Ignore backoff history for invites, aliases, and roomdirs

Add a param to the federation client which lets us ignore historical backoff
data for federation queries, and set it for a handful of operations.

synapse/util/retryutils.py

@@ -35,7 +35,8 @@ class NotRetryingDestination(Exception):
 
 
 @defer.inlineCallbacks
-def get_retry_limiter(destination, clock, store, **kwargs):
+def get_retry_limiter(destination, clock, store, ignore_backoff=False,
+                      **kwargs):
     """For a given destination check if we have previously failed to
     send a request there and are waiting before retrying the destination.
     If we are not ready to retry the destination, this will raise a
@@ -43,6 +44,14 @@ def get_retry_limiter(destination, clock, store, **kwargs):
     that will mark the destination as down if an exception is thrown (excluding
     CodeMessageException with code < 500)
 
+    Args:
+        destination (str): name of homeserver
+        clock (synapse.util.clock): timing source
+        store (synapse.storage.transactions.TransactionStore): datastore
+        ignore_backoff (bool): true to ignore the historical backoff data and
+            try the request anyway. We will still update the next
+            retry_interval on success/failure.
+
     Example usage:
 
         try:
@@ -66,7 +75,7 @@ def get_retry_limiter(destination, clock, store, **kwargs):
 
         now = int(clock.time_msec())
 
-        if retry_last_ts + retry_interval > now:
+        if not ignore_backoff and retry_last_ts + retry_interval > now:
             raise NotRetryingDestination(
                 retry_last_ts=retry_last_ts,
                 retry_interval=retry_interval,