summary refs log tree commit diff
path: root/synapse
diff options
context:
space:
mode:
Diffstat (limited to 'synapse')
-rw-r--r--synapse/app/frontend_proxy.py31
-rw-r--r--synapse/app/synchrotron.py3
-rwxr-xr-xsynapse/app/synctl.py24
-rw-r--r--synapse/federation/transaction_queue.py28
-rw-r--r--synapse/handlers/initial_sync.py1
-rw-r--r--synapse/handlers/message.py2
-rw-r--r--synapse/handlers/presence.py5
-rw-r--r--synapse/handlers/room_list.py2
-rw-r--r--synapse/handlers/room_member.py32
-rw-r--r--synapse/handlers/sync.py7
-rw-r--r--synapse/replication/slave/storage/client_ips.py2
-rw-r--r--synapse/replication/tcp/streams.py2
-rw-r--r--synapse/rest/client/v1/admin.py22
-rw-r--r--synapse/rest/client/v1/presence.py2
-rw-r--r--synapse/storage/client_ips.py2
-rw-r--r--synapse/storage/event_push_actions.py9
-rw-r--r--synapse/storage/events.py9
-rw-r--r--synapse/storage/roommember.py13
-rw-r--r--synapse/storage/search.py2
-rw-r--r--synapse/storage/state.py7
-rw-r--r--synapse/util/caches/__init__.py10
-rw-r--r--synapse/util/caches/descriptors.py4
22 files changed, 158 insertions, 61 deletions
diff --git a/synapse/app/frontend_proxy.py b/synapse/app/frontend_proxy.py
index de889357c3..b517739b8b 100644
--- a/synapse/app/frontend_proxy.py
+++ b/synapse/app/frontend_proxy.py
@@ -36,6 +36,7 @@ from synapse.replication.slave.storage.client_ips import SlavedClientIpStore
 from synapse.replication.slave.storage.devices import SlavedDeviceStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
 from synapse.replication.tcp.client import ReplicationClientHandler
+from synapse.rest.client.v1.base import ClientV1RestServlet, client_path_patterns
 from synapse.rest.client.v2_alpha._base import client_v2_patterns
 from synapse.server import HomeServer
 from synapse.storage.engines import create_engine
@@ -49,6 +50,35 @@ from twisted.web.resource import NoResource
 logger = logging.getLogger("synapse.app.frontend_proxy")
 
 
+class PresenceStatusStubServlet(ClientV1RestServlet):
+    PATTERNS = client_path_patterns("/presence/(?P<user_id>[^/]*)/status")
+
+    def __init__(self, hs):
+        super(PresenceStatusStubServlet, self).__init__(hs)
+        self.http_client = hs.get_simple_http_client()
+        self.auth = hs.get_auth()
+        self.main_uri = hs.config.worker_main_http_uri
+
+    @defer.inlineCallbacks
+    def on_GET(self, request, user_id):
+        # Pass through the auth headers, if any, in case the access token
+        # is there.
+        auth_headers = request.requestHeaders.getRawHeaders("Authorization", [])
+        headers = {
+            "Authorization": auth_headers,
+        }
+        result = yield self.http_client.get_json(
+            self.main_uri + request.uri,
+            headers=headers,
+        )
+        defer.returnValue((200, result))
+
+    @defer.inlineCallbacks
+    def on_PUT(self, request, user_id):
+        yield self.auth.get_user_by_req(request)
+        defer.returnValue((200, {}))
+
+
 class KeyUploadServlet(RestServlet):
     PATTERNS = client_v2_patterns("/keys/upload(/(?P<device_id>[^/]+))?$")
 
@@ -135,6 +165,7 @@ class FrontendProxyServer(HomeServer):
                 elif name == "client":
                     resource = JsonResource(self, canonical_json=False)
                     KeyUploadServlet(self).register(resource)
+                    PresenceStatusStubServlet(self).register(resource)
                     resources.update({
                         "/_matrix/client/r0": resource,
                         "/_matrix/client/unstable": resource,
diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index 508b66613d..cf0223e3b3 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -113,6 +113,7 @@ class SynchrotronPresence(object):
         logger.info("Presence process_id is %r", self.process_id)
 
     def send_user_sync(self, user_id, is_syncing, last_sync_ms):
+        return
         self.hs.get_tcp_replication().send_user_sync(user_id, is_syncing, last_sync_ms)
 
     def mark_as_coming_online(self, user_id):
@@ -210,6 +211,8 @@ class SynchrotronPresence(object):
         yield self.notify_from_replication(states, stream_id)
 
     def get_currently_syncing_users(self):
+        # presence is disabled on matrix.org, so we return the empty set
+        return set()
         return [
             user_id for user_id, count in self.user_to_num_current_syncs.iteritems()
             if count > 0
diff --git a/synapse/app/synctl.py b/synapse/app/synctl.py
index 712dfa870e..8c525459c7 100755
--- a/synapse/app/synctl.py
+++ b/synapse/app/synctl.py
@@ -108,7 +108,7 @@ def stop(pidfile, app):
 
 
 Worker = collections.namedtuple("Worker", [
-    "app", "configfile", "pidfile", "cache_factor"
+    "app", "configfile", "pidfile", "cache_factor", "cache_factors",
 ])
 
 
@@ -171,6 +171,10 @@ def main():
     if cache_factor:
         os.environ["SYNAPSE_CACHE_FACTOR"] = str(cache_factor)
 
+    cache_factors = config.get("synctl_cache_factors", {})
+    for cache_name, factor in cache_factors.iteritems():
+        os.environ["SYNAPSE_CACHE_FACTOR_" + cache_name.upper()] = str(factor)
+
     worker_configfiles = []
     if options.worker:
         start_stop_synapse = False
@@ -211,6 +215,10 @@ def main():
                 or pidfile
             )
             worker_cache_factor = worker_config.get("synctl_cache_factor") or cache_factor
+            worker_cache_factors = (
+                worker_config.get("synctl_cache_factors")
+                or cache_factors
+            )
             daemonize = worker_config.get("daemonize") or config.get("daemonize")
             assert daemonize, "Main process must have daemonize set to true"
 
@@ -226,8 +234,10 @@ def main():
             assert worker_daemonize, "In config %r: expected '%s' to be True" % (
                 worker_configfile, "worker_daemonize")
             worker_cache_factor = worker_config.get("synctl_cache_factor")
+            worker_cache_factors = worker_config.get("synctl_cache_factors", {})
         workers.append(Worker(
             worker_app, worker_configfile, worker_pidfile, worker_cache_factor,
+            worker_cache_factors,
         ))
 
     action = options.action
@@ -262,15 +272,19 @@ def main():
             start(configfile)
 
         for worker in workers:
+            env = os.environ.copy()
+
             if worker.cache_factor:
                 os.environ["SYNAPSE_CACHE_FACTOR"] = str(worker.cache_factor)
 
+            for cache_name, factor in worker.cache_factors.iteritems():
+                os.environ["SYNAPSE_CACHE_FACTOR_" + cache_name.upper()] = str(factor)
+
             start_worker(worker.app, configfile, worker.configfile)
 
-            if cache_factor:
-                os.environ["SYNAPSE_CACHE_FACTOR"] = str(cache_factor)
-            else:
-                os.environ.pop("SYNAPSE_CACHE_FACTOR", None)
+            # Reset env back to the original
+            os.environ.clear()
+            os.environ.update(env)
 
 
 if __name__ == "__main__":
diff --git a/synapse/federation/transaction_queue.py b/synapse/federation/transaction_queue.py
index a141ec9953..6da60718a9 100644
--- a/synapse/federation/transaction_queue.py
+++ b/synapse/federation/transaction_queue.py
@@ -184,17 +184,22 @@ class TransactionQueue(object):
                     if not is_mine and send_on_behalf_of is None:
                         continue
 
-                    # Get the state from before the event.
-                    # We need to make sure that this is the state from before
-                    # the event and not from after it.
-                    # Otherwise if the last member on a server in a room is
-                    # banned then it won't receive the event because it won't
-                    # be in the room after the ban.
-                    destinations = yield self.state.get_current_hosts_in_room(
-                        event.room_id, latest_event_ids=[
-                            prev_id for prev_id, _ in event.prev_events
-                        ],
-                    )
+                    try:
+                        # Get the state from before the event.
+                        # We need to make sure that this is the state from before
+                        # the event and not from after it.
+                        # Otherwise if the last member on a server in a room is
+                        # banned then it won't receive the event because it won't
+                        # be in the room after the ban.
+                        destinations = yield self.state.get_current_hosts_in_room(
+                            event.room_id, latest_event_ids=[
+                                prev_id for prev_id, _ in event.prev_events
+                            ],
+                        )
+                    except Exception:
+                        logger.exception("Failed to calculate hosts in room")
+                        continue
+
                     destinations = set(destinations)
 
                     if send_on_behalf_of is not None:
@@ -254,6 +259,7 @@ class TransactionQueue(object):
         Args:
             states (list(UserPresenceState))
         """
+        return
 
         # First we queue up the new presence by user ID, so multiple presence
         # updates in quick successtion are correctly handled
diff --git a/synapse/handlers/initial_sync.py b/synapse/handlers/initial_sync.py
index c5267b4b84..df56252185 100644
--- a/synapse/handlers/initial_sync.py
+++ b/synapse/handlers/initial_sync.py
@@ -372,6 +372,7 @@ class InitialSyncHandler(BaseHandler):
 
         @defer.inlineCallbacks
         def get_presence():
+            defer.returnValue([])
             states = yield presence_handler.get_states(
                 [m.user_id for m in room_members],
                 as_event=True,
diff --git a/synapse/handlers/message.py b/synapse/handlers/message.py
index 6de6e13b7b..2134741719 100644
--- a/synapse/handlers/message.py
+++ b/synapse/handlers/message.py
@@ -384,7 +384,7 @@ class MessageHandler(BaseHandler):
         # If this is an AS, double check that they are allowed to see the members.
         # This can either be because the AS user is in the room or becuase there
         # is a user in the room that the AS is "interested in"
-        if requester.app_service and user_id not in users_with_profile:
+        if False and requester.app_service and user_id not in users_with_profile:
             for uid in users_with_profile:
                 if requester.app_service.is_interested_in_user(uid):
                     break
diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py
index a5e501897c..18ab4b5c3b 100644
--- a/synapse/handlers/presence.py
+++ b/synapse/handlers/presence.py
@@ -373,6 +373,7 @@ class PresenceHandler(object):
         """We've seen the user do something that indicates they're interacting
         with the app.
         """
+        return
         user_id = user.to_string()
 
         bump_active_time_counter.inc()
@@ -402,6 +403,7 @@ class PresenceHandler(object):
                 Useful for streams that are not associated with an actual
                 client that is being used by a user.
         """
+        affect_presence = False
         if affect_presence:
             curr_sync = self.user_to_num_current_syncs.get(user_id, 0)
             self.user_to_num_current_syncs[user_id] = curr_sync + 1
@@ -444,6 +446,8 @@ class PresenceHandler(object):
         Returns:
             set(str): A set of user_id strings.
         """
+        # presence is disabled on matrix.org, so we return the empty set
+        return set()
         syncing_user_ids = {
             user_id for user_id, count in self.user_to_num_current_syncs.items()
             if count
@@ -463,6 +467,7 @@ class PresenceHandler(object):
             syncing_user_ids(set(str)): The set of user_ids that are
                 currently syncing on that server.
         """
+        return
 
         # Grab the previous list of user_ids that were syncing on that process
         prev_syncing_user_ids = (
diff --git a/synapse/handlers/room_list.py b/synapse/handlers/room_list.py
index 5d81f59b44..f5657a3521 100644
--- a/synapse/handlers/room_list.py
+++ b/synapse/handlers/room_list.py
@@ -44,7 +44,7 @@ EMTPY_THIRD_PARTY_ID = ThirdPartyInstanceID(None, None)
 class RoomListHandler(BaseHandler):
     def __init__(self, hs):
         super(RoomListHandler, self).__init__(hs)
-        self.response_cache = ResponseCache(hs)
+        self.response_cache = ResponseCache(hs, timeout_ms=10 * 60 * 1000)
         self.remote_response_cache = ResponseCache(hs, timeout_ms=30 * 1000)
 
     def get_local_public_room_list(self, limit=None, since_token=None,
diff --git a/synapse/handlers/room_member.py b/synapse/handlers/room_member.py
index 9977be8831..51206c6023 100644
--- a/synapse/handlers/room_member.py
+++ b/synapse/handlers/room_member.py
@@ -28,7 +28,7 @@ from synapse.api.constants import (
 )
 from synapse.api.errors import AuthError, SynapseError, Codes
 from synapse.types import UserID, RoomID
-from synapse.util.async import Linearizer
+from synapse.util.async import Linearizer, Limiter
 from synapse.util.distributor import user_left_room, user_joined_room
 
 
@@ -60,6 +60,7 @@ class RoomMemberHandler(object):
         self.event_creation_hander = hs.get_event_creation_handler()
 
         self.member_linearizer = Linearizer(name="member")
+        self.member_limiter = Limiter(20)
 
         self.clock = hs.get_clock()
         self.spam_checker = hs.get_spam_checker()
@@ -232,18 +233,23 @@ class RoomMemberHandler(object):
     ):
         key = (room_id,)
 
-        with (yield self.member_linearizer.queue(key)):
-            result = yield self._update_membership(
-                requester,
-                target,
-                room_id,
-                action,
-                txn_id=txn_id,
-                remote_room_hosts=remote_room_hosts,
-                third_party_signed=third_party_signed,
-                ratelimit=ratelimit,
-                content=content,
-            )
+        as_id = object()
+        if requester.app_service:
+            as_id = requester.app_service.id
+
+        with (yield self.member_limiter.queue(as_id)):
+            with (yield self.member_linearizer.queue(key)):
+                result = yield self._update_membership(
+                    requester,
+                    target,
+                    room_id,
+                    action,
+                    txn_id=txn_id,
+                    remote_room_hosts=remote_room_hosts,
+                    third_party_signed=third_party_signed,
+                    ratelimit=ratelimit,
+                    content=content,
+                )
 
         defer.returnValue(result)
 
diff --git a/synapse/handlers/sync.py b/synapse/handlers/sync.py
index 0f713ce038..37fae43f71 100644
--- a/synapse/handlers/sync.py
+++ b/synapse/handlers/sync.py
@@ -30,6 +30,7 @@ import itertools
 
 logger = logging.getLogger(__name__)
 
+SYNC_RESPONSE_CACHE_MS = 2 * 60 * 1000
 
 SyncConfig = collections.namedtuple("SyncConfig", [
     "user",
@@ -169,7 +170,9 @@ class SyncHandler(object):
         self.presence_handler = hs.get_presence_handler()
         self.event_sources = hs.get_event_sources()
         self.clock = hs.get_clock()
-        self.response_cache = ResponseCache(hs)
+        self.response_cache = ResponseCache(
+            hs, timeout_ms=SYNC_RESPONSE_CACHE_MS,
+        )
         self.state = hs.get_state_handler()
 
     def wait_for_sync_for_user(self, sync_config, since_token=None, timeout=0,
@@ -599,7 +602,7 @@ class SyncHandler(object):
             since_token is None and
             sync_config.filter_collection.blocks_all_presence()
         )
-        if not block_all_presence_data:
+        if False and not block_all_presence_data:
             yield self._generate_sync_entry_for_presence(
                 sync_result_builder, newly_joined_rooms, newly_joined_users
             )
diff --git a/synapse/replication/slave/storage/client_ips.py b/synapse/replication/slave/storage/client_ips.py
index 352c9a2aa8..fedf7a3188 100644
--- a/synapse/replication/slave/storage/client_ips.py
+++ b/synapse/replication/slave/storage/client_ips.py
@@ -42,6 +42,8 @@ class SlavedClientIpStore(BaseSlavedStore):
         if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY:
             return
 
+        self.client_ip_last_seen.prefill(key, now)
+
         self.hs.get_tcp_replication().send_user_ip(
             user_id, access_token, ip, user_agent, device_id, now
         )
diff --git a/synapse/replication/tcp/streams.py b/synapse/replication/tcp/streams.py
index 4c60bf79f9..07d99dd63c 100644
--- a/synapse/replication/tcp/streams.py
+++ b/synapse/replication/tcp/streams.py
@@ -33,7 +33,7 @@ import logging
 logger = logging.getLogger(__name__)
 
 
-MAX_EVENTS_BEHIND = 10000
+MAX_EVENTS_BEHIND = 500000
 
 
 EventStreamRow = namedtuple("EventStreamRow", (
diff --git a/synapse/rest/client/v1/admin.py b/synapse/rest/client/v1/admin.py
index 303419d281..ddba0f0fbc 100644
--- a/synapse/rest/client/v1/admin.py
+++ b/synapse/rest/client/v1/admin.py
@@ -296,17 +296,6 @@ class ShutdownRoomRestServlet(ClientV1RestServlet):
         )
         new_room_id = info["room_id"]
 
-        yield self.event_creation_handler.create_and_send_nonmember_event(
-            room_creator_requester,
-            {
-                "type": "m.room.message",
-                "content": {"body": message, "msgtype": "m.text"},
-                "room_id": new_room_id,
-                "sender": new_room_user_id,
-            },
-            ratelimit=False,
-        )
-
         requester_user_id = requester.user.to_string()
 
         logger.info("Shutting down room %r", room_id)
@@ -344,6 +333,17 @@ class ShutdownRoomRestServlet(ClientV1RestServlet):
 
             kicked_users.append(user_id)
 
+        yield self.event_creation_handler.create_and_send_nonmember_event(
+            room_creator_requester,
+            {
+                "type": "m.room.message",
+                "content": {"body": message, "msgtype": "m.text"},
+                "room_id": new_room_id,
+                "sender": new_room_user_id,
+            },
+            ratelimit=False,
+        )
+
         aliases_for_room = yield self.store.get_aliases_for_room(room_id)
 
         yield self.store.update_aliases_for_room(
diff --git a/synapse/rest/client/v1/presence.py b/synapse/rest/client/v1/presence.py
index 4a73813c58..b7e765fa0e 100644
--- a/synapse/rest/client/v1/presence.py
+++ b/synapse/rest/client/v1/presence.py
@@ -81,7 +81,7 @@ class PresenceStatusRestServlet(ClientV1RestServlet):
         except Exception:
             raise SynapseError(400, "Unable to parse state")
 
-        yield self.presence_handler.set_state(user, state)
+        # yield self.presence_handler.set_state(user, state)
 
         defer.returnValue((200, {}))
 
diff --git a/synapse/storage/client_ips.py b/synapse/storage/client_ips.py
index 7b44dae0fc..0bbbedb237 100644
--- a/synapse/storage/client_ips.py
+++ b/synapse/storage/client_ips.py
@@ -28,7 +28,7 @@ logger = logging.getLogger(__name__)
 # Number of msec of granularity to store the user IP 'last seen' time. Smaller
 # times give more inserts into the database even for readonly API hits
 # 120 seconds == 2 minutes
-LAST_SEEN_GRANULARITY = 120 * 1000
+LAST_SEEN_GRANULARITY = 10 * 60 * 1000
 
 
 class ClientIpStore(background_updates.BackgroundUpdateStore):
diff --git a/synapse/storage/event_push_actions.py b/synapse/storage/event_push_actions.py
index e78f8d0114..31719fe1b5 100644
--- a/synapse/storage/event_push_actions.py
+++ b/synapse/storage/event_push_actions.py
@@ -613,6 +613,9 @@ class EventPushActionsStore(EventPushActionsWorkerStore):
             self._rotate_notifs, 30 * 60 * 1000
         )
 
+        self._rotate_delay = 3
+        self._rotate_count = 10000
+
     def _set_push_actions_for_event_and_users_txn(self, txn, events_and_contexts,
                                                   all_events_and_contexts):
         """Handles moving push actions from staging table to main
@@ -809,7 +812,7 @@ class EventPushActionsStore(EventPushActionsWorkerStore):
                 )
                 if caught_up:
                     break
-                yield sleep(5)
+                yield sleep(self._rotate_delay)
         finally:
             self._doing_notif_rotation = False
 
@@ -830,8 +833,8 @@ class EventPushActionsStore(EventPushActionsWorkerStore):
         txn.execute("""
             SELECT stream_ordering FROM event_push_actions
             WHERE stream_ordering > ?
-            ORDER BY stream_ordering ASC LIMIT 1 OFFSET 50000
-        """, (old_rotate_stream_ordering,))
+            ORDER BY stream_ordering ASC LIMIT 1 OFFSET ?
+        """, (old_rotate_stream_ordering, self._rotate_count))
         stream_row = txn.fetchone()
         if stream_row:
             offset_stream_ordering, = stream_row
diff --git a/synapse/storage/events.py b/synapse/storage/events.py
index ece5e6c41f..5d1ef3ea46 100644
--- a/synapse/storage/events.py
+++ b/synapse/storage/events.py
@@ -1010,7 +1010,6 @@ class EventsStore(EventsWorkerStore):
                 "event_edge_hashes",
                 "event_edges",
                 "event_forward_extremities",
-                "event_push_actions",
                 "event_reference_hashes",
                 "event_search",
                 "event_signatures",
@@ -1030,6 +1029,14 @@ class EventsStore(EventsWorkerStore):
                 [(ev.event_id,) for ev, _ in events_and_contexts]
             )
 
+        for table in (
+            "event_push_actions",
+        ):
+            txn.executemany(
+                "DELETE FROM %s WHERE room_id = ? AND event_id = ?" % (table,),
+                [(ev.event_id,) for ev, _ in events_and_contexts]
+            )
+
     def _store_event_txn(self, txn, events_and_contexts):
         """Insert new events into the event and event_json tables
 
diff --git a/synapse/storage/roommember.py b/synapse/storage/roommember.py
index 6a861943a2..c353ef4eb9 100644
--- a/synapse/storage/roommember.py
+++ b/synapse/storage/roommember.py
@@ -65,7 +65,7 @@ class RoomMemberWorkerStore(EventsWorkerStore):
         hosts = frozenset(get_domain_from_id(user_id) for user_id in user_ids)
         defer.returnValue(hosts)
 
-    @cached(max_entries=100000, iterable=True)
+    @cachedInlineCallbacks(max_entries=100000, iterable=True)
     def get_users_in_room(self, room_id):
         def f(txn):
             sql = (
@@ -79,7 +79,14 @@ class RoomMemberWorkerStore(EventsWorkerStore):
 
             txn.execute(sql, (room_id, Membership.JOIN,))
             return [to_ascii(r[0]) for r in txn]
-        return self.runInteraction("get_users_in_room", f)
+        start_time = self._clock.time_msec()
+        result = yield self.runInteraction("get_users_in_room", f)
+        end_time = self._clock.time_msec()
+        logger.info(
+            "Fetched room membership for %s (%i users) in %i ms",
+            room_id, len(result), end_time - start_time,
+        )
+        defer.returnValue(result)
 
     @cached()
     def get_invited_rooms_for_user(self, user_id):
@@ -453,7 +460,7 @@ class RoomMemberWorkerStore(EventsWorkerStore):
 
         defer.returnValue(joined_hosts)
 
-    @cached(max_entries=10000, iterable=True)
+    @cached(max_entries=10000)
     def _get_joined_hosts_cache(self, room_id):
         return _JoinedHostsCache(self, room_id)
 
diff --git a/synapse/storage/search.py b/synapse/storage/search.py
index 426cbe6e1a..134f8dc870 100644
--- a/synapse/storage/search.py
+++ b/synapse/storage/search.py
@@ -721,7 +721,7 @@ def _parse_query(database_engine, search_term):
     results = re.findall(r"([\w\-]+)", search_term, re.UNICODE)
 
     if isinstance(database_engine, PostgresEngine):
-        return " & ".join(result + ":*" for result in results)
+        return " & ".join(result for result in results)
     elif isinstance(database_engine, Sqlite3Engine):
         return " & ".join(result + "*" for result in results)
     else:
diff --git a/synapse/storage/state.py b/synapse/storage/state.py
index ffa4246031..1123f5010b 100644
--- a/synapse/storage/state.py
+++ b/synapse/storage/state.py
@@ -20,7 +20,7 @@ from twisted.internet import defer
 
 from synapse.storage.background_updates import BackgroundUpdateStore
 from synapse.storage.engines import PostgresEngine
-from synapse.util.caches import intern_string, CACHE_SIZE_FACTOR
+from synapse.util.caches import intern_string, get_cache_factor_for
 from synapse.util.caches.descriptors import cached, cachedList
 from synapse.util.caches.dictionary_cache import DictionaryCache
 from synapse.util.stringutils import to_ascii
@@ -54,7 +54,7 @@ class StateGroupWorkerStore(SQLBaseStore):
         super(StateGroupWorkerStore, self).__init__(db_conn, hs)
 
         self._state_group_cache = DictionaryCache(
-            "*stateGroupCache*", 100000 * CACHE_SIZE_FACTOR
+            "*stateGroupCache*", 500000 * get_cache_factor_for("stateGroupCache")
         )
 
     @cached(max_entries=100000, iterable=True)
@@ -566,8 +566,7 @@ class StateGroupWorkerStore(SQLBaseStore):
                 state_dict = results[group]
 
                 state_dict.update(
-                    ((intern_string(k[0]), intern_string(k[1])), to_ascii(v))
-                    for k, v in group_state_dict.iteritems()
+                    group_state_dict.iteritems()
                 )
 
                 self._state_group_cache.update(
diff --git a/synapse/util/caches/__init__.py b/synapse/util/caches/__init__.py
index 4adae96681..6beec6a25e 100644
--- a/synapse/util/caches/__init__.py
+++ b/synapse/util/caches/__init__.py
@@ -18,6 +18,16 @@ import os
 
 CACHE_SIZE_FACTOR = float(os.environ.get("SYNAPSE_CACHE_FACTOR", 0.5))
 
+
+def get_cache_factor_for(cache_name):
+    env_var = "SYNAPSE_CACHE_FACTOR_" + cache_name.upper()
+    factor = os.environ.get(env_var)
+    if factor:
+        return float(factor)
+
+    return CACHE_SIZE_FACTOR
+
+
 metrics = synapse.metrics.get_metrics_for("synapse.util.caches")
 
 caches_by_name = {}
diff --git a/synapse/util/caches/descriptors.py b/synapse/util/caches/descriptors.py
index 68285a7594..f9bb8fbcf9 100644
--- a/synapse/util/caches/descriptors.py
+++ b/synapse/util/caches/descriptors.py
@@ -17,7 +17,7 @@ import logging
 
 from synapse.util.async import ObservableDeferred
 from synapse.util import unwrapFirstError, logcontext
-from synapse.util.caches import CACHE_SIZE_FACTOR
+from synapse.util.caches import get_cache_factor_for
 from synapse.util.caches.lrucache import LruCache
 from synapse.util.caches.treecache import TreeCache, iterate_tree_cache_entry
 from synapse.util.stringutils import to_ascii
@@ -310,7 +310,7 @@ class CacheDescriptor(_CacheDescriptorBase):
             orig, num_args=num_args, inlineCallbacks=inlineCallbacks,
             cache_context=cache_context)
 
-        max_entries = int(max_entries * CACHE_SIZE_FACTOR)
+        max_entries = int(max_entries * get_cache_factor_for(orig.__name__))
 
         self.max_entries = max_entries
         self.tree = tree