fix(clock): dedup tasks lock bug removal

polylan_submitter/clock/tasks.py

@@ -12,6 +12,7 @@ acts.
 """
 
 from datetime import timedelta
+from uuid import uuid4
 
 from asgiref.sync import async_to_sync
 from channels.layers import get_channel_layer
@@ -30,13 +31,20 @@ HEARTBEAT_INTERVAL_SECONDS = 1
 # One chain ticks every event with the game enabled - this lock keeps a
 # second chain from starting underneath an already-running one (e.g. two
 # EventGame saves close together). Refreshed every tick, wide enough that a
-# single slow tick doesn't let a second chain slip in.
+# single slow tick doesn't let a second chain slip in. The lock holds the
+# chain id: if it expires anyway (worker down/restarted for a few seconds) and
+# a new chain starts, the old one stops at its next tick (instead of both
+# ticking forever).
 HEARTBEAT_LOCK_KEY = "clock:heartbeat:lock"
 HEARTBEAT_LOCK_TIMEOUT_SECONDS = HEARTBEAT_INTERVAL_SECONDS * 5
 
 
 @task
-def clock_heartbeat_task():
+def clock_heartbeat_task(chain_id: str):
+    if cache.get(HEARTBEAT_LOCK_KEY) != chain_id:
+        # Another chain owns the lock (or it expired), stop this one
+        return
+
     event_games = list(
         EventGame.objects.filter(game__path="clock", enabled=True).select_related(
             "event"
@@ -46,7 +54,8 @@ def clock_heartbeat_task():
     if not event_games:
         # Nothing left to tick - let the chain die. start_clock_heartbeat()
         # starts a fresh one next time an event enables the game.
-        cache.delete(HEARTBEAT_LOCK_KEY)
+        if cache.get(HEARTBEAT_LOCK_KEY) == chain_id:
+            cache.delete(HEARTBEAT_LOCK_KEY)
         return
 
     channel_layer = get_channel_layer()
@@ -67,13 +76,14 @@ def clock_heartbeat_task():
             },
         )
 
-    cache.set(HEARTBEAT_LOCK_KEY, 1, timeout=HEARTBEAT_LOCK_TIMEOUT_SECONDS)
+    cache.set(HEARTBEAT_LOCK_KEY, chain_id, timeout=HEARTBEAT_LOCK_TIMEOUT_SECONDS)
     clock_heartbeat_task.using(
         run_after=timezone.now() + timedelta(seconds=HEARTBEAT_INTERVAL_SECONDS)
-    ).enqueue()
+    ).enqueue(chain_id)
 
 
 def start_clock_heartbeat():
     """Enqueues the heartbeat chain, unless one is already running."""
-    if cache.add(HEARTBEAT_LOCK_KEY, 1, timeout=HEARTBEAT_LOCK_TIMEOUT_SECONDS):
-        clock_heartbeat_task.enqueue()
+    chain_id = uuid4().hex
+    if cache.add(HEARTBEAT_LOCK_KEY, chain_id, timeout=HEARTBEAT_LOCK_TIMEOUT_SECONDS):
+        clock_heartbeat_task.enqueue(chain_id)

polylan_submitter/clock/tests.py

@@ -0,0 +1,57 @@
+from unittest import mock
+
+from django.core.cache import cache
+from django.test import TestCase, override_settings
+
+from events.models import Event
+from games.models import EventGame, Game
+
+from .tasks import HEARTBEAT_LOCK_KEY, clock_heartbeat_task, start_clock_heartbeat
+
+LOCMEM_CACHE = {"default": {"BACKEND": "django.core.cache.backends.locmem.LocMemCache"}}
+
+
+@override_settings(CACHES=LOCMEM_CACHE)
+@mock.patch("clock.tasks.get_channel_layer")
+class ClockHeartbeatTests(TestCase):
+    def setUp(self):
+        cache.clear()
+        event = Event.objects.create(name="Event", path="event")
+        clock = Game.objects.create(name="Clock", path="clock", custom_path=True)
+        self.event_game = EventGame.objects.create(event=event, game=clock)
+        cache.clear()  # post_save signal can start a chain
+
+    @mock.patch("clock.tasks.clock_heartbeat_task")
+    def test_only_one_chain_is_started(self, task, _):
+        start_clock_heartbeat()
+        start_clock_heartbeat()
+
+        task.enqueue.assert_called_once_with(cache.get(HEARTBEAT_LOCK_KEY))
+
+    def test_owning_chain_reschedules_itself(self, channel_layer):
+        channel_layer.return_value.group_send = mock.AsyncMock()
+        cache.set(HEARTBEAT_LOCK_KEY, "chain")
+        with mock.patch("clock.tasks.clock_heartbeat_task") as patched:
+            clock_heartbeat_task.call("chain")
+
+        patched.using.return_value.enqueue.assert_called_once_with("chain")
+        channel_layer.return_value.group_send.assert_awaited_once()
+        self.assertEqual(cache.get(HEARTBEAT_LOCK_KEY), "chain")
+
+    def test_superseded_chain_stops(self, channel_layer):
+        cache.set(HEARTBEAT_LOCK_KEY, "new-chain")
+        with mock.patch("clock.tasks.clock_heartbeat_task") as patched:
+            clock_heartbeat_task.call("old-chain")
+
+        patched.using.assert_not_called()
+        channel_layer.assert_not_called()
+        self.assertEqual(cache.get(HEARTBEAT_LOCK_KEY), "new-chain")
+
+    def test_chain_dies_and_releases_lock_when_game_disabled(self, _):
+        EventGame.objects.filter(pk=self.event_game.pk).update(enabled=False)
+        cache.set(HEARTBEAT_LOCK_KEY, "chain")
+        with mock.patch("clock.tasks.clock_heartbeat_task") as patched:
+            clock_heartbeat_task.call("chain")
+
+        patched.using.assert_not_called()
+        self.assertIsNone(cache.get(HEARTBEAT_LOCK_KEY))

polylan_submitter/games/admin.py

@@ -40,7 +40,7 @@ class GameAdmin(admin.ModelAdmin):
         game_ids = list(
             queryset.filter(steam_app_id__isnull=False).values_list("id", flat=True)
         )
-        refresh_steam_games_task.enqueue(game_ids=game_ids, reschedule=False)
+        refresh_steam_games_task.enqueue(game_ids=game_ids)
         self.message_user(
             request, f"Queued Steam store refresh for {len(game_ids)} game(s)."
         )

polylan_submitter/games/tasks.py

@@ -1,4 +1,5 @@
 from datetime import timedelta
+from uuid import uuid4
 
 from django.conf import settings
 from django.core.cache import cache
@@ -10,7 +11,9 @@ from .services.steam_store import refresh_steam_games
 # Set while a periodic refresh is queued, so there's only ever one loop running,
 # whatever the number of workers/containers trying to start it. It expires on
 # its own if the loop dies (worker killed mid-job, Redis queue flushed, ...),
-# letting the next schedule_steam_refresh() start a new one.
+# letting the next schedule_steam_refresh() start a new one. The guard holds the
+# loop id, so if it expires while the old loop is just late, the old one stops
+# at its next run (instead of 2 loops running side by side).
 STEAM_REFRESH_GUARD_KEY = "games:steam_refresh:scheduled"
 
 
@@ -25,28 +28,34 @@ def _guard_timeout() -> int:
 def schedule_steam_refresh() -> bool:
     """Start the periodic refresh loop right away, unless one is already queued.
     Safe to call from anywhere, any number of times."""
-    if not cache.add(STEAM_REFRESH_GUARD_KEY, 1, timeout=_guard_timeout()):
+    loop_id = uuid4().hex
+    if not cache.add(STEAM_REFRESH_GUARD_KEY, loop_id, timeout=_guard_timeout()):
         return False
 
-    refresh_steam_games_task.enqueue()
+    refresh_steam_games_task.enqueue(loop_id=loop_id)
     return True
 
 
 @task
 def refresh_steam_games_task(
-    game_ids: list[int] | None = None, reschedule: bool = True
+    game_ids: list[int] | None = None, loop_id: str | None = None
 ):
     """Refresh Steam store info, then re-enqueue itself every
     STEAM_STORE_REFRESH_MINUTES (same self-rescheduling idea as
     sync_event_game_scopes_task, no external scheduler needed).
 
-    reschedule=False is a one-off run (e.g. the admin action) that leaves the
-    periodic loop alone."""
+    Without loop_id it's a one-off run (e.g. the admin action) that leaves the
+    periodic loop alone. With loop_id, it only continues while the guard still
+    holds this loop_id."""
+    if loop_id is not None and cache.get(STEAM_REFRESH_GUARD_KEY) != loop_id:
+        # Another loop owns the guard (or it expired), stop this one
+        return
+
     try:
         refresh_steam_games(game_ids)
     finally:
-        if reschedule:
-            cache.set(STEAM_REFRESH_GUARD_KEY, 1, timeout=_guard_timeout())
+        if loop_id is not None:
+            cache.set(STEAM_REFRESH_GUARD_KEY, loop_id, timeout=_guard_timeout())
             refresh_steam_games_task.using(
                 run_after=timezone.now() + steam_refresh_interval()
-            ).enqueue()
+            ).enqueue(loop_id=loop_id)

polylan_submitter/games/tests.py

@@ -9,7 +9,11 @@ from events.models import Event
 
 from .models import EventGame, Game
 from .services.steam_store import refresh_steam_games
-from .tasks import refresh_steam_games_task, schedule_steam_refresh
+from .tasks import (
+    STEAM_REFRESH_GUARD_KEY,
+    refresh_steam_games_task,
+    schedule_steam_refresh,
+)
 
 LOCMEM_CACHE = {"default": {"BACKEND": "django.core.cache.backends.locmem.LocMemCache"}}
 
@@ -83,21 +87,32 @@ class ScheduleSteamRefreshTests(TestCase):
     def test_only_one_loop_is_scheduled(self, task):
         self.assertTrue(schedule_steam_refresh())
         self.assertFalse(schedule_steam_refresh())
-        task.enqueue.assert_called_once_with()
+        task.enqueue.assert_called_once_with(loop_id=cache.get(STEAM_REFRESH_GUARD_KEY))
 
     @mock.patch("games.tasks.refresh_steam_games")
     def test_task_reschedules_itself_even_on_failure(self, refresh):
+        cache.set(STEAM_REFRESH_GUARD_KEY, "loop")
         refresh.side_effect = RuntimeError("boom")
         with mock.patch("games.tasks.refresh_steam_games_task") as patched:
             with self.assertRaises(RuntimeError):
-                refresh_steam_games_task.call()
+                refresh_steam_games_task.call(loop_id="loop")
 
-        patched.using.return_value.enqueue.assert_called_once_with()
+        patched.using.return_value.enqueue.assert_called_once_with(loop_id="loop")
+
+    @mock.patch("games.tasks.refresh_steam_games")
+    def test_superseded_loop_stops(self, refresh):
+        cache.set(STEAM_REFRESH_GUARD_KEY, "new-loop")
+        with mock.patch("games.tasks.refresh_steam_games_task") as patched:
+            refresh_steam_games_task.call(loop_id="old-loop")
+
+        refresh.assert_not_called()
+        patched.using.assert_not_called()
+        self.assertEqual(cache.get(STEAM_REFRESH_GUARD_KEY), "new-loop")
 
     @mock.patch("games.tasks.refresh_steam_games")
     def test_one_off_run_does_not_reschedule(self, refresh):
         with mock.patch("games.tasks.refresh_steam_games_task") as patched:
-            refresh_steam_games_task.call(game_ids=[1], reschedule=False)
+            refresh_steam_games_task.call(game_ids=[1])
 
         refresh.assert_called_once_with([1])
         patched.using.assert_not_called()