Fix federated joins when the first server in the list is not in the room (#15074)

Previously we would give up upon receiving a 404 from the first server,
instead of trying the rest of the servers in the list.

Signed-off-by: Sean Quah <[email protected]>

Author: squahtx

changelog.d/15074.bugfix

@@ -0,0 +1 @@
+Fix a long-standing bug where federated joins would fail if the first server in the list of servers to try is not in the room.

synapse/federation/federation_client.py

@@ -884,7 +884,7 @@ async def _try_destination_list(
                 if 500 <= e.code < 600:
                     failover = True
 
-                elif e.code == 400 and synapse_error.errcode in failover_errcodes:
+                elif 400 <= e.code < 500 and synapse_error.errcode in failover_errcodes:
                     failover = True
 
                 elif failover_on_unknown_endpoint and self._is_unknown_endpoint(
@@ -999,14 +999,13 @@ async def send_request(destination: str) -> Tuple[str, EventBase, RoomVersion]:
 
             return destination, ev, room_version
 
+        failover_errcodes = {Codes.NOT_FOUND}
         # MSC3083 defines additional error codes for room joins. Unfortunately
         # we do not yet know the room version, assume these will only be returned
         # by valid room versions.
-        failover_errcodes = (
-            (Codes.UNABLE_AUTHORISE_JOIN, Codes.UNABLE_TO_GRANT_JOIN)
-            if membership == Membership.JOIN
-            else None
-        )
+        if membership == Membership.JOIN:
+            failover_errcodes.add(Codes.UNABLE_AUTHORISE_JOIN)
+            failover_errcodes.add(Codes.UNABLE_TO_GRANT_JOIN)
 
         return await self._try_destination_list(
             "make_" + membership,