Fix `wait_for_stream_position` for multiple waiters. (#8196)

This fixes a bug where having multiple callers waiting on the same
stream and position will cause it to try and compare two deferreds,
which fails (due to the sorted list having an entry of `Tuple[int,
Deferred]`).

changelog.d/8196.misc

@@ -0,0 +1 @@
+Fix `wait_for_stream_position` to allow multiple waiters on same stream ID.

synapse/python_dependencies.py

@@ -66,7 +66,9 @@ REQUIREMENTS = [
     "msgpack>=0.5.2",
     "phonenumbers>=8.2.0",
     "prometheus_client>=0.0.18,<0.9.0",
-    # we use attr.validators.deep_iterable, which arrived in 19.1.0
+    # we use attr.validators.deep_iterable, which arrived in 19.1.0 (Note:
+    # Fedora 31 only has 19.1, so if we want to upgrade we should wait until 33
+    # is out in November.)
     "attrs>=19.1.0",
     "netaddr>=0.7.18",
     "Jinja2>=2.9",

synapse/replication/tcp/client.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 """A replication client for use by synapse workers.
 """
-import heapq
 import logging
 from typing import TYPE_CHECKING, Dict, List, Tuple
 
@@ -219,9 +218,8 @@ class ReplicationDataHandler:
 
         waiting_list = self._streams_to_waiters.setdefault(stream_name, [])
 
-        # We insert into the list using heapq as it is more efficient than
-        # pushing then resorting each time.
-        heapq.heappush(waiting_list, (position, deferred))
+        waiting_list.append((position, deferred))
+        waiting_list.sort(key=lambda t: t[0])
 
         # We measure here to get in flight counts and average waiting time.
         with Measure(self._clock, "repl.wait_for_stream_position"):