Fix issues with multiple gets being suspended when a refresh starts

With a normal event, if multiple gets were suspended and a refresh
started, we'd continue the refresh after the first get finished and
set the event which would be the same behaviour as the one it tried
to fix. This is avoided by using a counter that's incremented every time
a context manager is entered around an event and only setting the event
when that counter reaches a zero after everything exited the context mgr

2 files changed, 34 insertions, 6 deletions

diff --git a/bot/exts/info/doc/_cog.py b/bot/exts/info/doc/_cog.py
index 7b9dad135..26694ae55 100644
--- a/bot/exts/info/doc/_cog.py
+++ b/bot/exts/info/doc/_cog.py
@@ -16,7 +16,7 @@ from bot.bot import Bot
 from bot.constants import MODERATION_ROLES, RedirectOutput
 from bot.converters import Inventory, PackageName, ValidURL
 from bot.pagination import LinePaginator
-from bot.utils.lock import lock
+from bot.utils.lock import SharedEvent, lock
 from bot.utils.messages import send_denial, wait_for_deletion
 from bot.utils.scheduling import Scheduler
 from . import PRIORITY_PACKAGES, doc_cache
@@ -70,8 +70,7 @@ class DocCog(commands.Cog):
 
         self.refresh_event = asyncio.Event()
         self.refresh_event.set()
-        self.symbol_get_event = asyncio.Event()
-        self.symbol_get_event.set()
+        self.symbol_get_event = SharedEvent()
 
         self.init_refresh_task = self.bot.loop.create_task(self.init_refresh_inventory())
 
@@ -252,9 +251,8 @@ class DocCog(commands.Cog):
             return None
         self.bot.stats.incr(f"doc_fetches.{symbol_info.package}")
 
-        self.symbol_get_event.clear()
-        markdown = await doc_cache.get(symbol_info)
-        self.symbol_get_event.set()
+        with self.symbol_get_event:
+            markdown = await doc_cache.get(symbol_info)
 
         if markdown is None:
             log.debug(f"Redis cache miss for symbol `{symbol}`.")
diff --git a/bot/utils/lock.py b/bot/utils/lock.py
index 997c653a1..b4bb0ebc7 100644
--- a/bot/utils/lock.py
+++ b/bot/utils/lock.py
@@ -1,3 +1,4 @@
+import asyncio
 import inspect
 import logging
 import types
@@ -18,6 +19,35 @@ _IdCallable = Callable[[function.BoundArgs], _IdCallableReturn]
 ResourceId = Union[Hashable, _IdCallable]
 
 
+class SharedEvent:
+    """
+    Context manager managing an internal event exposed through the wait coro.
+
+    While any code is executing in this context manager, the underyling event will not be set;
+    when all of the holders finish the event will be set.
+    """
+
+    def __init__(self):
+        self._active_count = 0
+        self._event = asyncio.Event()
+        self._event.set()
+
+    def __enter__(self):
+        """Increment the count of the active holders and clear the internal event."""
+        self._active_count += 1
+        self._event.clear()
+
+    def __exit__(self, _exc_type, _exc_val, _exc_tb):  # noqa: ANN001
+        """Decrement the count of the active holders; if 0 is reached set the internal event."""
+        self._active_count -= 1
+        if not self._active_count:
+            self._event.set()
+
+    async def wait(self) -> None:
+        """Wait for all active holders to exit."""
+        await self._event.wait()
+
+
 class LockGuard:
     """
     A context manager which acquires and releases a lock (mutex).