Change the way we get missing auth and state events

2 files changed, 106 insertions, 44 deletions

diff --git a/synapse/federation/replication.py b/synapse/federation/replication.py
index 6bfb30b42d..312d69fcaa 100644
--- a/synapse/federation/replication.py
+++ b/synapse/federation/replication.py
@@ -283,6 +283,22 @@ class ReplicationLayer(object):
 
     @defer.inlineCallbacks
     @log_function
+    def get_event_auth(self, destination, context, event_id):
+        res = yield self.transport_layer.get_event_auth(
+            destination, context, event_id,
+        )
+
+        auth_chain = [
+            self.event_from_pdu_json(p, outlier=True)
+            for p in res["auth_chain"]
+        ]
+
+        auth_chain.sort(key=lambda e: e.depth)
+
+        defer.returnValue(auth_chain)
+
+    @defer.inlineCallbacks
+    @log_function
     def on_backfill_request(self, origin, context, versions, limit):
         pdus = yield self.handler.on_backfill_request(
             origin, context, versions, limit
@@ -549,34 +565,34 @@ class ReplicationLayer(object):
         state = None
 
         # We need to make sure we have all the auth events.
-        for e_id, _ in pdu.auth_events:
-            exists = yield self._get_persisted_pdu(
-                origin,
-                e_id,
-                do_auth=False
-            )
-
-            if not exists:
-                try:
-                    logger.debug(
-                        "_handle_new_pdu fetch missing auth event %s from %s",
-                        e_id,
-                        origin,
-                    )
-
-                    yield self.get_pdu(
-                        origin,
-                        event_id=e_id,
-                        outlier=True,
-                    )
-
-                    logger.debug("Processed pdu %s", e_id)
-                except:
-                    logger.warn(
-                        "Failed to get auth event %s from %s",
-                        e_id,
-                        origin
-                    )
+        # for e_id, _ in pdu.auth_events:
+        #     exists = yield self._get_persisted_pdu(
+        #         origin,
+        #         e_id,
+        #         do_auth=False
+        #     )
+        #
+        #     if not exists:
+        #         try:
+        #             logger.debug(
+        #                 "_handle_new_pdu fetch missing auth event %s from %s",
+        #                 e_id,
+        #                 origin,
+        #             )
+        #
+        #             yield self.get_pdu(
+        #                 origin,
+        #                 event_id=e_id,
+        #                 outlier=True,
+        #             )
+        #
+        #             logger.debug("Processed pdu %s", e_id)
+        #         except:
+        #             logger.warn(
+        #                 "Failed to get auth event %s from %s",
+        #                 e_id,
+        #                 origin
+        #             )
 
         # Get missing pdus if necessary.
         if not pdu.outlier:
@@ -626,6 +642,7 @@ class ReplicationLayer(object):
 
         if not backfilled:
             ret = yield self.handler.on_receive_pdu(
+                origin,
                 pdu,
                 backfilled=backfilled,
                 state=state,
diff --git a/synapse/handlers/federation.py b/synapse/handlers/federation.py
index 7903494e08..0863fdb13e 100644
--- a/synapse/handlers/federation.py
+++ b/synapse/handlers/federation.py
@@ -101,7 +101,7 @@ class FederationHandler(BaseHandler):
 
     @log_function
     @defer.inlineCallbacks
-    def on_receive_pdu(self, pdu, backfilled, state=None):
+    def on_receive_pdu(self, origin, pdu, backfilled, state=None):
         """ Called by the ReplicationLayer when we have a new pdu. We need to
         do auth checks and put it through the StateHandler.
         """
@@ -149,14 +149,47 @@ class FederationHandler(BaseHandler):
         # FIXME (erikj): Awful hack to make the case where we are not currently
         # in the room work
         current_state = None
-        if state:
-            is_in_room = yield self.auth.check_host_in_room(
-                event.room_id,
-                self.server_name
+        is_in_room = yield self.auth.check_host_in_room(
+            event.room_id,
+            self.server_name
+        )
+        if not is_in_room:
+            logger.debug("Got event for room we're not in.")
+
+            replication_layer = self.replication_layer
+            auth_chain = yield replication_layer.get_event_auth(
+                origin,
+                context=event.room_id,
+                event_id=event.event_id,
+            )
+
+            current_state = yield replication_layer.get_state_for_context(
+                origin,
+                context=event.room_id,
+                event_id=event.event_id,
             )
-            if not is_in_room:
-                logger.debug("Got event for room we're not in.")
-                current_state = state
+
+            for e in auth_chain:
+                e.outlier = True
+                try:
+                    yield self._handle_new_event(e)
+                    yield self.notifier.on_new_room_event(e)
+                except:
+                    logger.exception(
+                        "Failed to parse auth event %s",
+                        e.event_id,
+                    )
+
+            for e in current_state:
+                e.outlier = True
+                try:
+                    yield self._handle_new_event(e)
+                    yield self.notifier.on_new_room_event(e)
+                except:
+                    logger.exception(
+                        "Failed to parse state event %s",
+                        e.event_id,
+                    )
 
         try:
             yield self._handle_new_event(
@@ -328,18 +361,30 @@ class FederationHandler(BaseHandler):
 
             for e in auth_chain:
                 e.outlier = True
-                yield self._handle_new_event(e)
-                yield self.notifier.on_new_room_event(
-                    e, extra_users=[joinee]
-                )
+                try:
+                    yield self._handle_new_event(e)
+                    yield self.notifier.on_new_room_event(
+                        e, extra_users=[joinee]
+                    )
+                except:
+                    logger.exception(
+                        "Failed to parse auth event %s",
+                        e.event_id,
+                    )
 
             for e in state:
                 # FIXME: Auth these.
                 e.outlier = True
-                yield self._handle_new_event(e)
-                yield self.notifier.on_new_room_event(
-                    e, extra_users=[joinee]
-                )
+                try:
+                    yield self._handle_new_event(e)
+                    yield self.notifier.on_new_room_event(
+                        e, extra_users=[joinee]
+                    )
+                except:
+                    logger.exception(
+                        "Failed to parse state event %s",
+                        e.event_id,
+                    )
 
             yield self._handle_new_event(
                 event,