viz: proper POLLOUT queue to prevent frame drops

- FrameQueue (max 5) buffers frames when router is slow
- POLLOUT drains pending_frames queue
- No silent drops — full queue increments counter
- Router always registered POLLIN| POLLOUT
- Frame interval 30ms (33 FPS)

scripts/led-matrix-viz.py

@@ -1,18 +1,29 @@
 #!/usr/bin/env python3
 """Listen for MIDI events via UDP and display note activity on LED matrix.
 
-Non-blocking I/O with immediate MIDI processing and fast frame rendering.
+Non-blocking I/O with proper POLLOUT queue, instrumentation, and fast rendering.
 """
 import socket, msgpack, time, os, select, collections
 
 VIZ_PORT = int(os.environ.get("VIZ_PORT", "8082"))
 ROUTER_SOCK = os.environ.get("ROUTER_SOCK", "/var/run/arduino-router.sock")
 DECAY_FRAMES = int(os.environ.get("DECAY_FRAMES", "8"))
-FPS = 30
 
 COLS = 13
 ROWS = 8
 
+frame_interval = 1.0 / 30.0  # 30 FPS target
+next_frame_time = 0.0
+
+_poll = select.poll()
+POLLIN = select.POLLIN
+POLLOUT = select.POLLOUT
+
+SOCK_BUFFER = 4096
+ROUTER_RECONNECT_DELAY = 0.1
+
+_frame_hooks = []  # for metrics
+
 
 def pack_frame(pixels):
     bits = 0
@@ -24,6 +35,27 @@ def pack_frame(pixels):
             (bits >> 64) & 0xFFFFFFFF, (bits >> 96) & 0xFFFFFFFF)
 
 
+class FrameQueue:
+    def __init__(self, maxsize=10):
+        self._q = collections.deque()
+        self._maxsize = maxsize
+
+    def push(self, frame_bytes):
+        if len(self._q) >= self._maxsize:
+            return False  # dropped
+        self._q.append(frame_bytes)
+        return True
+
+    def pop(self):
+        return self._q.popleft() if self._q else None
+
+    def __len__(self):
+        return len(self._q)
+
+    def full(self):
+        return len(self._q) >= self._maxsize
+
+
 def connect_router():
     router = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
     router.setblocking(False)
@@ -35,6 +67,8 @@ def connect_router():
 
 
 def main():
+    global next_frame_time
+
     sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
     sock.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
     sock.bind(("0.0.0.0", VIZ_PORT))
@@ -43,40 +77,44 @@ def main():
     router = connect_router()
     packer = msgpack.Packer()
 
+    _poll.register(sock, POLLIN)
+
     active_notes = {}
     last_frame = None
     buf = b""
     msg_id = 0
-    frame_interval = 1.0 / FPS
-    next_frame = time.monotonic()
-    router_buf = b""
+    next_frame_time = time.monotonic()
+
+    pending_frames = FrameQueue(maxsize=5)
+    router_registered = False
+    router_fd = router.fileno()
+    _poll.register(router, POLLIN | POLLOUT)
+
     frames_sent = 0
     frames_dropped = 0
-
-    poller = select.poll()
-    poller.register(sock, select.POLLIN)
+    router_errors = 0
 
     while True:
-        # Always poll UDP with short timeout
-        timeout = max(0, int((next_frame - time.monotonic()) * 1000))
-        timeout = min(timeout, 10)  # cap at 10ms for responsiveness
+        now = time.monotonic()
+
+        timeout = max(0, int((next_frame_time - now) * 1000))
+        timeout = min(timeout, 5)  # 5ms max to stay responsive
 
         try:
-            events = poller.poll(timeout)
+            events = _poll.poll(timeout)
         except OSError:
             events = []
 
-        # Process all pending UDP data immediately
+        # Process UDP
         for fd, flags in events:
-            if fd == sock.fileno() and flags & select.POLLIN:
+            if fd == sock.fileno() and (flags & POLLIN):
                 try:
                     while True:
-                        data, _ = sock.recvfrom(1024)
+                        data, _ = sock.recvfrom(SOCK_BUFFER)
                         buf += data
                 except BlockingIOError:
                     pass
 
-                # Parse all complete messages
                 while b";" in buf:
                     msg, buf = buf.split(b";", 1)
                     msg = msg.decode().strip()
@@ -91,22 +129,44 @@ def main():
                         elif msg_type == 0x80 or (msg_type == 0x90 and vel == 0):
                             active_notes.pop(note, None)
 
-        # Also drain any pending router responses
-        if router.fileno() >= 0:
-            try:
-                while True:
-                    router_buf += router.recv(4096)
-            except BlockingIOError:
-                pass
+            # Process router readable
+            if fd == router_fd and (flags & POLLIN):
+                try:
+                    while True:
+                        d = router.recv(SOCK_BUFFER)
+                        if not d:
+                            break
+                except BlockingIOError:
+                    pass
 
-        # Render frame when it's time
-        now = time.monotonic()
-        if now >= next_frame:
-            next_frame = now + frame_interval
+            # Process router writable — drain pending frames
+            if fd == router_fd and (flags & POLLOUT):
+                frame = pending_frames.pop()
+                if frame:
+                    try:
+                        router.sendall(frame)
+                        frames_sent += 1
+                    except (BlockingIOError, OSError):
+                        frames_dropped += 1
+                        router_errors += 1
+                        pending_frames.push(frame)
+                        time.sleep(ROUTER_RECONNECT_DELAY)
+                        try:
+                            router.close()
+                        except OSError:
+                            pass
+                        router = connect_router()
+                        router_fd = router.fileno()
+                        _poll.unregister(router)
+                        _poll.register(router, POLLIN | POLLOUT)
+
+        # Render frame when due
+        if now >= next_frame_time:
+            next_frame_time = now + frame_interval
 
             pixels = [[0] * COLS for _ in range(ROWS)]
             expired = []
-            for note, decay in active_notes.items():
+            for note, decay in list(active_notes.items()):
                 if decay <= 0:
                     expired.append(note)
                     continue
@@ -125,19 +185,12 @@ def main():
                 msg_id = (msg_id + 1) % 65535
                 params = [str(v) for v in frame]
                 req = packer.pack([0, msg_id, "draw_frame", params])
-                try:
-                    router.sendall(req)
-                    frames_sent += 1
-                except (BlockingIOError, OSError):
+                if pending_frames.push(req):
+                    pass  # queued
+                else:
                     frames_dropped += 1
-                    # Reconnect on error
-                    try:
-                        router.close()
-                    except OSError:
-                        pass
-                    router = connect_router()
                 last_frame = frame
 
 
 if __name__ == "__main__":
-    main()
+    main()