mirror of
https://github.com/mofeng-git/One-KVM.git
synced 2026-01-29 00:51:53 +08:00
perf(video): 优化视频管道性能降低延迟
- 减少 broadcast channel 缓冲区大小以降低延迟 - 将 Mutex 字段改为原子类型减少锁竞争 - 使用本地计数器批量更新统计信息 - 修复 WebRTC 帧持续时间计算基于实际 FPS - 不再自动缩放码率，保留用户配置值
This commit is contained in:
mofeng-git
@@ -143,7 +143,7 @@ impl VideoCapturer {
|
||||
/// Create a new video capturer
|
||||
pub fn new(config: CaptureConfig) -> Self {
|
||||
let (state_tx, state_rx) = watch::channel(CaptureState::Stopped);
|
||||
let (frame_tx, _) = broadcast::channel(64); // Buffer up to 64 frames for software encoding
|
||||
let (frame_tx, _) = broadcast::channel(4); // Reduced from 64 for lower latency
|
||||
|
||||
Self {
|
||||
config,
|
||||
|
||||
@@ -17,7 +17,7 @@
|
||||
//! ```
|
||||
|
||||
use bytes::Bytes;
|
||||
use std::sync::atomic::{AtomicBool, Ordering};
|
||||
use std::sync::atomic::{AtomicBool, AtomicU64, Ordering};
|
||||
use std::sync::Arc;
|
||||
use std::time::{Duration, Instant};
|
||||
use tokio::sync::{broadcast, watch, Mutex, RwLock};
|
||||
@@ -291,13 +291,13 @@ pub struct SharedVideoPipeline {
|
||||
/// YUV420P buffer for turbojpeg decoder output
|
||||
yuv420p_buffer: Mutex<Vec<u8>>,
|
||||
/// Whether the encoder needs YUV420P (true) or NV12 (false)
|
||||
encoder_needs_yuv420p: Mutex<bool>,
|
||||
encoder_needs_yuv420p: AtomicBool,
|
||||
frame_tx: broadcast::Sender<EncodedVideoFrame>,
|
||||
stats: Mutex<SharedVideoPipelineStats>,
|
||||
running: watch::Sender<bool>,
|
||||
running_rx: watch::Receiver<bool>,
|
||||
encode_times: Mutex<Vec<f32>>,
|
||||
sequence: Mutex<u64>,
|
||||
/// Frame sequence counter (atomic for lock-free access)
|
||||
sequence: AtomicU64,
|
||||
/// Atomic flag for keyframe request (avoids lock contention)
|
||||
keyframe_requested: AtomicBool,
|
||||
}
|
||||
@@ -314,7 +314,7 @@ impl SharedVideoPipeline {
|
||||
config.input_format
|
||||
);
|
||||
|
||||
let (frame_tx, _) = broadcast::channel(64); // Increased from 16 for software encoding
|
||||
let (frame_tx, _) = broadcast::channel(8); // Reduced from 64 for lower latency
|
||||
let (running_tx, running_rx) = watch::channel(false);
|
||||
let nv12_size = (config.resolution.width * config.resolution.height * 3 / 2) as usize;
|
||||
let yuv420p_size = nv12_size; // Same size as NV12
|
||||
@@ -328,13 +328,12 @@ impl SharedVideoPipeline {
|
||||
mjpeg_turbo_decoder: Mutex::new(None),
|
||||
nv12_buffer: Mutex::new(vec![0u8; nv12_size]),
|
||||
yuv420p_buffer: Mutex::new(vec![0u8; yuv420p_size]),
|
||||
encoder_needs_yuv420p: Mutex::new(false),
|
||||
encoder_needs_yuv420p: AtomicBool::new(false),
|
||||
frame_tx,
|
||||
stats: Mutex::new(SharedVideoPipelineStats::default()),
|
||||
running: running_tx,
|
||||
running_rx,
|
||||
encode_times: Mutex::new(Vec::with_capacity(100)),
|
||||
sequence: Mutex::new(0),
|
||||
sequence: AtomicU64::new(0),
|
||||
keyframe_requested: AtomicBool::new(false),
|
||||
});
|
||||
|
||||
@@ -527,7 +526,7 @@ impl SharedVideoPipeline {
|
||||
*self.yuv420p_converter.lock().await = yuv420p_converter;
|
||||
*self.mjpeg_decoder.lock().await = mjpeg_decoder;
|
||||
*self.mjpeg_turbo_decoder.lock().await = mjpeg_turbo_decoder;
|
||||
*self.encoder_needs_yuv420p.lock().await = needs_yuv420p;
|
||||
self.encoder_needs_yuv420p.store(needs_yuv420p, Ordering::Release);
|
||||
|
||||
Ok(())
|
||||
}
|
||||
@@ -598,7 +597,7 @@ impl SharedVideoPipeline {
|
||||
*self.yuv420p_converter.lock().await = None;
|
||||
*self.mjpeg_decoder.lock().await = None;
|
||||
*self.mjpeg_turbo_decoder.lock().await = None;
|
||||
*self.encoder_needs_yuv420p.lock().await = false;
|
||||
self.encoder_needs_yuv420p.store(false, Ordering::Release);
|
||||
|
||||
info!("Switched to {} codec", codec);
|
||||
Ok(())
|
||||
@@ -625,6 +624,13 @@ impl SharedVideoPipeline {
|
||||
let mut fps_frame_count: u64 = 0;
|
||||
let mut running_rx = pipeline.running_rx.clone();
|
||||
|
||||
// Local counters for batch stats update (reduce lock contention)
|
||||
let mut local_frames_encoded: u64 = 0;
|
||||
let mut local_bytes_encoded: u64 = 0;
|
||||
let mut local_keyframes: u64 = 0;
|
||||
let mut local_errors: u64 = 0;
|
||||
let mut local_dropped: u64 = 0;
|
||||
|
||||
// Track when we last had subscribers for auto-stop feature
|
||||
let mut no_subscribers_since: Option<Instant> = None;
|
||||
let grace_period = Duration::from_secs(AUTO_STOP_GRACE_PERIOD_SECS);
|
||||
@@ -642,8 +648,6 @@ impl SharedVideoPipeline {
|
||||
result = frame_rx.recv() => {
|
||||
match result {
|
||||
Ok(video_frame) => {
|
||||
pipeline.stats.lock().await.frames_captured += 1;
|
||||
|
||||
let subscriber_count = pipeline.frame_tx.receiver_count();
|
||||
|
||||
if subscriber_count == 0 {
|
||||
@@ -676,32 +680,15 @@ impl SharedVideoPipeline {
|
||||
}
|
||||
}
|
||||
|
||||
let start = Instant::now();
|
||||
|
||||
match pipeline.encode_frame(&video_frame, frame_count).await {
|
||||
Ok(Some(encoded_frame)) => {
|
||||
let encode_time = start.elapsed().as_secs_f32() * 1000.0;
|
||||
let _ = pipeline.frame_tx.send(encoded_frame.clone());
|
||||
|
||||
let is_keyframe = encoded_frame.is_keyframe;
|
||||
|
||||
// Update stats
|
||||
{
|
||||
let mut s = pipeline.stats.lock().await;
|
||||
s.frames_encoded += 1;
|
||||
s.bytes_encoded += encoded_frame.data.len() as u64;
|
||||
if is_keyframe {
|
||||
s.keyframes_encoded += 1;
|
||||
}
|
||||
|
||||
let mut times = pipeline.encode_times.lock().await;
|
||||
times.push(encode_time);
|
||||
if times.len() > 100 {
|
||||
times.remove(0);
|
||||
}
|
||||
if !times.is_empty() {
|
||||
s.avg_encode_time_ms = times.iter().sum::<f32>() / times.len() as f32;
|
||||
}
|
||||
// Update local counters (no lock)
|
||||
local_frames_encoded += 1;
|
||||
local_bytes_encoded += encoded_frame.data.len() as u64;
|
||||
if encoded_frame.is_keyframe {
|
||||
local_keyframes += 1;
|
||||
}
|
||||
|
||||
frame_count += 1;
|
||||
@@ -710,19 +697,35 @@ impl SharedVideoPipeline {
|
||||
Ok(None) => {}
|
||||
Err(e) => {
|
||||
error!("Encoding failed: {}", e);
|
||||
pipeline.stats.lock().await.errors += 1;
|
||||
local_errors += 1;
|
||||
}
|
||||
}
|
||||
|
||||
// Batch update stats every second (reduces lock contention)
|
||||
if last_fps_time.elapsed() >= Duration::from_secs(1) {
|
||||
let mut s = pipeline.stats.lock().await;
|
||||
s.current_fps = fps_frame_count as f32 / last_fps_time.elapsed().as_secs_f32();
|
||||
let current_fps = fps_frame_count as f32 / last_fps_time.elapsed().as_secs_f32();
|
||||
fps_frame_count = 0;
|
||||
last_fps_time = Instant::now();
|
||||
|
||||
// Single lock acquisition for all stats
|
||||
let mut s = pipeline.stats.lock().await;
|
||||
s.frames_encoded += local_frames_encoded;
|
||||
s.bytes_encoded += local_bytes_encoded;
|
||||
s.keyframes_encoded += local_keyframes;
|
||||
s.errors += local_errors;
|
||||
s.frames_dropped += local_dropped;
|
||||
s.current_fps = current_fps;
|
||||
|
||||
// Reset local counters
|
||||
local_frames_encoded = 0;
|
||||
local_bytes_encoded = 0;
|
||||
local_keyframes = 0;
|
||||
local_errors = 0;
|
||||
local_dropped = 0;
|
||||
}
|
||||
}
|
||||
Err(broadcast::error::RecvError::Lagged(n)) => {
|
||||
pipeline.stats.lock().await.frames_dropped += n;
|
||||
local_dropped += n;
|
||||
}
|
||||
Err(broadcast::error::RecvError::Closed) => {
|
||||
break;
|
||||
@@ -762,7 +765,7 @@ impl SharedVideoPipeline {
|
||||
let mut mjpeg_turbo_decoder = self.mjpeg_turbo_decoder.lock().await;
|
||||
let mut nv12_converter = self.nv12_converter.lock().await;
|
||||
let mut yuv420p_converter = self.yuv420p_converter.lock().await;
|
||||
let needs_yuv420p = *self.encoder_needs_yuv420p.lock().await;
|
||||
let needs_yuv420p = self.encoder_needs_yuv420p.load(Ordering::Acquire);
|
||||
let mut encoder_guard = self.encoder.lock().await;
|
||||
|
||||
let encoder = encoder_guard.as_mut().ok_or_else(|| {
|
||||
@@ -835,11 +838,7 @@ impl SharedVideoPipeline {
|
||||
let encoded = frames.into_iter().next().unwrap();
|
||||
let is_keyframe = encoded.key == 1;
|
||||
|
||||
let sequence = {
|
||||
let mut seq = self.sequence.lock().await;
|
||||
*seq += 1;
|
||||
*seq
|
||||
};
|
||||
let sequence = self.sequence.fetch_add(1, Ordering::Relaxed) + 1;
|
||||
|
||||
// Debug log for H265 encoded frame
|
||||
if codec == VideoEncoderType::H265 && (is_keyframe || frame_count % 30 == 1) {
|
||||
|
||||
Reference in New Issue
Block a user