somewhat stable

cht/stream/processor.py

@@ -22,7 +22,7 @@ import socket
 import time
 from pathlib import Path
 from queue import Queue, Empty
-from threading import Thread
+from threading import Thread, Event
 
 from cht.config import (
     AUDIO_EXTRACT_INTERVAL,
@@ -43,7 +43,7 @@ class SessionProcessor:
         self.frames_dir = session_dir / "frames"
         self.audio_dir = session_dir / "audio"
 
-        self._stop_flags: set[str] = set()
+        self._stop_event = Event()
         self._threads: dict[str, Thread] = {}
         self._on_new_frames = None
         self._on_new_audio = None
@@ -289,26 +289,26 @@ class SessionProcessor:
         socket_path = DATA_DIR / "scene.sock"
 
         # Wait for the socket to appear (server creates it on session start).
-        while "stop" not in self._stop_flags:
+        while not self._stop_event.is_set():
             if socket_path.exists():
                 break
             time.sleep(0.5)
-        if "stop" in self._stop_flags:
+        if self._stop_event.is_set():
             return
 
-        while "stop" not in self._stop_flags:
+        while not self._stop_event.is_set():
             try:
                 self._run_scene_session(socket_path, threshold)
             except Exception:
                 log.exception("Scene detector error")
-            if "stop" in self._stop_flags:
+            if self._stop_event.is_set():
                 break
             # If the socket is gone, the session ended — don't retry.
             if not socket_path.exists():
                 log.info("Scene detector: socket gone, session ended")
                 break
             log.info("Scene detector: reconnecting in 2s...")
-            time.sleep(2.0)
+            self._stop_event.wait(timeout=2.0)
 
         log.info("Scene detector stopped")
 
@@ -333,7 +333,7 @@ class SessionProcessor:
         # Thread: socket → ffmpeg stdin
         def _feed_stdin():
             try:
-                while "stop" not in self._stop_flags:
+                while not self._stop_event.is_set():
                     data = sock.recv(65536)
                     if not data:
                         break
@@ -411,10 +411,16 @@ class SessionProcessor:
         self._threads["audio_extractor"] = t
 
     def stop(self):
-        self._stop_flags.add("stop")
+        self._stop_event.set()
         for name, proc in getattr(self, "_procs", {}).items():
             ff.stop_proc(proc, timeout=3)
-        self._procs = {} if hasattr(self, "_procs") else {}
+        self._procs = {}
+        # Join all threads so caller knows they're done before starting a new session
+        for name, t in list(self._threads.items()):
+            t.join(timeout=5)
+            if t.is_alive():
+                log.warning("Thread %s still alive after stop timeout", name)
+        self._threads.clear()
 
     def _has_audio_stream(self, seg: Path) -> bool:
         try:
@@ -441,9 +447,7 @@ class SessionProcessor:
         chunk_num = 0
         current_source = None
 
-        while "stop" not in self._stop_flags:
-            time.sleep(AUDIO_EXTRACT_INTERVAL)
-
+        while not self._stop_event.wait(timeout=AUDIO_EXTRACT_INTERVAL):
             source = self._find_audio_source()
             if not source:
                 continue
@@ -461,6 +465,16 @@ class SessionProcessor:
             if safe_duration is None or safe_duration <= 0:
                 continue
 
+            # Fail-safe: processed_time can accumulate past the file if the
+            # source was recreated (e.g. server restarted same session).
+            if processed_time > safe_duration:
+                log.warning(
+                    "Audio extractor: processed_time %.1fs > file duration %.1fs — resetting",
+                    processed_time, safe_duration,
+                )
+                processed_time = 0.0
+                chunk_num = 0
+
             process_to = safe_duration - AUDIO_SAFETY_MARGIN
             if process_to <= processed_time + 1.0:
                 continue