hopefully does not lose keyboard control any more for users

streaming-agent/gst_agent.py

@@ -35,7 +35,7 @@ from gi.repository import Gst, GstWebRTC, GstSdp, GLib
 import requests
 import websockets
 
-logging.basicConfig(level=logging.INFO, format='[gst_agent] %(message)s', stream=sys.stderr)
+logging.basicConfig(level=logging.INFO, format='%(asctime)s [gst_agent] %(message)s', stream=sys.stderr)
 log = logging.getLogger(__name__)
 
 # ── Configuration ─────────────────────────────────────────────────────────────
@@ -345,16 +345,33 @@ class CSSStreamAgent:
 def css_watchdog():
     # If cstrike_win64.exe crashes (e.g. Wine c0000409 stack overrun), the
     # stream keeps running but shows a black screen forever with no recovery.
-    # This watchdog detects the crash and relaunches CSS directly via revLoader,
-    # the same way start.sh does on initial boot. gst_agent keeps running and
-    # will simply start capturing the relaunched game once it renders.
+    # This watchdog detects the crash and triggers a restart.
+    #
+    # IMPORTANT: this must NOT relaunch CSS directly itself. Doing so
+    # (a bare `wine revLoader.exe` Popen call) created a second, uncoordinated
+    # restart authority alongside the signaling server's own kill+relaunch
+    # logic (manual button / scheduled 5h restart). If both fired close
+    # together, two separate Wine processes could end up racing to launch
+    # CSS, producing duplicate/orphaned windows — xdotool's window search
+    # would then target an arbitrary one, causing keyboard input to land on
+    # a stale/backgrounded window while video (which captures the whole X
+    # display, not a specific window) kept working fine. That mismatch is
+    # exactly the "mouse/video fine, keyboard dead" symptom that was
+    # reported after a watchdog relaunch overlapped with a manual restart.
+    #
+    # So instead we just notify the signaling server, which has the single
+    # authoritative kill+relaunch implementation (performCssRestart) shared
+    # by the manual restart button, the scheduled 5-hour restart, and this
+    # watchdog — never more than one restart in flight at a time.
     #
     # We only start watching once CSS has been seen alive at least once, to
     # avoid racing its initial startup. Require several consecutive misses to
-    # avoid false positives during a normal controller-triggered relaunch
-    # (the signaling server kills+relaunches CSS on controller change).
-    home = os.path.expanduser('~')
-    revloader = os.path.join(home, 'Counter-Strike Source', 'revLoader.exe')
+    # avoid false positives during a normal controller-triggered handoff.
+    restart_url = SIG_URL.replace('ws://', 'http://').replace('wss://', 'https://')
+    if not restart_url.endswith('/'):
+        restart_url += '/'
+    restart_url += 'webclient-force-restart'
+
     seen_alive = False
     missing_count = 0
     while True:
@@ -381,17 +398,14 @@ def css_watchdog():
         # 4 consecutive misses (~20s) — long enough to not collide with the
         # signaling server's own kill+relaunch cycle on controller change.
         if missing_count >= 4:
-            log.error('CSS appears crashed — relaunching revLoader')
+            log.error('CSS appears crashed — requesting restart from signaling server')
             try:
-                env = dict(os.environ, DISPLAY=DISPLAY)
-                subprocess.Popen(['wine', revloader], env=env,
-                                 stdout=subprocess.DEVNULL,
-                                 stderr=subprocess.DEVNULL)
-                log.info('revLoader relaunched by watchdog')
+                resp = requests.post(restart_url, timeout=5)
+                log.info(f'restart request sent to {restart_url}, status={resp.status_code}')
             except Exception as e:
-                log.error(f'watchdog relaunch failed: {e}')
-            # Reset so we give the relaunched game time to come up before
-            # counting misses again.
+                log.error(f'watchdog restart request failed: {e}')
+            # Reset so we give the restart time to complete before counting
+            # misses again.
             seen_alive = False
             missing_count = 0