Have all instances correctly respond to REPLICATE command. (#7475)

Before all streams were only written to from master, so only master needed to respond to `REPLICATE` commands.

Before all instances wrote to the cache invalidation stream, but didn't respond to `REPLICATE`. This was a bug, which could lead to missed rows from cache invalidation stream if an instance is restarted, however all the caches would be empty in that case so it wasn't a problem.

1 files changed, 45 insertions, 10 deletions

diff --git a/synapse/replication/tcp/handler.py b/synapse/replication/tcp/handler.py
index e6a50aa74e..acfa66a7a8 100644
--- a/synapse/replication/tcp/handler.py
+++ b/synapse/replication/tcp/handler.py
@@ -36,7 +36,12 @@ from synapse.replication.tcp.commands import (
     UserSyncCommand,
 )
 from synapse.replication.tcp.protocol import AbstractConnection
-from synapse.replication.tcp.streams import STREAMS_MAP, Stream
+from synapse.replication.tcp.streams import (
+    STREAMS_MAP,
+    CachesStream,
+    FederationStream,
+    Stream,
+)
 from synapse.util.async_helpers import Linearizer
 
 logger = logging.getLogger(__name__)
@@ -73,6 +78,26 @@ class ReplicationCommandHandler:
             stream.NAME: stream(hs) for stream in STREAMS_MAP.values()
         }  # type: Dict[str, Stream]
 
+        # List of streams that this instance is the source of
+        self._streams_to_replicate = []  # type: List[Stream]
+
+        for stream in self._streams.values():
+            if stream.NAME == CachesStream.NAME:
+                # All workers can write to the cache invalidation stream.
+                self._streams_to_replicate.append(stream)
+                continue
+
+            # Only add any other streams if we're on master.
+            if hs.config.worker_app is not None:
+                continue
+
+            if stream.NAME == FederationStream.NAME and hs.config.send_federation:
+                # We only support federation stream if federation sending
+                # has been disabled on the master.
+                continue
+
+            self._streams_to_replicate.append(stream)
+
         self._position_linearizer = Linearizer(
             "replication_position", clock=self._clock
         )
@@ -150,6 +175,16 @@ class ReplicationCommandHandler:
             port = hs.config.worker_replication_port
             hs.get_reactor().connectTCP(host, port, self._factory)
 
+    def get_streams(self) -> Dict[str, Stream]:
+        """Get a map from stream name to all streams.
+        """
+        return self._streams
+
+    def get_streams_to_replicate(self) -> List[Stream]:
+        """Get a list of streams that this instances replicates.
+        """
+        return self._streams_to_replicate
+
     async def on_REPLICATE(self, conn: AbstractConnection, cmd: ReplicateCommand):
         self.send_positions_to_connection(conn)
 
@@ -158,15 +193,15 @@ class ReplicationCommandHandler:
         the connection.
         """
 
-        # We only want to announce positions by the writer of the streams.
-        # Currently this is just the master process.
-        if not self._is_master:
-            return
-
-        for stream_name, stream in self._streams.items():
-            current_token = stream.current_token(self._instance_name)
-            conn.send_command(
-                PositionCommand(stream_name, self._instance_name, current_token)
+        # We respond with current position of all streams this instance
+        # replicates.
+        for stream in self.get_streams_to_replicate():
+            self.send_command(
+                PositionCommand(
+                    stream.NAME,
+                    self._instance_name,
+                    stream.current_token(self._instance_name),
+                )
             )
 
     async def on_USER_SYNC(self, conn: AbstractConnection, cmd: UserSyncCommand):