Pass through list of room hosts from room alias query to federation so that it can retry against different room hosts

3 files changed, 22 insertions, 15 deletions

diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py
index eb36ec040b..9923b3fc0d 100644
--- a/synapse/federation/federation_client.py
+++ b/synapse/federation/federation_client.py
@@ -264,7 +264,9 @@ class FederationClient(FederationBase):
 
                 logger.debug("Got response to make_join: %s", pdu_dict)
 
-                defer.returnValue(self.event_from_pdu_json(pdu_dict))
+                defer.returnValue(
+                    (destination, self.event_from_pdu_json(pdu_dict))
+                )
                 break
             except CodeMessageException:
                 raise
@@ -313,6 +315,7 @@ class FederationClient(FederationBase):
                 defer.returnValue({
                     "state": signed_state,
                     "auth_chain": signed_auth,
+                    "origin": destination,
                 })
             except CodeMessageException:
                 raise
diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py
index 04a4689483..aba266c2bc 100644
--- a/synapse/handlers/federation.py
+++ b/synapse/handlers/federation.py
@@ -273,7 +273,7 @@ class FederationHandler(BaseHandler):
 
     @log_function
     @defer.inlineCallbacks
-    def do_invite_join(self, target_host, room_id, joinee, content, snapshot):
+    def do_invite_join(self, target_hosts, room_id, joinee, content, snapshot):
         """ Attempts to join the `joinee` to the room `room_id` via the
         server `target_host`.
 
@@ -287,8 +287,8 @@ class FederationHandler(BaseHandler):
         """
         logger.debug("Joining %s to %s", joinee, room_id)
 
-        pdu = yield self.replication_layer.make_join(
-            [target_host],
+        origin, pdu = yield self.replication_layer.make_join(
+            target_hosts,
             room_id,
             joinee
         )
@@ -330,11 +330,17 @@ class FederationHandler(BaseHandler):
 
             new_event = builder.build()
 
+            # Try the host we successfully got a response to /make_join/
+            # request first.
+            target_hosts.remove(origin)
+            target_hosts.insert(0, origin)
+
             ret = yield self.replication_layer.send_join(
-                [target_host],
+                target_hosts,
                 new_event
             )
 
+            origin = ret["origin"]
             state = ret["state"]
             auth_chain = ret["auth_chain"]
             auth_chain.sort(key=lambda e: e.depth)
@@ -371,7 +377,7 @@ class FederationHandler(BaseHandler):
                         if e.event_id in auth_ids
                     }
                     yield self._handle_new_event(
-                        target_host, e, auth_events=auth
+                        origin, e, auth_events=auth
                     )
                 except:
                     logger.exception(
@@ -391,7 +397,7 @@ class FederationHandler(BaseHandler):
                         if e.event_id in auth_ids
                     }
                     yield self._handle_new_event(
-                        target_host, e, auth_events=auth
+                        origin, e, auth_events=auth
                     )
                 except:
                     logger.exception(
@@ -406,7 +412,7 @@ class FederationHandler(BaseHandler):
             }
 
             yield self._handle_new_event(
-                target_host,
+                origin,
                 new_event,
                 state=state,
                 current_state=state,
diff --git a/synapse/handlers/room.py b/synapse/handlers/room.py
index 23821d321f..0369b907a5 100644
--- a/synapse/handlers/room.py
+++ b/synapse/handlers/room.py
@@ -389,8 +389,6 @@ class RoomMemberHandler(BaseHandler):
         if not hosts:
             raise SynapseError(404, "No known servers")
 
-        host = hosts[0]
-
         # If event doesn't include a display name, add one.
         yield self.distributor.fire(
             "collect_presencelike_data", joinee, content
@@ -407,12 +405,12 @@ class RoomMemberHandler(BaseHandler):
         })
         event, context = yield self._create_new_client_event(builder)
 
-        yield self._do_join(event, context, room_host=host, do_auth=True)
+        yield self._do_join(event, context, room_hosts=hosts, do_auth=True)
 
         defer.returnValue({"room_id": room_id})
 
     @defer.inlineCallbacks
-    def _do_join(self, event, context, room_host=None, do_auth=True):
+    def _do_join(self, event, context, room_hosts=None, do_auth=True):
         joinee = UserID.from_string(event.state_key)
         # room_id = RoomID.from_string(event.room_id, self.hs)
         room_id = event.room_id
@@ -441,7 +439,7 @@ class RoomMemberHandler(BaseHandler):
 
         if is_host_in_room:
             should_do_dance = False
-        elif room_host:  # TODO: Shouldn't this be remote_room_host?
+        elif room_hosts:  # TODO: Shouldn't this be remote_room_host?
             should_do_dance = True
         else:
             # TODO(markjh): get prev_state from snapshot
@@ -453,7 +451,7 @@ class RoomMemberHandler(BaseHandler):
                 inviter = UserID.from_string(prev_state.user_id)
 
                 should_do_dance = not self.hs.is_mine(inviter)
-                room_host = inviter.domain
+                room_hosts = [inviter.domain]
             else:
                 # return the same error as join_room_alias does
                 raise SynapseError(404, "No known servers")
@@ -461,7 +459,7 @@ class RoomMemberHandler(BaseHandler):
         if should_do_dance:
             handler = self.hs.get_handlers().federation_handler
             yield handler.do_invite_join(
-                room_host,
+                room_hosts,
                 room_id,
                 event.user_id,
                 event.get_dict()["content"],  # FIXME To get a non-frozen dict