tests/locks: add failing test for try_acquire_lock

This commit proves that there is a bug in LockStore.try_acquire_lock,
and provides a test case that must pass.

Signed-off-by: Sumner Evans <[email protected]>
Co-authored-by: Sean Quah <[email protected]>

tests/storage/databases/main/test_lock.py

@@ -12,6 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+from twisted.internet import defer, reactor
+from twisted.internet.base import ReactorBase
+from twisted.internet.defer import Deferred
+
 from synapse.server import HomeServer
 from synapse.storage.databases.main.lock import _LOCK_TIMEOUT_MS
 
@@ -22,6 +26,56 @@ class LockTestCase(unittest.HomeserverTestCase):
     def prepare(self, reactor, clock, hs: HomeServer):
         self.store = hs.get_datastores().main
 
+    def test_acquire_contention(self):
+        # Track the number of tasks holding the lock.
+        # Should be at most 1.
+        in_lock = 0
+        max_in_lock = 0
+
+        release_lock: "Deferred[None]" = Deferred()
+
+        async def task():
+            nonlocal in_lock
+            nonlocal max_in_lock
+
+            lock = await self.store.try_acquire_lock("name", "key")
+            if not lock:
+                return
+
+            async with lock:
+                in_lock += 1
+                max_in_lock = max(max_in_lock, in_lock)
+
+                # Block to allow other tasks to attempt to take the lock.
+                await release_lock
+
+                in_lock -= 1
+
+        # Start 3 tasks.
+        task1 = defer.ensureDeferred(task())
+        task2 = defer.ensureDeferred(task())
+        task3 = defer.ensureDeferred(task())
+
+        # Give the reactor a kick so that the database transaction returns.
+        self.pump()
+
+        release_lock.callback(None)
+
+        # Run the tasks to completion.
+        # To work around `Linearizer`s using a different reactor to sleep when
+        # contended (#12841), we call `runUntilCurrent` on
+        # `twisted.internet.reactor`, which is a different reactor to that used
+        # by the homeserver.
+        assert isinstance(reactor, ReactorBase)
+        self.get_success(task1)
+        reactor.runUntilCurrent()
+        self.get_success(task2)
+        reactor.runUntilCurrent()
+        self.get_success(task3)
+
+        # At most one task should have held the lock at a time.
+        self.assertEqual(max_in_lock, 1)
+
     def test_simple_lock(self):
         """Test that we can take out a lock and that while we hold it nobody
         else can take it out.