Merge tag 'v1.45.0' into babolivier/dinsic_1.48.0

Synapse 1.45.0 (2021-10-19)
===========================

No functional changes since Synapse 1.45.0rc2.

Known Issues
------------

- A suspected [performance regression](https://github.com/matrix-org/synapse/issues/11049) which was first reported after the release of 1.44.0 remains unresolved.

  We have not been able to identify a probable cause. Affected users report that setting up a federation sender worker appears to alleviate symptoms of the regression.

Improved Documentation
----------------------

- Reword changelog to clarify concerns about a suspected performance regression in 1.44.0. ([\#11117](https://github.com/matrix-org/synapse/issues/11117))

Synapse 1.45.0rc2 (2021-10-14)
==============================

This release candidate [fixes](https://github.com/matrix-org/synapse/issues/11053) a user directory [bug](https://github.com/matrix-org/synapse/issues/11025) present in 1.45.0rc1.

Known Issues
------------

- A suspected [performance regression](https://github.com/matrix-org/synapse/issues/11049) which was first reported after the release of 1.44.0 remains unresolved.

  We have not been able to identify a probable cause. Affected users report that setting up a federation sender worker appears to alleviate symptoms of the regression.

Bugfixes
--------

- Fix a long-standing bug when using multiple event persister workers where events were not correctly sent down `/sync` due to a race. ([\#11045](https://github.com/matrix-org/synapse/issues/11045))
- Fix a bug introduced in Synapse 1.45.0rc1 where the user directory would stop updating if it processed an event from a
  user not in the `users` table. ([\#11053](https://github.com/matrix-org/synapse/issues/11053))
- Fix a bug introduced in Synapse 1.44.0 when logging errors during oEmbed processing. ([\#11061](https://github.com/matrix-org/synapse/issues/11061))

Internal Changes
----------------

- Add an 'approximate difference' method to `StateFilter`. ([\#10825](https://github.com/matrix-org/synapse/issues/10825))
- Fix inconsistent behavior of `get_last_client_by_ip` when reporting data that has not been stored in the database yet. ([\#10970](https://github.com/matrix-org/synapse/issues/10970))
- Fix a bug introduced in Synapse 1.21.0 that causes opentracing and Prometheus metrics for replication requests to be measured incorrectly. ([\#10996](https://github.com/matrix-org/synapse/issues/10996))
- Ensure that cache config tests do not share state. ([\#11036](https://github.com/matrix-org/synapse/issues/11036))

Synapse 1.45.0rc1 (2021-10-12)
==============================

**Note:** Media storage providers module that read from Synapse's configuration need changes as of this version, see the [upgrade notes](https://matrix-org.github.io/synapse/develop/upgrade#upgrading-to-v1450) for more information.

Known Issues
------------

- We are investigating [a performance issue](https://github.com/matrix-org/synapse/issues/11049) which was reported after the release of 1.44.0.
- We are aware of [a bug](https://github.com/matrix-org/synapse/issues/11025) with the user directory when using application services. A second release candidate is expected which will resolve this.

Features
--------

- Add [MSC3069](https://github.com/matrix-org/matrix-doc/pull/3069) support to `/account/whoami`. ([\#9655](https://github.com/matrix-org/synapse/issues/9655))
- Support autodiscovery of oEmbed previews. ([\#10822](https://github.com/matrix-org/synapse/issues/10822))
- Add a `user_may_send_3pid_invite` spam checker callback for modules to allow or deny 3PID invites. ([\#10894](https://github.com/matrix-org/synapse/issues/10894))
- Add a spam checker callback to allow or deny room joins. ([\#10910](https://github.com/matrix-org/synapse/issues/10910))
- Include an `update_synapse_database` script in the distribution. Contributed by @Fizzadar at Beeper. ([\#10954](https://github.com/matrix-org/synapse/issues/10954))
- Include exception information in JSON logging output. Contributed by @Fizzadar at Beeper. ([\#11028](https://github.com/matrix-org/synapse/issues/11028))

Bugfixes
--------

- Fix a minor bug in the response to `/_matrix/client/r0/voip/turnServer`. Contributed by @lukaslihotzki. ([\#10922](https://github.com/matrix-org/synapse/issues/10922))
- Fix a bug where empty `yyyy-mm-dd/` directories would be left behind in the media store's `url_cache_thumbnails/` directory. ([\#10924](https://github.com/matrix-org/synapse/issues/10924))
- Fix a bug introduced in Synapse v1.40.0 where the signature checks for room version 8 and 9 could be applied to earlier room versions in some situations. ([\#10927](https://github.com/matrix-org/synapse/issues/10927))
- Fix a long-standing bug wherein deactivated users still count towards the monthly active users limit. ([\#10947](https://github.com/matrix-org/synapse/issues/10947))
- Fix a long-standing bug which meant that events received over federation were sometimes incorrectly accepted into the room state. ([\#10956](https://github.com/matrix-org/synapse/issues/10956))
- Fix a long-standing bug where rebuilding the user directory wouldn't exclude support and deactivated users. ([\#10960](https://github.com/matrix-org/synapse/issues/10960))
- Fix [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) `/batch_send` endpoint rejecting subsequent batches with unknown batch ID error in existing room versions from the room creator. ([\#10962](https://github.com/matrix-org/synapse/issues/10962))
- Fix a bug that could leak local users' per-room nicknames and avatars when the user directory is rebuilt. ([\#10981](https://github.com/matrix-org/synapse/issues/10981))
- Fix a long-standing bug where the remainder of a batch of user directory changes would be silently dropped if the server left a room early in the batch. ([\#10982](https://github.com/matrix-org/synapse/issues/10982))
- Correct a bugfix introduced in Synapse v1.44.0 that would catch the wrong error if a connection is lost before a response could be written to it. ([\#10995](https://github.com/matrix-org/synapse/issues/10995))
- Fix a long-standing bug where local users' per-room nicknames/avatars were visible to anyone who could see you in the user directory. ([\#11002](https://github.com/matrix-org/synapse/issues/11002))
- Fix a long-standing bug where a user's per-room nickname/avatar would overwrite their profile in the user directory when a room was made public. ([\#11003](https://github.com/matrix-org/synapse/issues/11003))
- Work around a regression, introduced in Synapse v1.39.0, that caused `SynapseError`s raised by the experimental third-party rules module callback `check_event_allowed` to be ignored. ([\#11042](https://github.com/matrix-org/synapse/issues/11042))
- Fix a bug in [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) insertion events in rooms that could cause cross-talk/conflicts between batches. ([\#10877](https://github.com/matrix-org/synapse/issues/10877))

Improved Documentation
----------------------

- Change wording ("reference homeserver") in Synapse repository documentation. Contributed by @maxkratz. ([\#10971](https://github.com/matrix-org/synapse/issues/10971))
- Fix a dead URL in development documentation (SAML) and change wording from "Riot" to "Element". Contributed by @maxkratz. ([\#10973](https://github.com/matrix-org/synapse/issues/10973))
- Add additional content to the Welcome and Overview page of the documentation. ([\#10990](https://github.com/matrix-org/synapse/issues/10990))
- Update links to MSCs in documentation. Contributed by @dklimpel. ([\#10991](https://github.com/matrix-org/synapse/issues/10991))

Internal Changes
----------------

- Improve type hinting in `synapse.util`. ([\#10888](https://github.com/matrix-org/synapse/issues/10888))
- Add further type hints to `synapse.storage.util`. ([\#10892](https://github.com/matrix-org/synapse/issues/10892))
- Fix type hints to be compatible with an upcoming change to Twisted. ([\#10895](https://github.com/matrix-org/synapse/issues/10895))
- Update utility code to handle C implementations of frozendict. ([\#10902](https://github.com/matrix-org/synapse/issues/10902))
- Drop old functionality which maintained database compatibility with Synapse versions before v1.31. ([\#10903](https://github.com/matrix-org/synapse/issues/10903))
- Clean-up configuration helper classes for the `ServerConfig` class. ([\#10915](https://github.com/matrix-org/synapse/issues/10915))
- Use direct references to config flags. ([\#10916](https://github.com/matrix-org/synapse/issues/10916), [\#10959](https://github.com/matrix-org/synapse/issues/10959), [\#10985](https://github.com/matrix-org/synapse/issues/10985))
- Clean up some of the federation event authentication code for clarity. ([\#10926](https://github.com/matrix-org/synapse/issues/10926), [\#10940](https://github.com/matrix-org/synapse/issues/10940), [\#10986](https://github.com/matrix-org/synapse/issues/10986), [\#10987](https://github.com/matrix-org/synapse/issues/10987), [\#10988](https://github.com/matrix-org/synapse/issues/10988), [\#11010](https://github.com/matrix-org/synapse/issues/11010), [\#11011](https://github.com/matrix-org/synapse/issues/11011))
- Refactor various parts of the codebase to use `RoomVersion` objects instead of room version identifier strings. ([\#10934](https://github.com/matrix-org/synapse/issues/10934))
- Refactor user directory tests in preparation for upcoming changes. ([\#10935](https://github.com/matrix-org/synapse/issues/10935))
- Include the event id in the logcontext when handling PDUs received over federation. ([\#10936](https://github.com/matrix-org/synapse/issues/10936))
- Fix logged errors in unit tests. ([\#10939](https://github.com/matrix-org/synapse/issues/10939))
- Fix a broken test to ensure that consent configuration works during registration. ([\#10945](https://github.com/matrix-org/synapse/issues/10945))
- Add type hints to filtering classes. ([\#10958](https://github.com/matrix-org/synapse/issues/10958))
- Add type-hint to `HomeserverTestcase.setup_test_homeserver`. ([\#10961](https://github.com/matrix-org/synapse/issues/10961))
- Fix the test utility function `create_room_as` so that `is_public=True` will explicitly set the `visibility` parameter of room creation requests to `public`. Contributed by @AndrewFerr. ([\#10963](https://github.com/matrix-org/synapse/issues/10963))
- Make the release script more robust and transparent. ([\#10966](https://github.com/matrix-org/synapse/issues/10966))
- Refactor [MSC2716](https://github.com/matrix-org/matrix-doc/pull/2716) `/batch_send` mega function into smaller handler functions. ([\#10974](https://github.com/matrix-org/synapse/issues/10974))
- Log stack traces when a missing opentracing span is detected. ([\#10983](https://github.com/matrix-org/synapse/issues/10983))
- Update GHA config to run tests against Python 3.10 and PostgreSQL 14. ([\#10992](https://github.com/matrix-org/synapse/issues/10992))
- Fix a long-standing bug where `ReadWriteLock`s could drop logging contexts on exit. ([\#10993](https://github.com/matrix-org/synapse/issues/10993))
- Add a `CODEOWNERS` file to automatically request reviews from the `@matrix-org/synapse-core` team on new pull requests. ([\#10994](https://github.com/matrix-org/synapse/issues/10994))
- Add further type hints to `synapse.state`. ([\#11004](https://github.com/matrix-org/synapse/issues/11004))
- Remove the deprecated `BaseHandler` object. ([\#11005](https://github.com/matrix-org/synapse/issues/11005))
- Bump mypy version for CI to 0.910, and pull in new type stubs for dependencies. ([\#11006](https://github.com/matrix-org/synapse/issues/11006))
- Fix CI to run the unit tests without optional deps. ([\#11017](https://github.com/matrix-org/synapse/issues/11017))
- Ensure that cache config tests do not share state. ([\#11019](https://github.com/matrix-org/synapse/issues/11019))
- Add additional type hints to `synapse.server_notices`. ([\#11021](https://github.com/matrix-org/synapse/issues/11021))
- Add additional type hints for `synapse.push`. ([\#11023](https://github.com/matrix-org/synapse/issues/11023))
- When installing the optional developer dependencies, also include the dependencies needed for type-checking and unit testing. ([\#11034](https://github.com/matrix-org/synapse/issues/11034))
- Remove unnecessary list comprehension from `synapse_port_db` to satisfy code style requirements. ([\#11043](https://github.com/matrix-org/synapse/issues/11043))

10 files changed, 974 insertions, 69 deletions

diff --git a/tests/storage/databases/main/test_room.py b/tests/storage/databases/main/test_room.py
index ffee707153..7496974da3 100644
--- a/tests/storage/databases/main/test_room.py
+++ b/tests/storage/databases/main/test_room.py
@@ -79,12 +79,7 @@ class RoomBackgroundUpdateStoreTestCase(HomeserverTestCase):
         self.store.db_pool.updates._all_done = False
 
         # Now let's actually drive the updates to completion
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         # Make sure the background update filled in the room creator
         room_creator_after = self.get_success(
diff --git a/tests/storage/test_appservice.py b/tests/storage/test_appservice.py
index cf9748f218..f26d5acf9c 100644
--- a/tests/storage/test_appservice.py
+++ b/tests/storage/test_appservice.py
@@ -126,7 +126,7 @@ class ApplicationServiceTransactionStoreTestCase(unittest.TestCase):
         self.db_pool = database._db_pool
         self.engine = database.engine
 
-        db_config = hs.config.get_single_database()
+        db_config = hs.config.database.get_single_database()
         self.store = TestTransactionStore(
             database, make_conn(db_config, self.engine, "test"), hs
         )
diff --git a/tests/storage/test_cleanup_extrems.py b/tests/storage/test_cleanup_extrems.py
index 7cc5e621ba..a59c28f896 100644
--- a/tests/storage/test_cleanup_extrems.py
+++ b/tests/storage/test_cleanup_extrems.py
@@ -66,12 +66,7 @@ class CleanupExtremBackgroundUpdateStoreTestCase(HomeserverTestCase):
         # Ugh, have to reset this flag
         self.store.db_pool.updates._all_done = False
 
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
     def test_soft_failed_extremities_handled_correctly(self):
         """Test that extremities are correctly calculated in the presence of
diff --git a/tests/storage/test_client_ips.py b/tests/storage/test_client_ips.py
index 3cc8038f1e..0e4013ebea 100644
--- a/tests/storage/test_client_ips.py
+++ b/tests/storage/test_client_ips.py
@@ -147,6 +147,49 @@ class ClientIpStoreTestCase(unittest.HomeserverTestCase):
         )
 
     @parameterized.expand([(False,), (True,)])
+    def test_get_last_client_ip_by_device(self, after_persisting: bool):
+        """Test `get_last_client_ip_by_device` for persisted and unpersisted data"""
+        self.reactor.advance(12345678)
+
+        user_id = "@user:id"
+        device_id = "MY_DEVICE"
+
+        # Insert a user IP
+        self.get_success(
+            self.store.store_device(
+                user_id,
+                device_id,
+                "display name",
+            )
+        )
+        self.get_success(
+            self.store.insert_client_ip(
+                user_id, "access_token", "ip", "user_agent", device_id
+            )
+        )
+
+        if after_persisting:
+            # Trigger the storage loop
+            self.reactor.advance(10)
+
+        result = self.get_success(
+            self.store.get_last_client_ip_by_device(user_id, device_id)
+        )
+
+        self.assertEqual(
+            result,
+            {
+                (user_id, device_id): {
+                    "user_id": user_id,
+                    "device_id": device_id,
+                    "ip": "ip",
+                    "user_agent": "user_agent",
+                    "last_seen": 12345678000,
+                },
+            },
+        )
+
+    @parameterized.expand([(False,), (True,)])
     def test_get_user_ip_and_agents(self, after_persisting: bool):
         """Test `get_user_ip_and_agents` for persisted and unpersisted data"""
         self.reactor.advance(12345678)
@@ -242,12 +285,7 @@ class ClientIpStoreTestCase(unittest.HomeserverTestCase):
 
     def test_devices_last_seen_bg_update(self):
         # First make sure we have completed all updates.
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         user_id = "@user:id"
         device_id = "MY_DEVICE"
@@ -311,12 +349,7 @@ class ClientIpStoreTestCase(unittest.HomeserverTestCase):
         self.store.db_pool.updates._all_done = False
 
         # Now let's actually drive the updates to completion
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         # We should now get the correct result again
         result = self.get_success(
@@ -337,12 +370,7 @@ class ClientIpStoreTestCase(unittest.HomeserverTestCase):
 
     def test_old_user_ips_pruned(self):
         # First make sure we have completed all updates.
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         user_id = "@user:id"
         device_id = "MY_DEVICE"
diff --git a/tests/storage/test_event_chain.py b/tests/storage/test_event_chain.py
index 93136f0717..b31c5eb5ec 100644
--- a/tests/storage/test_event_chain.py
+++ b/tests/storage/test_event_chain.py
@@ -578,12 +578,7 @@ class EventChainBackgroundUpdateTestCase(HomeserverTestCase):
         # Ugh, have to reset this flag
         self.store.db_pool.updates._all_done = False
 
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         # Test that the `has_auth_chain_index` has been set
         self.assertTrue(self.get_success(self.store.has_auth_chain_index(room_id)))
@@ -619,12 +614,7 @@ class EventChainBackgroundUpdateTestCase(HomeserverTestCase):
         # Ugh, have to reset this flag
         self.store.db_pool.updates._all_done = False
 
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         # Test that the `has_auth_chain_index` has been set
         self.assertTrue(self.get_success(self.store.has_auth_chain_index(room_id1)))
diff --git a/tests/storage/test_monthly_active_users.py b/tests/storage/test_monthly_active_users.py
index 944dbc34a2..d6b4cdd788 100644
--- a/tests/storage/test_monthly_active_users.py
+++ b/tests/storage/test_monthly_active_users.py
@@ -51,7 +51,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
 
     @override_config({"max_mau_value": 3, "mau_limit_reserved_threepids": gen_3pids(3)})
     def test_initialise_reserved_users(self):
-        threepids = self.hs.config.mau_limits_reserved_threepids
+        threepids = self.hs.config.server.mau_limits_reserved_threepids
 
         # register three users, of which two have reserved 3pids, and a third
         # which is a support user.
@@ -101,9 +101,9 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         # XXX some of this is redundant. poking things into the config shouldn't
         # work, and in any case it's not obvious what we expect to happen when
         # we advance the reactor.
-        self.hs.config.max_mau_value = 0
+        self.hs.config.server.max_mau_value = 0
         self.reactor.advance(FORTY_DAYS)
-        self.hs.config.max_mau_value = 5
+        self.hs.config.server.max_mau_value = 5
 
         self.get_success(self.store.reap_monthly_active_users())
 
@@ -183,7 +183,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         self.get_success(d)
 
         count = self.get_success(self.store.get_monthly_active_count())
-        self.assertEqual(count, self.hs.config.max_mau_value)
+        self.assertEqual(count, self.hs.config.server.max_mau_value)
 
         self.reactor.advance(FORTY_DAYS)
 
@@ -199,7 +199,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
     def test_reap_monthly_active_users_reserved_users(self):
         """Tests that reaping correctly handles reaping where reserved users are
         present"""
-        threepids = self.hs.config.mau_limits_reserved_threepids
+        threepids = self.hs.config.server.mau_limits_reserved_threepids
         initial_users = len(threepids)
         reserved_user_number = initial_users - 1
         for i in range(initial_users):
@@ -234,7 +234,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
         self.get_success(d)
 
         count = self.get_success(self.store.get_monthly_active_count())
-        self.assertEqual(count, self.hs.config.max_mau_value)
+        self.assertEqual(count, self.hs.config.server.max_mau_value)
 
     def test_populate_monthly_users_is_guest(self):
         # Test that guest users are not added to mau list
@@ -294,7 +294,7 @@ class MonthlyActiveUsersTestCase(unittest.HomeserverTestCase):
             {"medium": "email", "address": user2_email},
         ]
 
-        self.hs.config.mau_limits_reserved_threepids = threepids
+        self.hs.config.server.mau_limits_reserved_threepids = threepids
         d = self.store.db_pool.runInteraction(
             "initialise", self.store._initialise_reserved_users, threepids
         )
diff --git a/tests/storage/test_roommember.py b/tests/storage/test_roommember.py
index c72dc40510..2873e22ccf 100644
--- a/tests/storage/test_roommember.py
+++ b/tests/storage/test_roommember.py
@@ -169,12 +169,7 @@ class CurrentStateMembershipUpdateTestCase(unittest.HomeserverTestCase):
 
     def test_can_rerun_update(self):
         # First make sure we have completed all updates.
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
 
         # Now let's create a room, which will insert a membership
         user = UserID("alice", "test")
@@ -197,9 +192,4 @@ class CurrentStateMembershipUpdateTestCase(unittest.HomeserverTestCase):
         self.store.db_pool.updates._all_done = False
 
         # Now let's actually drive the updates to completion
-        while not self.get_success(
-            self.store.db_pool.updates.has_completed_background_updates()
-        ):
-            self.get_success(
-                self.store.db_pool.updates.do_next_background_update(100), by=0.1
-            )
+        self.wait_for_background_updates()
diff --git a/tests/storage/test_state.py b/tests/storage/test_state.py
index 32060f2abd..70d52b088c 100644
--- a/tests/storage/test_state.py
+++ b/tests/storage/test_state.py
@@ -21,7 +21,7 @@ from synapse.api.room_versions import RoomVersions
 from synapse.storage.state import StateFilter
 from synapse.types import RoomID, UserID
 
-from tests.unittest import HomeserverTestCase
+from tests.unittest import HomeserverTestCase, TestCase
 
 logger = logging.getLogger(__name__)
 
@@ -105,7 +105,6 @@ class StateStoreTestCase(HomeserverTestCase):
         self.assertEqual({ev.event_id for ev in state_list}, {e1.event_id, e2.event_id})
 
     def test_get_state_for_event(self):
-
         # this defaults to a linear DAG as each new injection defaults to whatever
         # forward extremities are currently in the DB for this room.
         e1 = self.inject_state_event(self.room, self.u_alice, EventTypes.Create, "", {})
@@ -483,3 +482,513 @@ class StateStoreTestCase(HomeserverTestCase):
 
         self.assertEqual(is_all, True)
         self.assertDictEqual({(e5.type, e5.state_key): e5.event_id}, state_dict)
+
+
+class StateFilterDifferenceTestCase(TestCase):
+    def assert_difference(
+        self, minuend: StateFilter, subtrahend: StateFilter, expected: StateFilter
+    ):
+        self.assertEqual(
+            minuend.approx_difference(subtrahend),
+            expected,
+            f"StateFilter difference not correct:\n\n\t{minuend!r}\nminus\n\t{subtrahend!r}\nwas\n\t{minuend.approx_difference(subtrahend)}\nexpected\n\t{expected}",
+        )
+
+    def test_state_filter_difference_no_include_other_minus_no_include_other(self):
+        """
+        Tests the StateFilter.approx_difference method
+        where, in a.approx_difference(b), both a and b do not have the
+        include_others flag set.
+        """
+        # (wildcard on state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.Create: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.CanonicalAlias: None},
+                include_others=False,
+            ),
+            StateFilter.freeze({EventTypes.Create: None}, include_others=False),
+        )
+
+        # (wildcard on state keys) - (specific state keys)
+        # This one is an over-approximation because we can't represent
+        # 'all state keys except a few named examples'
+        self.assert_difference(
+            StateFilter.freeze({EventTypes.Member: None}, include_others=False),
+            StateFilter.freeze(
+                {EventTypes.Member: {"@wombat:spqr"}},
+                include_others=False,
+            ),
+            StateFilter.freeze({EventTypes.Member: None}, include_others=False),
+        )
+
+        # (wildcard on state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {EventTypes.CanonicalAlias: {""}},
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (specific state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr"},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+        )
+
+    def test_state_filter_difference_include_other_minus_no_include_other(self):
+        """
+        Tests the StateFilter.approx_difference method
+        where, in a.approx_difference(b), only a has the include_others flag set.
+        """
+        # (wildcard on state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.Create: None},
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.CanonicalAlias: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Create: None,
+                    EventTypes.Member: set(),
+                    EventTypes.CanonicalAlias: set(),
+                },
+                include_others=True,
+            ),
+        )
+
+        # (wildcard on state keys) - (specific state keys)
+        # This one is an over-approximation because we can't represent
+        # 'all state keys except a few named examples'
+        # This also shows that the resultant state filter is normalised.
+        self.assert_difference(
+            StateFilter.freeze({EventTypes.Member: None}, include_others=True),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr"},
+                    EventTypes.Create: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter(types=frozendict(), include_others=True),
+        )
+
+        # (wildcard on state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=False,
+            ),
+            StateFilter(
+                types=frozendict(),
+                include_others=True,
+            ),
+        )
+
+        # (specific state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.CanonicalAlias: {""},
+                    EventTypes.Member: set(),
+                },
+                include_others=True,
+            ),
+        )
+
+        # (specific state keys) - (specific state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr"},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+        )
+
+        # (specific state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+        )
+
+    def test_state_filter_difference_include_other_minus_include_other(self):
+        """
+        Tests the StateFilter.approx_difference method
+        where, in a.approx_difference(b), both a and b have the include_others
+        flag set.
+        """
+        # (wildcard on state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.Create: None},
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.CanonicalAlias: None},
+                include_others=True,
+            ),
+            StateFilter(types=frozendict(), include_others=False),
+        )
+
+        # (wildcard on state keys) - (specific state keys)
+        # This one is an over-approximation because we can't represent
+        # 'all state keys except a few named examples'
+        self.assert_difference(
+            StateFilter.freeze({EventTypes.Member: None}, include_others=True),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.CanonicalAlias: None},
+                include_others=False,
+            ),
+        )
+
+        # (wildcard on state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=True,
+            ),
+            StateFilter(
+                types=frozendict(),
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (specific state keys)
+        # This one is an over-approximation because we can't represent
+        # 'all state keys except a few named examples'
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                    EventTypes.Create: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr"},
+                    EventTypes.Create: set(),
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@spqr:spqr"},
+                    EventTypes.Create: {""},
+                },
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                },
+                include_others=False,
+            ),
+        )
+
+    def test_state_filter_difference_no_include_other_minus_include_other(self):
+        """
+        Tests the StateFilter.approx_difference method
+        where, in a.approx_difference(b), only b has the include_others flag set.
+        """
+        # (wildcard on state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.Create: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None, EventTypes.CanonicalAlias: None},
+                include_others=True,
+            ),
+            StateFilter(types=frozendict(), include_others=False),
+        )
+
+        # (wildcard on state keys) - (specific state keys)
+        # This one is an over-approximation because we can't represent
+        # 'all state keys except a few named examples'
+        self.assert_difference(
+            StateFilter.freeze({EventTypes.Member: None}, include_others=False),
+            StateFilter.freeze(
+                {EventTypes.Member: {"@wombat:spqr"}},
+                include_others=True,
+            ),
+            StateFilter.freeze({EventTypes.Member: None}, include_others=False),
+        )
+
+        # (wildcard on state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (wildcard on state keys):
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {EventTypes.Member: None},
+                include_others=True,
+            ),
+            StateFilter(
+                types=frozendict(),
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (specific state keys)
+        # This one is an over-approximation because we can't represent
+        # 'all state keys except a few named examples'
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr"},
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@spqr:spqr"},
+                },
+                include_others=False,
+            ),
+        )
+
+        # (specific state keys) - (no state keys)
+        self.assert_difference(
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                    EventTypes.CanonicalAlias: {""},
+                },
+                include_others=False,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: set(),
+                },
+                include_others=True,
+            ),
+            StateFilter.freeze(
+                {
+                    EventTypes.Member: {"@wombat:spqr", "@spqr:spqr"},
+                },
+                include_others=False,
+            ),
+        )
+
+    def test_state_filter_difference_simple_cases(self):
+        """
+        Tests some very simple cases of the StateFilter approx_difference,
+        that are not explicitly tested by the more in-depth tests.
+        """
+
+        self.assert_difference(StateFilter.all(), StateFilter.all(), StateFilter.none())
+
+        self.assert_difference(
+            StateFilter.all(),
+            StateFilter.none(),
+            StateFilter.all(),
+        )
diff --git a/tests/storage/test_txn_limit.py b/tests/storage/test_txn_limit.py
index 6ff3ebb137..ace82cbf42 100644
--- a/tests/storage/test_txn_limit.py
+++ b/tests/storage/test_txn_limit.py
@@ -22,7 +22,7 @@ class SQLTransactionLimitTestCase(unittest.HomeserverTestCase):
         return self.setup_test_homeserver(db_txn_limit=1000)
 
     def test_config(self):
-        db_config = self.hs.config.get_single_database()
+        db_config = self.hs.config.database.get_single_database()
         self.assertEqual(db_config.config["txn_limit"], 1000)
 
     def test_select(self):
diff --git a/tests/storage/test_user_directory.py b/tests/storage/test_user_directory.py
index 222e5d129d..be3ed64f5e 100644
--- a/tests/storage/test_user_directory.py
+++ b/tests/storage/test_user_directory.py
@@ -11,7 +11,22 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+from typing import Any, Dict, List, Set, Tuple
+from unittest import mock
+from unittest.mock import Mock, patch
 
+from twisted.test.proto_helpers import MemoryReactor
+
+from synapse.api.constants import EventTypes, Membership, UserTypes
+from synapse.appservice import ApplicationService
+from synapse.rest import admin
+from synapse.rest.client import login, register, room
+from synapse.server import HomeServer
+from synapse.storage import DataStore
+from synapse.storage.roommember import ProfileInfo
+from synapse.util import Clock
+
+from tests.test_utils.event_injection import inject_member_event
 from tests.unittest import HomeserverTestCase, override_config
 
 ALICE = "@alice:a"
@@ -21,8 +36,391 @@ BOBBY = "@bobby:a"
 BELA = "@somenickname:a"
 
 
+class GetUserDirectoryTables:
+    """Helper functions that we want to reuse in tests/handlers/test_user_directory.py"""
+
+    def __init__(self, store: DataStore):
+        self.store = store
+
+    def _compress_shared(
+        self, shared: List[Dict[str, str]]
+    ) -> Set[Tuple[str, str, str]]:
+        """
+        Compress a list of users who share rooms dicts to a list of tuples.
+        """
+        r = set()
+        for i in shared:
+            r.add((i["user_id"], i["other_user_id"], i["room_id"]))
+        return r
+
+    async def get_users_in_public_rooms(self) -> List[Tuple[str, str]]:
+        """Fetch the entire `users_in_public_rooms` table.
+
+        Returns a list of tuples (user_id, room_id) where room_id is public and
+        contains the user with the given id.
+        """
+        r = await self.store.db_pool.simple_select_list(
+            "users_in_public_rooms", None, ("user_id", "room_id")
+        )
+
+        retval = []
+        for i in r:
+            retval.append((i["user_id"], i["room_id"]))
+        return retval
+
+    async def get_users_who_share_private_rooms(self) -> List[Dict[str, str]]:
+        """Fetch the entire `users_who_share_private_rooms` table.
+
+        Returns a dict containing "user_id", "other_user_id" and "room_id" keys.
+        The dicts can be flattened to Tuples with the `_compress_shared` method.
+        (This seems a little awkward---maybe we could clean this up.)
+        """
+
+        return await self.store.db_pool.simple_select_list(
+            "users_who_share_private_rooms",
+            None,
+            ["user_id", "other_user_id", "room_id"],
+        )
+
+    async def get_users_in_user_directory(self) -> Set[str]:
+        """Fetch the set of users in the `user_directory` table.
+
+        This is useful when checking we've correctly excluded users from the directory.
+        """
+        result = await self.store.db_pool.simple_select_list(
+            "user_directory",
+            None,
+            ["user_id"],
+        )
+        return {row["user_id"] for row in result}
+
+    async def get_profiles_in_user_directory(self) -> Dict[str, ProfileInfo]:
+        """Fetch users and their profiles from the `user_directory` table.
+
+        This is useful when we want to inspect display names and avatars.
+        It's almost the entire contents of the `user_directory` table: the only
+        thing missing is an unused room_id column.
+        """
+        rows = await self.store.db_pool.simple_select_list(
+            "user_directory",
+            None,
+            ("user_id", "display_name", "avatar_url"),
+        )
+        return {
+            row["user_id"]: ProfileInfo(
+                display_name=row["display_name"], avatar_url=row["avatar_url"]
+            )
+            for row in rows
+        }
+
+
+class UserDirectoryInitialPopulationTestcase(HomeserverTestCase):
+    """Ensure that rebuilding the directory writes the correct data to the DB.
+
+    See also tests/handlers/test_user_directory.py for similar checks. They
+    test the incremental updates, rather than the big rebuild.
+    """
+
+    servlets = [
+        login.register_servlets,
+        admin.register_servlets,
+        room.register_servlets,
+        register.register_servlets,
+    ]
+
+    def make_homeserver(self, reactor: MemoryReactor, clock: Clock) -> HomeServer:
+        self.appservice = ApplicationService(
+            token="i_am_an_app_service",
+            hostname="test",
+            id="1234",
+            namespaces={"users": [{"regex": r"@as_user.*", "exclusive": True}]},
+            sender="@as:test",
+        )
+
+        mock_load_appservices = Mock(return_value=[self.appservice])
+        with patch(
+            "synapse.storage.databases.main.appservice.load_appservices",
+            mock_load_appservices,
+        ):
+            hs = super().make_homeserver(reactor, clock)
+        return hs
+
+    def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None:
+        self.store = hs.get_datastore()
+        self.user_dir_helper = GetUserDirectoryTables(self.store)
+
+    def _purge_and_rebuild_user_dir(self) -> None:
+        """Nuke the user directory tables, start the background process to
+        repopulate them, and wait for the process to complete. This allows us
+        to inspect the outcome of the background process alone, without any of
+        the other incremental updates.
+        """
+        self.get_success(self.store.update_user_directory_stream_pos(None))
+        self.get_success(self.store.delete_all_from_user_dir())
+
+        shares_private = self.get_success(
+            self.user_dir_helper.get_users_who_share_private_rooms()
+        )
+        public_users = self.get_success(
+            self.user_dir_helper.get_users_in_public_rooms()
+        )
+
+        # Nothing updated yet
+        self.assertEqual(shares_private, [])
+        self.assertEqual(public_users, [])
+
+        # Ugh, have to reset this flag
+        self.store.db_pool.updates._all_done = False
+
+        self.get_success(
+            self.store.db_pool.simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_user_directory_createtables",
+                    "progress_json": "{}",
+                },
+            )
+        )
+        self.get_success(
+            self.store.db_pool.simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_user_directory_process_rooms",
+                    "progress_json": "{}",
+                    "depends_on": "populate_user_directory_createtables",
+                },
+            )
+        )
+        self.get_success(
+            self.store.db_pool.simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_user_directory_process_users",
+                    "progress_json": "{}",
+                    "depends_on": "populate_user_directory_process_rooms",
+                },
+            )
+        )
+        self.get_success(
+            self.store.db_pool.simple_insert(
+                "background_updates",
+                {
+                    "update_name": "populate_user_directory_cleanup",
+                    "progress_json": "{}",
+                    "depends_on": "populate_user_directory_process_users",
+                },
+            )
+        )
+
+        self.wait_for_background_updates()
+
+    def test_initial(self) -> None:
+        """
+        The user directory's initial handler correctly updates the search tables.
+        """
+        u1 = self.register_user("user1", "pass")
+        u1_token = self.login(u1, "pass")
+        u2 = self.register_user("user2", "pass")
+        u2_token = self.login(u2, "pass")
+        u3 = self.register_user("user3", "pass")
+        u3_token = self.login(u3, "pass")
+
+        room = self.helper.create_room_as(u1, is_public=True, tok=u1_token)
+        self.helper.invite(room, src=u1, targ=u2, tok=u1_token)
+        self.helper.join(room, user=u2, tok=u2_token)
+
+        private_room = self.helper.create_room_as(u1, is_public=False, tok=u1_token)
+        self.helper.invite(private_room, src=u1, targ=u3, tok=u1_token)
+        self.helper.join(private_room, user=u3, tok=u3_token)
+
+        # Do the initial population of the user directory via the background update
+        self._purge_and_rebuild_user_dir()
+
+        shares_private = self.get_success(
+            self.user_dir_helper.get_users_who_share_private_rooms()
+        )
+        public_users = self.get_success(
+            self.user_dir_helper.get_users_in_public_rooms()
+        )
+
+        # User 1 and User 2 are in the same public room
+        self.assertEqual(set(public_users), {(u1, room), (u2, room)})
+
+        # User 1 and User 3 share private rooms
+        self.assertEqual(
+            self.user_dir_helper._compress_shared(shares_private),
+            {(u1, u3, private_room), (u3, u1, private_room)},
+        )
+
+        # All three should have entries in the directory
+        users = self.get_success(self.user_dir_helper.get_users_in_user_directory())
+        self.assertEqual(users, {u1, u2, u3})
+
+    # The next four tests (test_population_excludes_*) all set up
+    #   - A normal user included in the user dir
+    #   - A public and private room created by that user
+    #   - A user excluded from the room dir, belonging to both rooms
+
+    # They match similar logic in handlers/test_user_directory.py But that tests
+    # updating the directory; this tests rebuilding it from scratch.
+
+    def _create_rooms_and_inject_memberships(
+        self, creator: str, token: str, joiner: str
+    ) -> Tuple[str, str]:
+        """Create a public and private room as a normal user.
+        Then get the `joiner` into those rooms.
+        """
+        public_room = self.helper.create_room_as(
+            creator,
+            is_public=True,
+            # See https://github.com/matrix-org/synapse/issues/10951
+            extra_content={"visibility": "public"},
+            tok=token,
+        )
+        private_room = self.helper.create_room_as(creator, is_public=False, tok=token)
+
+        # HACK: get the user into these rooms
+        self.get_success(inject_member_event(self.hs, public_room, joiner, "join"))
+        self.get_success(inject_member_event(self.hs, private_room, joiner, "join"))
+
+        return public_room, private_room
+
+    def _check_room_sharing_tables(
+        self, normal_user: str, public_room: str, private_room: str
+    ) -> None:
+        # After rebuilding the directory, we should only see the normal user.
+        users = self.get_success(self.user_dir_helper.get_users_in_user_directory())
+        self.assertEqual(users, {normal_user})
+        in_public_rooms = self.get_success(
+            self.user_dir_helper.get_users_in_public_rooms()
+        )
+        self.assertEqual(set(in_public_rooms), {(normal_user, public_room)})
+        in_private_rooms = self.get_success(
+            self.user_dir_helper.get_users_who_share_private_rooms()
+        )
+        self.assertEqual(in_private_rooms, [])
+
+    def test_population_excludes_support_user(self) -> None:
+        # Create a normal and support user.
+        user = self.register_user("user", "pass")
+        token = self.login(user, "pass")
+        support = "@support1:test"
+        self.get_success(
+            self.store.register_user(
+                user_id=support, password_hash=None, user_type=UserTypes.SUPPORT
+            )
+        )
+
+        # Join the support user to rooms owned by the normal user.
+        public, private = self._create_rooms_and_inject_memberships(
+            user, token, support
+        )
+
+        # Rebuild the directory.
+        self._purge_and_rebuild_user_dir()
+
+        # Check the support user is not in the directory.
+        self._check_room_sharing_tables(user, public, private)
+
+    def test_population_excludes_deactivated_user(self) -> None:
+        user = self.register_user("naughty", "pass")
+        admin = self.register_user("admin", "pass", admin=True)
+        admin_token = self.login(admin, "pass")
+
+        # Deactivate the user.
+        channel = self.make_request(
+            "PUT",
+            f"/_synapse/admin/v2/users/{user}",
+            access_token=admin_token,
+            content={"deactivated": True},
+        )
+        self.assertEqual(channel.code, 200)
+        self.assertEqual(channel.json_body["deactivated"], True)
+
+        # Join the deactivated user to rooms owned by the admin.
+        # Is this something that could actually happen outside of a test?
+        public, private = self._create_rooms_and_inject_memberships(
+            admin, admin_token, user
+        )
+
+        # Rebuild the user dir. The deactivated user should be missing.
+        self._purge_and_rebuild_user_dir()
+        self._check_room_sharing_tables(admin, public, private)
+
+    def test_population_excludes_appservice_user(self) -> None:
+        # Register an AS user.
+        user = self.register_user("user", "pass")
+        token = self.login(user, "pass")
+        as_user = self.register_appservice_user("as_user_potato", self.appservice.token)
+
+        # Join the AS user to rooms owned by the normal user.
+        public, private = self._create_rooms_and_inject_memberships(
+            user, token, as_user
+        )
+
+        # Rebuild the directory.
+        self._purge_and_rebuild_user_dir()
+
+        # Check the AS user is not in the directory.
+        self._check_room_sharing_tables(user, public, private)
+
+    def test_population_excludes_appservice_sender(self) -> None:
+        user = self.register_user("user", "pass")
+        token = self.login(user, "pass")
+
+        # Join the AS sender to rooms owned by the normal user.
+        public, private = self._create_rooms_and_inject_memberships(
+            user, token, self.appservice.sender
+        )
+
+        # Rebuild the directory.
+        self._purge_and_rebuild_user_dir()
+
+        # Check the AS sender is not in the directory.
+        self._check_room_sharing_tables(user, public, private)
+
+    def test_population_conceals_private_nickname(self) -> None:
+        # Make a private room, and set a nickname within
+        user = self.register_user("aaaa", "pass")
+        user_token = self.login(user, "pass")
+        private_room = self.helper.create_room_as(user, is_public=False, tok=user_token)
+        self.helper.send_state(
+            private_room,
+            EventTypes.Member,
+            state_key=user,
+            body={"membership": Membership.JOIN, "displayname": "BBBB"},
+            tok=user_token,
+        )
+
+        # Rebuild the user directory. Make the rescan of the `users` table a no-op
+        # so we only see the effect of scanning the `room_memberships` table.
+        async def mocked_process_users(*args: Any, **kwargs: Any) -> int:
+            await self.store.db_pool.updates._end_background_update(
+                "populate_user_directory_process_users"
+            )
+            return 1
+
+        with mock.patch.dict(
+            self.store.db_pool.updates._background_update_handlers,
+            populate_user_directory_process_users=mocked_process_users,
+        ):
+            self._purge_and_rebuild_user_dir()
+
+        # Local users are ignored by the scan over rooms
+        users = self.get_success(self.user_dir_helper.get_profiles_in_user_directory())
+        self.assertEqual(users, {})
+
+        # Do a full rebuild including the scan over the `users` table. The local
+        # user should appear with their profile name.
+        self._purge_and_rebuild_user_dir()
+        users = self.get_success(self.user_dir_helper.get_profiles_in_user_directory())
+        self.assertEqual(
+            users, {user: ProfileInfo(display_name="aaaa", avatar_url=None)}
+        )
+
+
 class UserDirectoryStoreTestCase(HomeserverTestCase):
-    def prepare(self, reactor, clock, hs):
+    def prepare(self, reactor: MemoryReactor, clock: Clock, hs: HomeServer) -> None:
         self.store = hs.get_datastore()
 
         # alice and bob are both in !room_id. bobby is not but shares
@@ -33,7 +431,7 @@ class UserDirectoryStoreTestCase(HomeserverTestCase):
         self.get_success(self.store.update_profile_in_user_dir(BELA, "Bela", None))
         self.get_success(self.store.add_users_in_public_rooms("!room:id", (ALICE, BOB)))
 
-    def test_search_user_dir(self):
+    def test_search_user_dir(self) -> None:
         # normally when alice searches the directory she should just find
         # bob because bobby doesn't share a room with her.
         r = self.get_success(self.store.search_user_dir(ALICE, "bob", 10))
@@ -44,7 +442,7 @@ class UserDirectoryStoreTestCase(HomeserverTestCase):
         )
 
     @override_config({"user_directory": {"search_all_users": True}})
-    def test_search_user_dir_all_users(self):
+    def test_search_user_dir_all_users(self) -> None:
         r = self.get_success(self.store.search_user_dir(ALICE, "bob", 10))
         self.assertFalse(r["limited"])
         self.assertEqual(2, len(r["results"]))
@@ -58,7 +456,7 @@ class UserDirectoryStoreTestCase(HomeserverTestCase):
         )
 
     @override_config({"user_directory": {"search_all_users": True}})
-    def test_search_user_dir_stop_words(self):
+    def test_search_user_dir_stop_words(self) -> None:
         """Tests that a user can look up another user by searching for the start if its
         display name even if that name happens to be a common English word that would
         usually be ignored in full text searches.