Fix handling of failures when calling /event_auth.

When processing an incoming event over federation, we may try and
resolve any unexpected differences in auth events. This is a
non-essential process and so should not stop the processing of the event
if it fails (e.g. due to the remote disappearing or not implementing the
necessary endpoints).

Fixes #3330

synapse/handlers/federation.py

@@ -35,6 +35,7 @@ from synapse.api.errors import (
     CodeMessageException,
     FederationDeniedError,
     FederationError,
+    RequestSendFailed,
     StoreError,
     SynapseError,
 )
@@ -2027,9 +2028,15 @@ class FederationHandler(BaseHandler):
         """
         room_version = yield self.store.get_room_version(event.room_id)
 
+        try:
             yield self._update_auth_events_and_context_for_auth(
                 origin, event, context, auth_events
             )
+        except Exception:
+            # We don't really mind if the above fails, so lets not fail
+            # processing if it does.
+            logger.exception("Failed to call _update_auth_events_and_context_for_auth")
+
         try:
             self.auth.check(room_version, event, auth_events=auth_events)
         except AuthError as e:
@@ -2042,6 +2049,15 @@ class FederationHandler(BaseHandler):
     ):
         """Helper for do_auth. See there for docs.
 
+        Checks whether a given event has the expected auth events. If it
+        doesn't then we talk to the remote server to compare state to see if
+        we can come to a consensus (e.g. if one server missed some valid
+        state).
+
+        This attempts to resovle any potential divergence of state between
+        servers, but is not essential and so failures should not block further
+        processing of the event.
+
         Args:
             origin (str):
             event (synapse.events.EventBase):
@@ -2087,10 +2103,15 @@ class FederationHandler(BaseHandler):
                 "auth_events contains unknown events: %s",
                 missing_auth,
             )
+            try:
                 try:
                     remote_auth_chain = yield self.federation_client.get_event_auth(
                         origin, event.room_id, event.event_id
                     )
+                except RequestSendFailed:
+                    # The other side isn't around or doesn't implement the
+                    # endpoint, so lets just bail out.
+                    return
 
                 seen_remotes = yield self.store.have_seen_events(
                     [e.event_id for e in remote_auth_chain]
@@ -2236,12 +2257,17 @@ class FederationHandler(BaseHandler):
 
         try:
             # 2. Get remote difference.
+            try:
                 result = yield self.federation_client.query_auth(
                     origin,
                     event.room_id,
                     event.event_id,
                     local_auth_chain,
                 )
+            except RequestSendFailed:
+                # The other side isn't around or doesn't implement the
+                # endpoint, so lets just bail out.
+                return
 
             seen_remotes = yield self.store.have_seen_events(
                 [e.event_id for e in result["auth_chain"]]