health checks for vc

Signed-off-by: salmonik <raisersoft@proton.me>

1 files changed, 89 insertions, 33 deletions

diff --git a/cogs/voice.py b/cogs/voice.py
index 046c2a8..6f9a269 100644
--- a/cogs/voice.py
+++ b/cogs/voice.py
@@ -24,60 +24,116 @@ def get_voice_manager(bot) -> "VoiceManager":
     return _voice_managers[bot_id]
 
 
+def is_voice_client_alive(voice_client, guild):
+    """Check if a voice client connection is healthy.
+
+    Verifies the client is connected, the channel still exists,
+    and the underlying voice socket is still open.
+    """
+    if voice_client is None:
+        return False
+    if not voice_client.is_connected():
+        return False
+    if voice_client.channel is None:
+        return False
+    if guild.get_channel(voice_client.channel.id) is None:
+        return False
+
+    try:
+        ws = voice_client._connection
+        s = ws.socket
+        if s is None or s.fileno() == -1:
+            return False
+    except Exception:
+        return False
+
+    return True
+
+
 @dataclass
 class VoiceManager:
     bot: Any  # type: ignore[no-any-import]
     _voice_lock: asyncio.Lock = field(default_factory=asyncio.Lock, init=False)
+    _connecting: set[int] = field(default_factory=set, init=False)
 
     async def join_channel(self, channel):
         """Join a voice channel by channel object."""
         async with self._voice_lock:
             guild = channel.guild
+            guild_id = guild.id
+
+            if guild_id in self._connecting:
+                logger.info(f"Connection already in progress for guild {guild_id}, skipping")
+                return None
+
             voice_client = discord.utils.get(self.bot.voice_clients, guild=guild)
 
             if voice_client and voice_client.is_connected():
                 if voice_client.channel == channel:
-                    logger.info(f"Already connected to {channel}")
-                    return voice_client
-                try:
-                    await voice_client.move_to(channel)
-                    logger.info(f"Moved to voice channel: {channel}")
-                    return voice_client
-                except Exception as e:
-                    logger.error(f"Failed to move voice client: {e}")
-                    logger.info("Disconnecting after failed move attempt")
+                    if is_voice_client_alive(voice_client, guild):
+                        logger.info(f"Already connected to {channel}")
+                        return voice_client
+                    logger.warning(f"Stale connection to {channel}, reconnecting")
                     try:
                         await voice_client.disconnect()
                     except Exception:
                         pass
-
-            await asyncio.sleep(0.3)
-            try:
-                voice_client = await channel.connect(timeout=30.0, reconnect=False)
-                logger.info(f"Connected to voice channel: {channel}")
-                return voice_client
-            except discord.ClientException as e:
-                if "Already connected" in str(e):
-                    logger.warning(f"Stale voice state detected, forcing disconnect and reconnect: {e}")
-                    voice_client = discord.utils.get(self.bot.voice_clients, guild=guild)
-                    if voice_client:
+                else:
+                    try:
+                        await voice_client.move_to(channel)
+                        logger.info(f"Moved to voice channel: {channel}")
+                        return voice_client
+                    except Exception as e:
+                        logger.error(f"Failed to move voice client: {e}")
+                        logger.info("Disconnecting after failed move attempt")
                         try:
                             await voice_client.disconnect()
                         except Exception:
                             pass
-                    await asyncio.sleep(0.3)
-                    try:
-                        voice_client = await channel.connect(timeout=30.0, reconnect=False)
-                        logger.info(f"Reconnected to voice channel: {channel}")
-                        return voice_client
-                    except Exception as e2:
-                        logger.error(f"Failed to reconnect after stale cleanup: {e2}")
-                        return None
-                logger.error(f"Failed to connect to voice channel: {e}")
-                return None
-            except Exception as e:
-                logger.error(f"Failed to connect to voice channel: {e}")
-                return None
+            elif voice_client and not voice_client.is_connected():
+                logger.info("Voice client exists but disconnected, cleaning up stale entry")
+                try:
+                    await voice_client.disconnect()
+                except Exception:
+                    pass
+                voice_client = None
+
+            self._connecting.add(guild_id)
+            try:
+                await asyncio.sleep(0.3)
+                try:
+                    voice_client = await channel.connect(timeout=30.0, reconnect=False)
+                    logger.info(f"Connected to voice channel: {channel}")
+                    return voice_client
+                except discord.ClientException as e:
+                    if "Already connected" in str(e):
+                        logger.warning(f"Stale voice state detected, forcing disconnect and reconnect: {e}")
+                        voice_client = discord.utils.get(self.bot.voice_clients, guild=guild)
+                        if voice_client:
+                            try:
+                                await voice_client.disconnect()
+                            except Exception:
+                                pass
+                        else:
+                            try:
+                                channel._state._remove_voice_client(guild.id)
+                            except Exception:
+                                pass
+                        await asyncio.sleep(0.3)
+                        try:
+                            voice_client = await channel.connect(timeout=30.0, reconnect=False)
+                            logger.info(f"Reconnected to voice channel: {channel}")
+                            return voice_client
+                        except Exception as e2:
+                            logger.error(f"Failed to reconnect after stale cleanup: {e2}")
+                            return None
+                    logger.error(f"Failed to connect to voice channel: {e}")
+                    return None
+                except Exception as e:
+                    logger.error(f"Failed to connect to voice channel: {e}")
+                    return None
+            finally:
+                self._connecting.discard(guild_id)
 
     async def join_by_channel_id(self, channel_id):
         """Join a voice channel by channel ID."""