From 1a282f5ec6dc31bb0da12e2ef8877e8acc5dc324 Mon Sep 17 00:00:00 2001
From: Stan Grams <stanislawgrams@gmail.com>
Date: Mon, 23 Feb 2026 13:24:55 +0100
Subject: [PATCH] [fix](trx-server): prevent audio thread crash on ALSA EPIPE

CPAL error callbacks can fire millions of times per second on ALSA
EPIPE (errno -32). Previously each invocation did a string allocation
and mutex lock, saturating CPU and eventually crashing the server.

- Use atomic swap in both input and output error callbacks so only the
  first error fires the expensive log+notify path; all subsequent hits
  cost a single atomic op and return immediately.
- Replace stream.play()? with explicit error handling in run_playback
  so a play failure triggers stream recreation instead of permanently
  terminating the playback thread.

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Signed-off-by: Stan Grams <stanislawgrams@gmail.com>
---
 src/trx-server/src/audio.rs | 35 ++++++++++++++++++++++++-----------
 1 file changed, 24 insertions(+), 11 deletions(-)

diff --git a/src/trx-server/src/audio.rs b/src/trx-server/src/audio.rs
index c2bed3d..b3a86f6 100644
--- a/src/trx-server/src/audio.rs
+++ b/src/trx-server/src/audio.rs
@@ -317,9 +317,13 @@ fn run_capture(
                 let stream_failed = stream_failed.clone();
                 let stream_err_tx = stream_err_tx.clone();
                 move |err| {
-                    input_err_logger.log(&err.to_string());
-                    stream_failed.store(true, Ordering::SeqCst);
-                    let _ = stream_err_tx.try_send(());
+                    // swap ensures only the first error does expensive work;
+                    // subsequent callbacks (can fire millions/s on ALSA EPIPE)
+                    // return immediately after a single atomic op.
+                    if !stream_failed.swap(true, Ordering::SeqCst) {
+                        input_err_logger.log(&err.to_string());
+                        let _ = stream_err_tx.try_send(());
+                    }
                 }
             },
             None,
@@ -503,9 +507,13 @@ fn run_playback(
                 let stream_failed = stream_failed.clone();
                 let stream_err_tx = stream_err_tx.clone();
                 move |err| {
-                    output_err_logger.log(&err.to_string());
-                    stream_failed.store(true, Ordering::SeqCst);
-                    let _ = stream_err_tx.try_send(());
+                    // swap ensures only the first error does expensive work;
+                    // subsequent callbacks (can fire millions/s on ALSA EPIPE)
+                    // return immediately after a single atomic op.
+                    if !stream_failed.swap(true, Ordering::SeqCst) {
+                        output_err_logger.log(&err.to_string());
+                        let _ = stream_err_tx.try_send(());
+                    }
                 }
             },
             None,
@@ -521,10 +529,12 @@ fn run_playback(
             }
         };
 
-        if !playing {
-            // stay paused until packets arrive
-        } else {
-            stream.play()?;
+        if playing {
+            if let Err(e) = stream.play() {
+                warn!("Audio playback: stream.play failed, recreating: {}", e);
+                std::thread::sleep(AUDIO_STREAM_RECOVERY_DELAY);
+                continue;
+            }
         }
 
         loop {
@@ -544,7 +554,10 @@ fn run_playback(
             match rx.try_recv() {
                 Ok(packet) => {
                     if !playing {
-                        stream.play()?;
+                        if let Err(e) = stream.play() {
+                            warn!("Audio playback: stream.play failed, recreating: {}", e);
+                            break;
+                        }
                         playing = true;
                         info!("Audio playback: started");
                     }