Record replay audio

images/chromium-headful/Dockerfile

@@ -366,8 +366,8 @@ COPY --from=xorg-deps /usr/local/lib/xorg/modules/drivers/dummy_drv.so /usr/lib/
 COPY --from=xorg-deps /usr/local/lib/xorg/modules/input/neko_drv.so /usr/lib/xorg/modules/input/neko_drv.so
 
 COPY images/chromium-headful/image-chromium/ /
-COPY images/chromium-headful/start-pulseaudio.sh /images/chromium-headful/start-pulseaudio.sh
-RUN chmod +x /images/chromium-headful/start-pulseaudio.sh
+COPY shared/start-pulseaudio.sh /usr/local/bin/start-pulseaudio.sh
+RUN chmod +x /usr/local/bin/start-pulseaudio.sh
 COPY images/chromium-headful/supervisord.conf /etc/supervisor/supervisord.conf
 COPY images/chromium-headful/supervisor/services/ /etc/supervisor/conf.d/services/
 COPY shared/envoy/supervisor-envoy.conf /etc/supervisor/conf.d/services/envoy.conf

images/chromium-headful/client/src/components/video.vue

@@ -541,6 +541,8 @@
 
       /* Initialize Guacamole Keyboard */
       this.keyboard.onkeydown = (key: number) => {
+        this.unmuteOnInteraction()
+
         if (!this.hosting || this.locked) {
           return true
         }
@@ -670,6 +672,18 @@
       this.$accessor.video.setMuted(false)
     }
 
+    // The autoplay policy only lets us unmute inside a real user gesture, and
+    // this client hides the unmute overlay. Piggyback on the existing input
+    // handlers so the first interaction with the live view unmutes. Guarded on
+    // the current state so it is a cheap no-op once unmuted, yet re-applies if a
+    // reconnect re-mutes the element. (mousemove is intentionally not used: it
+    // is not a user-activation event, so it cannot unlock audio.)
+    unmuteOnInteraction() {
+      if (this.muted) {
+        this.unmute()
+      }
+    }
+
     toggleControl() {
       if (!this.playable) {
         return
@@ -789,6 +803,8 @@
     }
 
     onMouseDown(e: MouseEvent) {
+      this.unmuteOnInteraction()
+
       if (!this.hosting) {
         this.$emit('control-attempt', e)
       }

images/chromium-headful/neko.yaml

@@ -42,6 +42,14 @@ session:
     # needed for legacy API
     enabled: false
 
+capture:
+  audio:
+    # Capture from the monitor of the recorder's playback sink so live view
+    # streams the same browser audio that gets recorded. Neko defaults to
+    # "audio_output.monitor", which does not exist here (our sink is
+    # KernelOutput), so without this override live view has no audio.
+    device: "KernelOutput.monitor"
+
 plugins:
   enabled: false
 

images/chromium-headful/supervisor/services/kernel-images-api.conf

@@ -1,5 +1,6 @@
 [program:kernel-images-api]
-command=/bin/bash -lc 'mkdir -p "${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" && PORT="${KERNEL_IMAGES_API_PORT:-10001}" FRAME_RATE="${KERNEL_IMAGES_API_FRAME_RATE:-10}" DISPLAY_NUM="${KERNEL_IMAGES_API_DISPLAY_NUM:-${DISPLAY_NUM:-1}}" MAX_SIZE_MB="${KERNEL_IMAGES_API_MAX_SIZE_MB:-500}" OUTPUT_DIR="${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" LOG_CDP_MESSAGES="${LOG_CDP_MESSAGES:-false}" S2_BASIN="${S2_BASIN:-}" S2_ACCESS_TOKEN="${S2_ACCESS_TOKEN:-}" S2_STREAM="${S2_STREAM:-}" exec /usr/local/bin/kernel-images-api'
+; AUDIO_SOURCE and PULSE_SERVER defaults must match shared/start-pulseaudio.sh.
+command=/bin/bash -lc 'mkdir -p "${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" && PORT="${KERNEL_IMAGES_API_PORT:-10001}" FRAME_RATE="${KERNEL_IMAGES_API_FRAME_RATE:-10}" DISPLAY_NUM="${KERNEL_IMAGES_API_DISPLAY_NUM:-${DISPLAY_NUM:-1}}" MAX_SIZE_MB="${KERNEL_IMAGES_API_MAX_SIZE_MB:-500}" OUTPUT_DIR="${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" AUDIO_SOURCE="${KERNEL_IMAGES_API_AUDIO_SOURCE:-${AUDIO_SOURCE:-KernelOutput.monitor}}" PULSE_SERVER="${PULSE_SERVER:-unix:/tmp/pulse/native}" LOG_CDP_MESSAGES="${LOG_CDP_MESSAGES:-false}" S2_BASIN="${S2_BASIN:-}" S2_ACCESS_TOKEN="${S2_ACCESS_TOKEN:-}" S2_STREAM="${S2_STREAM:-}" exec /usr/local/bin/kernel-images-api'
 autostart=false
 autorestart=true
 startsecs=0

images/chromium-headful/supervisor/services/neko.conf

@@ -3,5 +3,8 @@ command=/usr/bin/neko serve --server.static /var/www --server.bind 0.0.0.0:8080
 autostart=false
 autorestart=true
 startsecs=0
+# neko's gstreamer pulsesrc needs the shared PulseAudio socket to capture
+# KernelOutput.monitor for live-view audio
+environment=PULSE_SERVER="unix:/tmp/pulse/native"
 stdout_logfile=/var/log/supervisord/neko
 redirect_stderr=true

images/chromium-headful/supervisor/services/pulseaudio.conf

@@ -1,7 +1,7 @@
 [program:pulseaudio]
-command=/bin/bash -lc '/images/chromium-headful/start-pulseaudio.sh'
+command=/bin/bash -lc '/usr/local/bin/start-pulseaudio.sh'
 autostart=false
-autorestart=false
+autorestart=true
 startsecs=0
 exitcodes=0
 stdout_logfile=/var/log/supervisord/pulseaudio

images/chromium-headless/image/Dockerfile

@@ -149,6 +149,7 @@ RUN --mount=type=cache,target=/var/cache/apt,sharing=locked,id=$CACHEIDPREFIX-ap
     gpg-agent \
     dbus \
     dbus-x11 \
+    pulseaudio \
     xvfb \
     x11-utils \
     x11-xserver-utils \
@@ -235,6 +236,8 @@ RUN useradd -m -s /bin/bash kernel
 # supervisor  start scripts
 COPY images/chromium-headless/image/start-xvfb.sh /images/chromium-headless/image/start-xvfb.sh
 RUN chmod +x /images/chromium-headless/image/start-xvfb.sh
+COPY shared/start-pulseaudio.sh /usr/local/bin/start-pulseaudio.sh
+RUN chmod +x /usr/local/bin/start-pulseaudio.sh
 
 # Container entrypoint wrapper (Go binary, replaces wrapper.sh)
 COPY --from=server-builder /out/wrapper /wrapper

images/chromium-headless/image/supervisor/services/kernel-images-api.conf

@@ -1,5 +1,6 @@
 [program:kernel-images-api]
-command=/bin/bash -lc 'mkdir -p "${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" && PORT="${KERNEL_IMAGES_API_PORT:-10001}" FRAME_RATE="${KERNEL_IMAGES_API_FRAME_RATE:-10}" DISPLAY_NUM="${KERNEL_IMAGES_API_DISPLAY_NUM:-${DISPLAY_NUM:-1}}" MAX_SIZE_MB="${KERNEL_IMAGES_API_MAX_SIZE_MB:-500}" OUTPUT_DIR="${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" LOG_CDP_MESSAGES="${LOG_CDP_MESSAGES:-false}" S2_BASIN="${S2_BASIN:-}" S2_ACCESS_TOKEN="${S2_ACCESS_TOKEN:-}" S2_STREAM="${S2_STREAM:-}" exec /usr/local/bin/kernel-images-api'
+; AUDIO_SOURCE and PULSE_SERVER defaults must match shared/start-pulseaudio.sh.
+command=/bin/bash -lc 'mkdir -p "${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" && PORT="${KERNEL_IMAGES_API_PORT:-10001}" FRAME_RATE="${KERNEL_IMAGES_API_FRAME_RATE:-10}" DISPLAY_NUM="${KERNEL_IMAGES_API_DISPLAY_NUM:-${DISPLAY_NUM:-1}}" MAX_SIZE_MB="${KERNEL_IMAGES_API_MAX_SIZE_MB:-500}" OUTPUT_DIR="${KERNEL_IMAGES_API_OUTPUT_DIR:-/recordings}" AUDIO_SOURCE="${KERNEL_IMAGES_API_AUDIO_SOURCE:-${AUDIO_SOURCE:-KernelOutput.monitor}}" PULSE_SERVER="${PULSE_SERVER:-unix:/tmp/pulse/native}" LOG_CDP_MESSAGES="${LOG_CDP_MESSAGES:-false}" S2_BASIN="${S2_BASIN:-}" S2_ACCESS_TOKEN="${S2_ACCESS_TOKEN:-}" S2_STREAM="${S2_STREAM:-}" exec /usr/local/bin/kernel-images-api'
 autostart=false
 autorestart=true
 startsecs=0

images/chromium-headless/image/supervisor/services/pulseaudio.conf

@@ -0,0 +1,8 @@
+[program:pulseaudio]
+command=/bin/bash -lc '/usr/local/bin/start-pulseaudio.sh'
+autostart=false
+autorestart=true
+startsecs=0
+exitcodes=0
+stdout_logfile=/var/log/supervisord/pulseaudio
+redirect_stderr=true

server/cmd/api/api/api.go

@@ -150,6 +150,7 @@ func (s *ApiService) StartRecording(ctx context.Context, req oapi.StartRecording
 		params.FrameRate = req.Body.Framerate
 		params.MaxSizeInMB = req.Body.MaxFileSizeInMB
 		params.MaxDurationInSeconds = req.Body.MaxDurationInSeconds
+		params.RecordAudio = req.Body.RecordAudio
 	}
 
 	// Determine recorder ID (use default if none provided)
@@ -161,6 +162,10 @@ func (s *ApiService) StartRecording(ctx context.Context, req oapi.StartRecording
 	// Create, register, and start a new recorder
 	rec, err := s.factory(recorderID, params)
 	if err != nil {
+		if errors.Is(err, recorder.ErrInvalidParams) {
+			log.Warn("invalid recording parameters", "err", err, "recorder_id", recorderID)
+			return oapi.StartRecording400JSONResponse{BadRequestErrorJSONResponse: oapi.BadRequestErrorJSONResponse{Message: err.Error()}}, nil
+		}
 		log.Error("failed to create recorder", "err", err, "recorder_id", recorderID)
 		return oapi.StartRecording500JSONResponse{InternalErrorJSONResponse: oapi.InternalErrorJSONResponse{Message: "failed to create recording"}}, nil
 	}

server/cmd/api/main.go

@@ -72,6 +72,8 @@ func main() {
 		FrameRate:   &config.FrameRate,
 		MaxSizeInMB: &config.MaxSizeInMB,
 		OutputDir:   &config.OutputDir,
+		AudioSource: &config.AudioSource,
+		PulseServer: &config.PulseServer,
 	}
 	if err := defaultParams.Validate(); err != nil {
 		slogger.Error("invalid default recording parameters", "err", err)

server/cmd/chromium-launcher/main.go

@@ -16,6 +16,18 @@ import (
 	"github.com/kernel/kernel-images/server/lib/x11"
 )
 
+// shared/start-pulseaudio.sh is the authority for the audio topology. These are
+// the fixed contract values it creates, not overridable defaults: chromium must
+// connect to the same socket and play into the same sink the daemon sets up.
+// Keep them in sync with start-pulseaudio.sh.
+const (
+	// pulseServer is the PulseAudio socket the recorder and chromium share.
+	pulseServer = "unix:/tmp/pulse/native"
+	// pulseSink is the null sink chromium plays into; the recorder captures
+	// its .monitor source.
+	pulseSink = "KernelOutput"
+)
+
 func main() {
 	headless := flag.Bool("headless", false, "Run Chromium with headless flags")
 	chromiumPath := flag.String("chromium", "chromium", "Chromium binary path (default: chromium)")
@@ -89,11 +101,15 @@ func main() {
 
 	runAsRoot := strings.EqualFold(strings.TrimSpace(os.Getenv("RUN_AS_ROOT")), "true")
 
-	// Prepare environment
+	// Prepare environment. PULSE_SERVER/PULSE_SINK route chromium's audio into the
+	// recorder's sink; the root path below relies on this inherited env, while the
+	// non-root path re-asserts them in its runuser env allowlist.
 	env := os.Environ()
 	env = append(env,
 		"DISPLAY=:1",
 		"DBUS_SESSION_BUS_ADDRESS=unix:path=/run/dbus/system_bus_socket",
+		"PULSE_SERVER="+pulseServer,
+		"PULSE_SINK="+pulseSink,
 	)
 
 	if runAsRoot {
@@ -118,10 +134,18 @@ func main() {
 	}
 
 	// Build: runuser -u kernel -- env DISPLAY=... DBUS_... XDG_... HOME=... chromium <args>
+	// PULSE_SERVER tells libpulse which daemon socket to connect to; without it
+	// chromium-as-kernel-user can't reach the recorder's PulseAudio instance and
+	// has no audio output at all. PULSE_SINK then selects which sink within that
+	// daemon playback lands on: Chromium's AudioManagerPulse honors it to redirect
+	// playback into KernelOutput (see media/audio/pulse/audio_manager_pulse.cc
+	// GetDefaultOutputDeviceID), which is the sink the recorder captures.
 	inner := []string{
 		"env",
 		"DISPLAY=:1",
 		"DBUS_SESSION_BUS_ADDRESS=unix:path=/run/dbus/system_bus_socket",
+		"PULSE_SERVER=" + pulseServer,
+		"PULSE_SINK=" + pulseSink,
 		"XDG_CONFIG_HOME=/home/kernel/.config",
 		"XDG_CACHE_HOME=/home/kernel/.cache",
 		"HOME=/home/kernel",

server/cmd/config/config.go

@@ -18,6 +18,12 @@ type Config struct {
 	DisplayNum  int    `envconfig:"DISPLAY_NUM" default:"1"`
 	MaxSizeInMB int    `envconfig:"MAX_SIZE_MB" default:"500"`
 	OutputDir   string `envconfig:"OUTPUT_DIR" default:"."`
+	// AudioSource and PulseServer default to empty, i.e. video-only. Setting both
+	// enables audio capture; their values must match the topology defined in
+	// shared/start-pulseaudio.sh (the authority for the sink/source/socket). The
+	// image's supervisor conf sets both.
+	AudioSource string `envconfig:"AUDIO_SOURCE" default:""`
+	PulseServer string `envconfig:"PULSE_SERVER" default:""`
 
 	// Absolute or relative path to the ffmpeg binary. If empty the code falls back to "ffmpeg" on $PATH.
 	PathToFFmpeg string `envconfig:"FFMPEG_PATH" default:"ffmpeg"`
@@ -55,6 +61,8 @@ func (c *Config) LogValue() slog.Value {
 		slog.Int("display_num", c.DisplayNum),
 		slog.Int("max_size_mb", c.MaxSizeInMB),
 		slog.String("output_dir", c.OutputDir),
+		slog.String("audio_source", c.AudioSource),
+		slog.String("pulse_server", c.PulseServer),
 		slog.String("ffmpeg_path", c.PathToFFmpeg),
 		slog.Int("devtools_proxy_port", c.DevToolsProxyPort),
 		slog.Bool("log_cdp_messages", c.LogCDPMessages),

server/cmd/config/config_test.go

@@ -23,6 +23,8 @@ func TestLoad(t *testing.T) {
 				DisplayNum:               1,
 				MaxSizeInMB:              500,
 				OutputDir:                ".",
+				AudioSource:              "",
+				PulseServer:              "",
 				PathToFFmpeg:             "ffmpeg",
 				DevToolsProxyPort:        9222,
 				ScaleToZeroCooldown:      time.Second,
@@ -39,6 +41,8 @@ func TestLoad(t *testing.T) {
 				"DISPLAY_NUM":                "2",
 				"MAX_SIZE_MB":                "250",
 				"OUTPUT_DIR":                 "/tmp",
+				"AUDIO_SOURCE":               "CustomOutput.monitor",
+				"PULSE_SERVER":               "unix:/tmp/pulse/native",
 				"FFMPEG_PATH":                "/usr/local/bin/ffmpeg",
 				"DEVTOOLS_PROXY_PORT":        "9876",
 				"SCALE_TO_ZERO_COOLDOWN":     "5s",
@@ -51,6 +55,8 @@ func TestLoad(t *testing.T) {
 				DisplayNum:               2,
 				MaxSizeInMB:              250,
 				OutputDir:                "/tmp",
+				AudioSource:              "CustomOutput.monitor",
+				PulseServer:              "unix:/tmp/pulse/native",
 				PathToFFmpeg:             "/usr/local/bin/ffmpeg",
 				DevToolsProxyPort:        9876,
 				ScaleToZeroCooldown:      5 * time.Second,
@@ -71,6 +77,8 @@ func TestLoad(t *testing.T) {
 				DisplayNum:               1,
 				MaxSizeInMB:              500,
 				OutputDir:                ".",
+				AudioSource:              "",
+				PulseServer:              "",
 				PathToFFmpeg:             "ffmpeg",
 				DevToolsProxyPort:        7777,
 				ScaleToZeroCooldown:      time.Second,

server/cmd/wrapper/chromium.go

@@ -47,7 +47,6 @@ func applyHeadlessDefaultFlags() {
 		"--hide-crash-restore-bubble",
 		"--hide-scrollbars",
 		"--metrics-recording-only",
-		"--mute-audio",
 		"--no-default-browser-check",
 		"--no-first-run",
 		"--no-sandbox",

server/cmd/wrapper/main.go

@@ -27,6 +27,9 @@ const (
 	dbusSocket     = "/run/dbus/system_bus_socket"
 	defaultDisplay = ":1"
 	defaultIntPort = "9223"
+	// pulseSocket must match the socket path created in shared/start-pulseaudio.sh
+	// (the authority for the audio topology); the wrapper only waits on it here.
+	pulseSocket = "/tmp/pulse/native"
 )
 
 type profile int
@@ -172,11 +175,13 @@ func main() {
 	_ = os.WriteFile(filepath.Join(supervisordLogD, "chromium"), nil, 0o644)
 
 	browserStart := time.Now()
-	startAll(xServer, "dbus", "chromedriver", "chromium")
+	startAll(xServer, "dbus", "chromedriver", "pulseaudio")
 	waitForX(defaultDisplay, 20*time.Second)
 	if prof == profileHeadful {
 		startAll("mutter")
 	}
+	waitForSocket(pulseSocket, 10*time.Second)
+	startAll("chromium")
 	waitForSocket(dbusSocket, 10*time.Second)
 	if prof == profileHeadful && webrtc {
 		startAll("neko")
@@ -223,12 +228,6 @@ func main() {
 		identityDone.Sub(identityStart).Truncate(time.Millisecond),
 		formatProbeDurations(probeDurations))
 
-	// Cosmetic + non-critical services come up off the hot path. Headless has
-	// no audio stack.
-	if prof == profileHeadful {
-		go startAll("pulseaudio")
-	}
-
 	// Re-enable scale-to-zero now that the hot path is up — unless the caller
 	// asked to keep it disabled via ENABLE_STZ=false/0.
 	if stzManaged {