Fix logging context warnings when losing replication connection

changelog.d/10984.misc

@@ -0,0 +1 @@
+Fix spurious warnings about losing the logging context on the `ReplicationCommandHandler` when losing the replication connection.

synapse/replication/tcp/protocol.py

@@ -182,9 +182,13 @@ def __init__(self, clock: Clock, handler: "ReplicationCommandHandler"):
 
         # a logcontext which we use for processing incoming commands. We declare it as a
         # background process so that the CPU stats get reported to prometheus.
-        self._logging_context = BackgroundProcessLoggingContext(
-            "replication-conn", self.conn_id
-        )
+        with PreserveLoggingContext():
+            # thanks to `PreserveLoggingContext()`, the new logcontext is guaranteed to
+            # capture the sentinel context as its containing context and won't prevent
+            # GC of / unintentionally reactivate what would be the current context.
+            self._logging_context = BackgroundProcessLoggingContext(
+                "replication-conn", self.conn_id
+            )
 
     def connectionMade(self):
         logger.info("[%s] Connection established", self.id())
@@ -434,8 +438,12 @@ def on_connection_closed(self):
         if self.transport:
             self.transport.unregisterProducer()
 
-        # mark the logging context as finished
-        self._logging_context.__exit__(None, None, None)
+        # mark the logging context as finished by triggering `__exit__()`
+        with PreserveLoggingContext():
+            with self._logging_context:
+                pass
+            # the sentinel context is now active, which may not be correct.
+            # PreserveLoggingContext() will restore the correct logging context.
 
     def __str__(self):
         addr = None

synapse/replication/tcp/redis.py

@@ -100,9 +100,13 @@ def __init__(self, *args, **kwargs):
 
         # a logcontext which we use for processing incoming commands. We declare it as a
         # background process so that the CPU stats get reported to prometheus.
-        self._logging_context = BackgroundProcessLoggingContext(
-            "replication_command_handler"
-        )
+        with PreserveLoggingContext():
+            # thanks to `PreserveLoggingContext()`, the new logcontext is guaranteed to
+            # capture the sentinel context as its containing context and won't prevent
+            # GC of / unintentionally reactivate what would be the current context.
+            self._logging_context = BackgroundProcessLoggingContext(
+                "replication_command_handler"
+            )
 
     def connectionMade(self):
         logger.info("Connected to redis")
@@ -182,8 +186,12 @@ def connectionLost(self, reason):
         super().connectionLost(reason)
         self.synapse_handler.lost_connection(self)
 
-        # mark the logging context as finished
-        self._logging_context.__exit__(None, None, None)
+        # mark the logging context as finished by triggering `__exit__()`
+        with PreserveLoggingContext():
+            with self._logging_context:
+                pass
+            # the sentinel context is now active, which may not be correct.
+            # PreserveLoggingContext() will restore the correct logging context.
 
     def send_command(self, cmd: Command):
         """Send a command if connection has been established.