fix(backend): listen connections reconnect after a dropped connection or a postgres restart and catch up on what they missed

backend/src/watcher/runner.py

@@ -6,7 +6,7 @@ import asyncpg
 import uvloop
 
 from dependencies.container import container
-from utils.env import env
+from utils.listen import Listener
 from utils.logging import logger, setup_logging
 from utils.read.bots import BOTS_CHANGED_CHANNEL
 from utils.storage import ContentAddressedStorage
@@ -24,19 +24,13 @@ async def _retry_loop(registry: BotRegistry) -> None:
             await registry.sync()
 
 
-async def _listen_changes(
-    registry: BotRegistry, tasks: set[asyncio.Task]
-) -> asyncpg.Connection:
-    def on_change(
-        _conn: asyncpg.Connection, _pid: int, _channel: str, _payload: str
-    ) -> None:
+def _change_listener(registry: BotRegistry, tasks: set[asyncio.Task]) -> Listener:
+    def on_change(_payload: str) -> None:
         task = asyncio.create_task(registry.sync())
         tasks.add(task)
         task.add_done_callback(tasks.discard)
 
-    conn = await asyncpg.connect(dsn=env.db.connection_url)
-    await conn.add_listener(BOTS_CHANGED_CHANNEL, on_change)
-    return conn
+    return Listener("watcher", {BOTS_CHANGED_CHANNEL: on_change}, lambda: on_change(""))
 
 
 async def runner() -> None:
@@ -44,7 +38,7 @@ async def runner() -> None:
     storage = await container.get(ContentAddressedStorage)
     registry = BotRegistry(pool, storage)
     tasks: set[asyncio.Task] = set()
-    listen_conn: asyncpg.Connection | None = None
+    listener = _change_listener(registry, tasks)
     stop = asyncio.Event()
     asyncio.get_running_loop().add_signal_handler(signal.SIGTERM, stop.set)
     retry = asyncio.create_task(_retry_loop(registry))
@@ -52,16 +46,14 @@ async def runner() -> None:
         await registry.sync()
         if not registry.running:
             logger.warning("[yellow]No bots yet. Add a bot token in the web UI.[/]")
-        listen_conn = await _listen_changes(registry, tasks)
+        await listener.start()
         logger.info("[green]Bot watcher running.[/]")
         await stop.wait()
     finally:
         retry.cancel()
         with contextlib.suppress(asyncio.CancelledError):
             await retry
-        if listen_conn is not None:
-            with contextlib.suppress(Exception):
-                await listen_conn.close()
+        await listener.stop()
         await registry.close()
         await container.close()