Faster joins: Support for calling `/federation/v1/state` (#12013)

This is an endpoint that we have server-side support for, but no client-side support. It's going to be useful for resyncing partial-stated rooms, so let's introduce it.

1 files changed, 81 insertions, 12 deletions

diff --git a/synapse/federation/federation_client.py b/synapse/federation/federation_client.py
index 48c90bf0bb..c2997997da 100644
--- a/synapse/federation/federation_client.py
+++ b/synapse/federation/federation_client.py
@@ -419,26 +419,90 @@ class FederationClient(FederationBase):
 
         return state_event_ids, auth_event_ids
 
+    async def get_room_state(
+        self,
+        destination: str,
+        room_id: str,
+        event_id: str,
+        room_version: RoomVersion,
+    ) -> Tuple[List[EventBase], List[EventBase]]:
+        """Calls the /state endpoint to fetch the state at a particular point
+        in the room.
+
+        Any invalid events (those with incorrect or unverifiable signatures or hashes)
+        are filtered out from the response, and any duplicate events are removed.
+
+        (Size limits and other event-format checks are *not* performed.)
+
+        Note that the result is not ordered, so callers must be careful to process
+        the events in an order that handles dependencies.
+
+        Returns:
+            a tuple of (state events, auth events)
+        """
+        result = await self.transport_layer.get_room_state(
+            room_version,
+            destination,
+            room_id,
+            event_id,
+        )
+        state_events = result.state
+        auth_events = result.auth_events
+
+        # we may as well filter out any duplicates from the response, to save
+        # processing them multiple times. (In particular, events may be present in
+        # `auth_events` as well as `state`, which is redundant).
+        #
+        # We don't rely on the sort order of the events, so we can just stick them
+        # in a dict.
+        state_event_map = {event.event_id: event for event in state_events}
+        auth_event_map = {
+            event.event_id: event
+            for event in auth_events
+            if event.event_id not in state_event_map
+        }
+
+        logger.info(
+            "Processing from /state: %d state events, %d auth events",
+            len(state_event_map),
+            len(auth_event_map),
+        )
+
+        valid_auth_events = await self._check_sigs_and_hash_and_fetch(
+            destination, auth_event_map.values(), room_version
+        )
+
+        valid_state_events = await self._check_sigs_and_hash_and_fetch(
+            destination, state_event_map.values(), room_version
+        )
+
+        return valid_state_events, valid_auth_events
+
     async def _check_sigs_and_hash_and_fetch(
         self,
         origin: str,
         pdus: Collection[EventBase],
         room_version: RoomVersion,
     ) -> List[EventBase]:
-        """Takes a list of PDUs and checks the signatures and hashes of each
-        one. If a PDU fails its signature check then we check if we have it in
-        the database and if not then request if from the originating server of
-        that PDU.
+        """Checks the signatures and hashes of a list of events.
+
+        If a PDU fails its signature check then we check if we have it in
+        the database, and if not then request it from the sender's server (if that
+        is different from `origin`). If that still fails, the event is omitted from
+        the returned list.
 
         If a PDU fails its content hash check then it is redacted.
 
-        The given list of PDUs are not modified, instead the function returns
+        Also runs each event through the spam checker; if it fails, redacts the event
+        and flags it as soft-failed.
+
+        The given list of PDUs are not modified; instead the function returns
         a new list.
 
         Args:
-            origin
-            pdu
-            room_version
+            origin: The server that sent us these events
+            pdus: The events to be checked
+            room_version: the version of the room these events are in
 
         Returns:
             A list of PDUs that have valid signatures and hashes.
@@ -469,11 +533,16 @@ class FederationClient(FederationBase):
         origin: str,
         room_version: RoomVersion,
     ) -> Optional[EventBase]:
-        """Takes a PDU and checks its signatures and hashes. If the PDU fails
-        its signature check then we check if we have it in the database and if
-        not then request if from the originating server of that PDU.
+        """Takes a PDU and checks its signatures and hashes.
+
+        If the PDU fails its signature check then we check if we have it in the
+        database; if not, we then request it from sender's server (if that is not the
+        same as `origin`). If that still fails, we return None.
+
+        If the PDU fails its content hash check, it is redacted.
 
-        If then PDU fails its content hash check then it is redacted.
+        Also runs the event through the spam checker; if it fails, redacts the event
+        and flags it as soft-failed.
 
         Args:
             origin