession(ctx context.Context, setup *apolloRTSPSetup, configureMedia func(*apolloMediaCodec)) (*nativeApolloSession, error) {
if setup == nil || len(setup.streamKey) != 16 || setup.controlPort == 0 || setup.audioPort == 0 || setup.videoPort == 0 {
return nil, ErrProviderMalformed
}
@@ -360,6 +364,9 @@ func newNativeApolloProviderSession(ctx context.Context, setup *apolloRTSPSetup)
peer.close(err)
return nil, err
}
if configureMedia != nil {
configureMedia(media)
}
session := newNativeApolloSession(setup.sessionID)
managementClient, err := newPinnedApolloHTTPClient(setup.providerWork)
if err != nil {
@@ -391,6 +398,12 @@ func newNativeApolloProviderSession(ctx context.Context, setup *apolloRTSPSetup)
peer.close(err)
return nil, err
}
if err := videoConn.SetReadBuffer(nativeApolloVideoReadBuffer); err != nil {
_ = audioConn.Close()
_ = videoConn.Close()
peer.close(err)
return nil, fmt.Errorf("set Apollo video receive buffer: %w", err)
}
if _, err := audioConn.Write(apolloMediaPing(setup.audioPing, 1)); err != nil {
_ = audioConn.Close()
_ = videoConn.Close()
@@ -896,19 +909,20 @@ func (s *nativeApolloSession) readUDPMedia() {
s.closeMediaChannels()
return
}
var readers sync.WaitGroup
readers.Add(2)
read := func(conn *net.UDPConn, output chan ProviderMedia, video bool) {
defer readers.Done()
videoIngress := newNativeApolloVideoIngress()
var workers sync.WaitGroup
workers.Add(3)
go func() {
defer workers.Done()
buffer := make([]byte, apolloMediaMaximumPacket+1)
for {
if s.mediaQuiesced.Load() {
return
}
if err := conn.SetReadDeadline(time.Now().Add(250 * time.Millisecond)); err != nil {
if err := s.audioConn.SetReadDeadline(time.Now().Add(250 * time.Millisecond)); err != nil {
return
}
count, err := conn.Read(buffer)
count, err := s.audioConn.Read(buffer)
receivedAt := time.Now()
if err != nil {
if networkErr, ok := err.(net.Error); ok && networkErr.Timeout() {
@@ -928,45 +942,121 @@ func (s *nativeApolloSession) readUDPMedia() {
return
}
s.mediaIngress.Add(1)
var payloads [][]byte
if video {
shard, openErr := s.media.OpenVideo(buffer[:count])
if openErr != nil {
continue
}
payload, err := s.videoFEC.Add(shard)
if err != nil || len(payload) == 0 {
continue
}
payloads = [][]byte{payload}
} else {
shard, openErr := s.media.OpenAudio(buffer[:count])
if openErr != nil {
continue
}
var evicted bool
payloads, evicted, err = s.audioFEC.Add(s.media, shard)
if evicted {
s.mediaDrops.Add(1)
}
shard, openErr := s.media.OpenAudio(buffer[:count])
if openErr != nil {
continue
}
payloads, evicted, err := s.audioFEC.Add(s.media, shard)
if evicted {
s.mediaDrops.Add(1)
}
if err != nil {
continue
}
for _, payload := range payloads {
s.enqueueMedia(output, payload, receivedAt)
s.enqueueMedia(s.audio, payload, receivedAt)
}
}
}
go read(s.audioConn, s.audio, false)
go read(s.videoConn, s.video, true)
}()
go func() {
readers.Wait()
defer workers.Done()
s.drainApolloVideo(videoIngress)
}()
go func() {
defer workers.Done()
s.processApolloVideo(videoIngress)
}()
go func() {
workers.Wait()
close(s.readDone)
s.closeMediaChannels()
}()
}
type nativeApolloVideoIngressSlot struct {
packet [apolloMediaMaximumPacket + 1]byte
count int
receivedAt time.Time
}
type nativeApolloVideoIngress struct {
slots [nativeApolloVideoIngressSlots]nativeApolloVideoIngressSlot
free chan uint16
ready chan uint16
scratch [apolloMediaMaximumPacket + 1]byte
}
func newNativeApolloVideoIngress() *nativeApolloVideoIngress {
ingress := &nativeApolloVideoIngress{
free: make(chan uint16, nativeApolloVideoIngressSlots),
ready: make(chan uint16, nativeApolloVideoIngressSlots),
}
for index := range nativeApolloVideoIngressSlots {
ingress.free <- uint16(index)
}
return ingress
}
func (s *nativeApolloSession) drainApolloVideo(ingress *nativeApolloVideoIngress) {
defer close(ingress.ready)
for {
if s.mediaQuiesced.Load() {
return
}
select {
case index := <-ingress.free:
slot := &ingress.slots[index]
count, err := s.videoConn.Read(slot.packet[:])
receivedAt := time.Now()
if err != nil {
ingress.free <- index
return
}
if count > apolloMediaMaximumPacket {
ingress.free <- index
continue
}
if s.mediaQuiesced.Load() {
ingress.free <- index
return
}
s.mediaIngress.Add(1)
slot.count, slot.receivedAt = count, receivedAt
ingress.ready <- index
default:
count, err := s.videoConn.Read(ingress.scratch[:])
if err != nil {
return
}
if count > apolloMediaMaximumPacket {
continue
}
if s.mediaQuiesced.Load() {
return
}
s.mediaIngress.Add(1)
s.mediaDrops.Add(1)
}
}
}
func (s *nativeApolloSession) processApolloVideo(ingress *nativeApolloVideoIngress) {
for index := range ingress.ready {
slot := &ingress.slots[index]
if !s.mediaQuiesced.Load() {
shard, err := s.media.OpenVideo(slot.packet[:slot.count])
if err == nil {
payload, fecErr := s.videoFEC.Add(shard)
if fecErr == nil && len(payload) > 0 {
s.enqueueMedia(s.video, payload, slot.receivedAt)
}
}
}
slot.count, slot.receivedAt = 0, time.Time{}
ingress.free <- index
}
}
func pushLatest[T any](channel chan T, payload T) bool {
select {
case channel <- payload:
+190
View File
@@ -17,8 +17,10 @@ import (
"net"
"net/http"
"net/http/httptest"
"runtime"
"strconv"
"strings"
"sync"
"testing"
"time"
@@ -760,6 +762,34 @@ func TestNativeApolloSessionRelaysOnlyAuthenticatedEncodedUDPMedia(t *testing.T)
if drops := session.Telemetry().MediaDrops; drops < 2 {
t.Fatalf("stale FEC eviction drops = %d, want at least 2", drops)
}
beforeIngress := session.mediaIngress.Load()
beforeDrops := session.mediaDrops.Load()
if _, err := videoServer.WriteToUDP(make([]byte, apolloMediaMaximumPacket+2), videoClient.LocalAddr().(*net.UDPAddr)); err != nil {
t.Fatal(err)
}
marker := []byte{0xde, 0xad, 0xbe, 0xef}
markerPayload := make([]byte, apolloVideoShardPayloadSize)
markerPayload[0], markerPayload[3] = 0x01, 0x01
binary.LittleEndian.PutUint16(markerPayload[4:6], uint16(8+len(marker)))
copy(markerPayload[8:], marker)
markerPacket := sourceEncryptVideoRaw(t, key, sourceShapedVideoRaw(44, 103, 103, 0x07, 1, 0, 0, markerPayload), "0123456789dV")
if _, err := videoServer.WriteToUDP(markerPacket, videoClient.LocalAddr().(*net.UDPAddr)); err != nil {
t.Fatal(err)
}
select {
case media := <-session.Video():
if !bytes.Equal(media.Payload, marker) {
t.Fatalf("post-oversize marker relay = %x, want %x", media.Payload, marker)
}
case <-time.After(time.Second):
t.Fatal("post-oversize marker was not relayed")
}
if got := session.mediaIngress.Load() - beforeIngress; got != 1 {
t.Fatalf("post-oversize video ingress = %d, want 1 valid marker", got)
}
if got := session.mediaDrops.Load(); got != beforeDrops {
t.Fatalf("post-oversize video drops = %d, want unchanged %d", got, beforeDrops)
}
terminateCtx, cancel := context.WithTimeout(context.Background(), time.Second)
defer cancel()
if err := session.Terminate(terminateCtx); err != nil {
@@ -767,6 +797,166 @@ func TestNativeApolloSessionRelaysOnlyAuthenticatedEncodedUDPMedia(t *testing.T)
}
}
func TestNativeApolloVideoIngressSaturationIsBoundedAndAccounted(t *testing.T) {
if nativeApolloVideoIngressSlots != 2048 || nativeApolloVideoReadBuffer != 2_195_456 {
t.Fatalf("video ingress bounds = %d slots/%d bytes, want 2048/2195456", nativeApolloVideoIngressSlots, nativeApolloVideoReadBuffer)
}
ingress := newNativeApolloVideoIngress()
if len(ingress.slots) != 2048 || len(ingress.free) != 2048 || cap(ingress.ready) != 2048 || len(ingress.slots[0].packet) != apolloMediaMaximumPacket+1 {
t.Fatalf("video ingress pool = slots:%d free:%d ready-cap:%d packet:%d",
len(ingress.slots), len(ingress.free), cap(ingress.ready), len(ingress.slots[0].packet))
}
key := []byte("0123456789abcdef")
session, audioServer, videoServer := newNativeApolloMediaTestSession(t, key)
defer audioServer.Close()
defer videoServer.Close()
blockCtx, cancelBlock := context.WithCancel(context.Background())
defer cancelBlock()
blocked := make(chan struct{})
release := make(chan struct{})
var releaseOnce sync.Once
releaseAEAD := func() { releaseOnce.Do(func() { close(release) }) }
defer releaseAEAD()
session.media.aead = &qualificationBlockingAEAD{
AEAD: session.media.aead, ctx: blockCtx, blocked: blocked, release: release,
}
go session.readUDPMedia()
packet := sourceShapedEncryptedVideoPacket(t, key, []byte{1})
if _, err := videoServer.WriteToUDP(packet, session.videoConn.LocalAddr().(*net.UDPAddr)); err != nil {
t.Fatal(err)
}
select {
case <-blocked:
case <-time.After(time.Second):
t.Fatal("video processor did not block")
}
waitIngress := func(want uint64) {
t.Helper()
deadline := time.NewTimer(time.Second)
defer deadline.Stop()
for session.mediaIngress.Load() < want {
select {
case <-deadline.C:
t.Fatalf("video ingress = %d, want at least %d", session.mediaIngress.Load(), want)
default:
runtime.Gosched()
}
}
}
waitIngress(1)
const overflow = 33
sent := uint64(1)
for remaining := nativeApolloVideoIngressSlots - 1 + overflow; remaining > 0; {
batch := min(32, remaining)
for range batch {
if _, err := videoServer.WriteToUDP([]byte{1}, session.videoConn.LocalAddr().(*net.UDPAddr)); err != nil {
t.Fatal(err)
}
}
sent += uint64(batch)
waitIngress(sent)
remaining -= batch
}
if got := session.mediaDrops.Load(); got != overflow {
t.Fatalf("saturated video ingress drops = %d, want %d", got, overflow)
}
session.quiesceMedia()
cancelBlock()
select {
case <-session.readDone:
case <-time.After(time.Second):
t.Fatal("video ingress workers did not stop")
}
}
func TestNativeApolloTerminateStopsActiveVideoDrainAndProcessor(t *testing.T) {
key := []byte("0123456789abcdef")
session, audioServer, videoServer := newNativeApolloMediaTestSession(t, key)
defer audioServer.Close()
defer videoServer.Close()
blockCtx, cancelBlock := context.WithCancel(context.Background())
t.Cleanup(cancelBlock)
blocked := make(chan struct{})
session.media.aead = &qualificationBlockingAEAD{
AEAD: session.media.aead, ctx: blockCtx, blocked: blocked, release: make(chan struct{}),
}
go session.readUDPMedia()
if _, err := videoServer.WriteToUDP(sourceShapedEncryptedVideoPacket(t, key, []byte{1}), session.videoConn.LocalAddr().(*net.UDPAddr)); err != nil {
t.Fatal(err)
}
select {
case <-blocked:
case <-time.After(time.Second):
t.Fatal("video processor did not block")
}
terminateCtx, cancelTerminate := context.WithTimeout(context.Background(), time.Second)
defer cancelTerminate()
terminated := make(chan error, 1)
go func() { terminated <- session.Terminate(terminateCtx) }()
cancelBlock()
if err := <-terminated; err != nil {
t.Fatalf("Terminate() error = %v", err)
}
select {
case <-session.readDone:
default:
t.Fatal("Terminate returned before video drain and processor stopped")
}
select {
case _, ok := <-session.Video():
if ok {
t.Fatal("video channel remained open after worker shutdown")
}
default:
t.Fatal("video channel was not closed after worker shutdown")
}
}
func newNativeApolloMediaTestSession(t *testing.T, key []byte) (*nativeApolloSession, *net.UDPConn, *net.UDPConn) {
t.Helper()
audioServer, err := net.ListenUDP("udp", &net.UDPAddr{IP: net.ParseIP("127.0.0.1")})
if err != nil {
t.Fatal(err)
}
videoServer, err := net.ListenUDP("udp", &net.UDPAddr{IP: net.ParseIP("127.0.0.1")})
if err != nil {
_ = audioServer.Close()
t.Fatal(err)
}
audioClient, err := net.DialUDP("udp", nil, audioServer.LocalAddr().(*net.UDPAddr))
if err != nil {
_ = audioServer.Close()
_ = videoServer.Close()
t.Fatal(err)
}
videoClient, err := net.DialUDP("udp", nil, videoServer.LocalAddr().(*net.UDPAddr))
if err != nil {
_ = audioClient.Close()
_ = audioServer.Close()
_ = videoServer.Close()
t.Fatal(err)
}
session := newNativeApolloSession("session-video-ingress")
session.media, err = newApolloMediaCodec(key, 1)
if err != nil {
_ = audioClient.Close()
_ = videoClient.Close()
_ = audioServer.Close()
_ = videoServer.Close()
t.Fatal(err)
}
session.audioConn, session.videoConn = audioClient, videoClient
t.Cleanup(func() {
_ = audioClient.Close()
_ = videoClient.Close()
})
return session, audioServer, videoServer
}
func TestNativeApolloTerminateReleasesPressedProviderInput(t *testing.T) {
server, err := net.ListenUDP("udp", &net.UDPAddr{IP: net.ParseIP("127.0.0.1")})
if err != nil {
+114
View File
@@ -4,6 +4,7 @@ import (
"bytes"
"compress/gzip"
"context"
"crypto/cipher"
"encoding/binary"
"errors"
"fmt"
@@ -15,10 +16,35 @@ import (
"runtime"
"strconv"
"strings"
"sync"
"testing"
"time"
)
type qualificationBlockingAEAD struct {
cipher.AEAD
ctx context.Context
blocked chan struct{}
release chan struct{}
once sync.Once
waitErr error
}
func (a *qualificationBlockingAEAD) Open(dst, nonce, ciphertext, additionalData []byte) ([]byte, error) {
a.once.Do(func() {
close(a.blocked)
select {
case <-a.release:
case <-a.ctx.Done():
a.waitErr = a.ctx.Err()
}
})
if a.waitErr != nil {
return nil, a.waitErr
}
return a.AEAD.Open(dst, nonce, ciphertext, additionalData)
}
func TestQualificationCatalogMatchesSection7(t *testing.T) {
media := qualificationMediaProfiles()
if len(media) != 3 {
@@ -757,6 +783,94 @@ func TestQualificationLateSourceBatchDoesNotCollapseThroughPublicPath(t *testing
path.session.mediaRecovered.Load(), path.session.mediaEnqueued.Load())
}
func TestQualificationBlockedVideoAEADDoesNotBlockProviderIngress(t *testing.T) {
profile := qualificationMediaProfiles()[2]
blockCtx, cancelBlock := context.WithCancel(context.Background())
defer cancelBlock()
blocked := make(chan struct{})
release := make(chan struct{})
var releaseOnce sync.Once
releaseAEAD := func() { releaseOnce.Do(func() { close(release) }) }
t.Cleanup(releaseAEAD)
native := NewNativeApolloBackend()
native.configureMedia = func(media *apolloMediaCodec) {
media.aead = &qualificationBlockingAEAD{
AEAD: media.aead, ctx: blockCtx, blocked: blocked, release: release,
}
}
path := newQualificationPathWithNativeBackend(t, profile, profile.BitrateKbps, nil, native)
t.Cleanup(func() {
path.Close()
select {
case <-path.session.readDone:
case <-time.After(time.Second):
t.Error("native media workers did not stop during qualification cleanup")
}
})
payload := qualificationFramePayload(profile, 0)
packets := qualificationSourceVideoPackets(t, path.key, 1, payload)
if len(payload) != 666_664 || len(packets) != 662 {
t.Fatalf("4K60 keyframe = %d bytes/%d shards, want 666664/662", len(payload), len(packets))
}
beforeMetrics := path.server.Metrics()
beforeIngress := path.session.mediaIngress.Load()
beforeRecovered := path.session.mediaRecovered.Load()
beforeEnqueued := path.session.mediaEnqueued.Load()
beforePacer := path.server.pacer.reservations.Load()
if _, err := path.emit(t, payload); err != nil {
t.Fatal(err)
}
select {
case <-blocked:
case <-time.After(time.Second):
t.Fatal("video AEAD did not block")
}
if sent := path.fixture.sentPackets.Load(); sent != 662 {
t.Fatalf("fixture sent packets = %d, want 662", sent)
}
deadline := time.NewTimer(250 * time.Millisecond)
defer deadline.Stop()
for path.session.mediaIngress.Load()-beforeIngress != 662 {
select {
case <-deadline.C:
t.Fatalf("blocked-AEAD ingress = %d, want 662 after 662 successful fixture writes",
path.session.mediaIngress.Load()-beforeIngress)
default:
runtime.Gosched()
}
}
releaseAEAD()
recovered, err := path.receivePayload(context.Background())
if err != nil {
t.Fatal(err)
}
afterMetrics := path.server.Metrics()
stageDeadline := time.Now().Add(2 * time.Second)
for (path.session.mediaRecovered.Load() <= beforeRecovered ||
path.session.mediaEnqueued.Load() <= beforeEnqueued ||
afterMetrics.ProcessingSamples <= beforeMetrics.ProcessingSamples ||
afterMetrics.MediaPackets <= beforeMetrics.MediaPackets) && time.Now().Before(stageDeadline) {
runtime.Gosched()
afterMetrics = path.server.Metrics()
}
if path.backend.setups.Load() != 1 || path.backend.opens.Load() != 1 ||
path.session.mediaRecovered.Load()-beforeRecovered != 1 ||
path.session.mediaEnqueued.Load()-beforeEnqueued != 1 ||
afterMetrics.ProcessingSamples <= beforeMetrics.ProcessingSamples ||
path.server.pacer.reservations.Load() <= beforePacer ||
afterMetrics.MediaPackets <= beforeMetrics.MediaPackets || !bytes.Equal(recovered, payload) {
t.Fatalf("blocked-AEAD public path: setup=%d open=%d ingress=%d recovered=%d enqueued=%d processing=%d pacer=%d media=%d payload=%t",
path.backend.setups.Load(), path.backend.opens.Load(), path.session.mediaIngress.Load()-beforeIngress,
path.session.mediaRecovered.Load()-beforeRecovered, path.session.mediaEnqueued.Load()-beforeEnqueued,
afterMetrics.ProcessingSamples-beforeMetrics.ProcessingSamples,
path.server.pacer.reservations.Load()-beforePacer, afterMetrics.MediaPackets-beforeMetrics.MediaPackets,
bytes.Equal(recovered, payload))
}
}
func TestQualificationProcessingRetainsProviderIngressDiagnostics(t *testing.T) {
profile := qualificationMediaProfiles()[2]
path := newQualificationProcessingPath(t, profile, profile.BitrateKbps)
+5 -1
View File
@@ -1055,13 +1055,17 @@ func newQualificationImpairedPath(t *testing.T, profile qualificationMediaProfil
}
func newQualificationPathWithImpairment(t *testing.T, profile qualificationMediaProfile, pacerKbps int64, impairment *qualificationImpairmentProfile) *qualificationPath {
return newQualificationPathWithNativeBackend(t, profile, pacerKbps, impairment, NewNativeApolloBackend())
}
func newQualificationPathWithNativeBackend(t *testing.T, profile qualificationMediaProfile, pacerKbps int64, impairment *qualificationImpairmentProfile, native *NativeApolloBackend) *qualificationPath {
t.Helper()
serverTLS, clientTLS := testTLS(t)
fixture := newQualificationApolloFixture(t, serverTLS, clientTLS, "qualification-session", profile)
if impairment != nil {
fixture.setControlImpairment(*impairment)
}
backend := &qualificationTracingBackend{native: NewNativeApolloBackend()}
backend := &qualificationTracingBackend{native: native}
provider := NewApolloAdapter(backend, ProviderIdentity{})
authority := protocol.SessionAuthority{
Version: "1", SessionID: "qualification-session", GatewayID: "gateway-1",