Merge branch 'release-v0.9.2'

Thumbnail images on a seperate thread

CHANGES.rst

@@ -1,3 +1,26 @@
+Changes in synapse v0.9.2 (2015-06-12)
+======================================
+
+General:
+
+* Use ultrajson for json (de)serialisation when a canonical encoding is not
+  required. Ultrajson is significantly faster than simplejson in certain
+  circumstances.
+* Use connection pools for outgoing HTTP connections.
+* Process thumbnails on separate threads.
+
+Configuration:
+
+* Add option, ``gzip_responses``, to disable HTTP response compression.
+
+Federation:
+
+* Improve resilience of backfill by ensuring we fetch any missing auth events.
+* Improve performance of backfill and joining remote rooms by removing
+  unnecessary computations. This included handling events we'd previously
+  handled as well as attempting to compute the current state for outliers.
+
+
 Changes in synapse v0.9.1 (2015-05-26)
 ======================================
 

synapse/__init__.py

@@ -16,4 +16,4 @@
 """ This is a reference implementation of a Matrix home server.
 """
 
-__version__ = "0.9.1"
+__version__ = "0.9.2"

synapse/app/homeserver.py

@@ -87,10 +87,16 @@ class SynapseHomeServer(HomeServer):
         return MatrixFederationHttpClient(self)
 
     def build_resource_for_client(self):
-        return gz_wrap(ClientV1RestResource(self))
+        res = ClientV1RestResource(self)
+        if self.config.gzip_responses:
+            res = gz_wrap(res)
+        return res
 
     def build_resource_for_client_v2_alpha(self):
-        return gz_wrap(ClientV2AlphaRestResource(self))
+        res = ClientV2AlphaRestResource(self)
+        if self.config.gzip_responses:
+            res = gz_wrap(res)
+        return res
 
     def build_resource_for_federation(self):
         return JsonResource(self)

synapse/config/server.py

@@ -29,6 +29,7 @@ class ServerConfig(Config):
         self.soft_file_limit = config["soft_file_limit"]
         self.daemonize = config.get("daemonize")
         self.use_frozen_dicts = config.get("use_frozen_dicts", True)
+        self.gzip_responses = config["gzip_responses"]
 
         # Attempt to guess the content_addr for the v0 content repostitory
         content_addr = config.get("content_addr")
@@ -86,6 +87,11 @@ class ServerConfig(Config):
         # Turn on the twisted telnet manhole service on localhost on the given
         # port.
         #manhole: 9000
+
+        # Should synapse compress HTTP responses to clients that support it?
+        # This should be disabled if running synapse behind a load balancer
+        # that can do automatic compression.
+        gzip_responses: True
         """ % locals()
 
     def read_arguments(self, args):

synapse/federation/federation_base.py

@@ -18,8 +18,6 @@ from twisted.internet import defer
 
 from synapse.events.utils import prune_event
 
-from syutil.jsonutil import encode_canonical_json
-
 from synapse.crypto.event_signing import check_event_content_hash
 
 from synapse.api.errors import SynapseError
@@ -120,16 +118,15 @@ class FederationBase(object):
             )
         except SynapseError:
             logger.warn(
-                "Signature check failed for %s redacted to %s",
-                encode_canonical_json(pdu.get_pdu_json()),
-                encode_canonical_json(redacted_pdu_json),
+                "Signature check failed for %s",
+                pdu.event_id,
             )
             raise
 
         if not check_event_content_hash(pdu):
             logger.warn(
-                "Event content has been tampered, redacting %s, %s",
-                pdu.event_id, encode_canonical_json(pdu.get_dict())
+                "Event content has been tampered, redacting.",
+                pdu.event_id,
             )
             defer.returnValue(redacted_event)
 

synapse/federation/transport/server.py

@@ -93,6 +93,8 @@ class TransportLayerServer(object):
 
         yield self.keyring.verify_json_for_server(origin, json_request)
 
+        logger.info("Request from %s", origin)
+
         defer.returnValue((origin, content))
 
     @log_function

synapse/handlers/_base.py

@@ -78,7 +78,9 @@ class BaseHandler(object):
         context = yield state_handler.compute_event_context(builder)
 
         if builder.is_state():
-            builder.prev_state = context.prev_state_events
+            builder.prev_state = yield self.store.add_event_hashes(
+                context.prev_state_events
+            )
 
         yield self.auth.add_auth_events(builder, context)
 

synapse/handlers/federation.py

@@ -247,9 +247,15 @@ class FederationHandler(BaseHandler):
             if set(e_id for e_id, _ in ev.prev_events) - event_ids
         ]
 
+        logger.info(
+            "backfill: Got %d events with %d edges",
+            len(events), len(edges),
+        )
+
         # For each edge get the current state.
 
         auth_events = {}
+        state_events = {}
         events_to_state = {}
         for e_id in edges:
             state, auth = yield self.replication_layer.get_state_for_room(
@@ -258,12 +264,46 @@ class FederationHandler(BaseHandler):
                 event_id=e_id
             )
             auth_events.update({a.event_id: a for a in auth})
+            auth_events.update({s.event_id: s for s in state})
+            state_events.update({s.event_id: s for s in state})
             events_to_state[e_id] = state
 
+        seen_events = yield self.store.have_events(
+            set(auth_events.keys()) | set(state_events.keys())
+        )
+
+        all_events = events + state_events.values() + auth_events.values()
+        required_auth = set(
+            a_id for event in all_events for a_id, _ in event.auth_events
+        )
+
+        missing_auth = required_auth - set(auth_events)
+        results = yield defer.gatherResults(
+            [
+                self.replication_layer.get_pdu(
+                    [dest],
+                    event_id,
+                    outlier=True,
+                    timeout=10000,
+                )
+                for event_id in missing_auth
+            ],
+            consumeErrors=True
+        ).addErrback(unwrapFirstError)
+        auth_events.update({a.event_id: a for a in results})
+
         yield defer.gatherResults(
             [
-                self._handle_new_event(dest, a)
+                self._handle_new_event(
+                    dest, a,
+                    auth_events={
+                        (auth_events[a_id].type, auth_events[a_id].state_key):
+                        auth_events[a_id]
+                        for a_id, _ in a.auth_events
+                    },
+                )
                 for a in auth_events.values()
+                if a.event_id not in seen_events
             ],
             consumeErrors=True,
         ).addErrback(unwrapFirstError)
@@ -274,6 +314,11 @@ class FederationHandler(BaseHandler):
                     dest, event_map[e_id],
                     state=events_to_state[e_id],
                     backfilled=True,
+                    auth_events={
+                        (auth_events[a_id].type, auth_events[a_id].state_key):
+                        auth_events[a_id]
+                        for a_id, _ in event_map[e_id].auth_events
+                    },
                 )
                 for e_id in events_to_state
             ],
@@ -900,8 +945,10 @@ class FederationHandler(BaseHandler):
             event.event_id, event.signatures,
         )
 
+        outlier = event.internal_metadata.is_outlier()
+
         context = yield self.state_handler.compute_event_context(
-            event, old_state=state
+            event, old_state=state, outlier=outlier,
         )
 
         if not auth_events:
@@ -912,7 +959,7 @@ class FederationHandler(BaseHandler):
             event.event_id, auth_events,
         )
 
-        is_new_state = not event.internal_metadata.is_outlier()
+        is_new_state = not outlier
 
         # This is a hack to fix some old rooms where the initial join event
         # didn't reference the create event in its auth events.

synapse/push/__init__.py

@@ -24,6 +24,7 @@ import baserules
 import logging
 import simplejson as json
 import re
+import random
 
 logger = logging.getLogger(__name__)
 
@@ -256,134 +257,154 @@ class Pusher(object):
             logger.info("Pusher %s for user %s starting from token %s",
                         self.pushkey, self.user_name, self.last_token)
 
+        wait = 0
         while self.alive:
-            from_tok = StreamToken.from_string(self.last_token)
-            config = PaginationConfig(from_token=from_tok, limit='1')
-            chunk = yield self.evStreamHandler.get_stream(
-                self.user_name, config,
-                timeout=100*365*24*60*60*1000, affect_presence=False
-            )
-
-            # limiting to 1 may get 1 event plus 1 presence event, so
-            # pick out the actual event
-            single_event = None
-            for c in chunk['chunk']:
-                if 'event_id' in c:  # Hmmm...
-                    single_event = c
-                    break
-            if not single_event:
-                self.last_token = chunk['end']
-                continue
-
-            if not self.alive:
-                continue
-
-            processed = False
-            actions = yield self._actions_for_event(single_event)
-            tweaks = _tweaks_for_actions(actions)
-
-            if len(actions) == 0:
-                logger.warn("Empty actions! Using default action.")
-                actions = Pusher.DEFAULT_ACTIONS
-
-            if 'notify' not in actions and 'dont_notify' not in actions:
-                logger.warn("Neither notify nor dont_notify in actions: adding default")
-                actions.extend(Pusher.DEFAULT_ACTIONS)
-
-            if 'dont_notify' in actions:
-                logger.debug(
-                    "%s for %s: dont_notify",
-                    single_event['event_id'], self.user_name
+            try:
+                if wait > 0:
+                    yield synapse.util.async.sleep(wait)
+                yield self.get_and_dispatch()
+                wait = 0
+            except:
+                if wait == 0:
+                    wait = 1
+                else:
+                    wait = min(wait * 2, 1800)
+                logger.exception(
+                    "Exception in pusher loop for pushkey %s. Pausing for %ds",
+                    self.pushkey, wait
                 )
+
+    @defer.inlineCallbacks
+    def get_and_dispatch(self):
+        from_tok = StreamToken.from_string(self.last_token)
+        config = PaginationConfig(from_token=from_tok, limit='1')
+        timeout = (300 + random.randint(-60, 60)) * 1000
+        chunk = yield self.evStreamHandler.get_stream(
+            self.user_name, config,
+            timeout=timeout, affect_presence=False
+        )
+
+        # limiting to 1 may get 1 event plus 1 presence event, so
+        # pick out the actual event
+        single_event = None
+        for c in chunk['chunk']:
+            if 'event_id' in c:  # Hmmm...
+                single_event = c
+                break
+        if not single_event:
+            self.last_token = chunk['end']
+            logger.debug("Event stream timeout for pushkey %s", self.pushkey)
+            return
+
+        if not self.alive:
+            return
+
+        processed = False
+        actions = yield self._actions_for_event(single_event)
+        tweaks = _tweaks_for_actions(actions)
+
+        if len(actions) == 0:
+            logger.warn("Empty actions! Using default action.")
+            actions = Pusher.DEFAULT_ACTIONS
+
+        if 'notify' not in actions and 'dont_notify' not in actions:
+            logger.warn("Neither notify nor dont_notify in actions: adding default")
+            actions.extend(Pusher.DEFAULT_ACTIONS)
+
+        if 'dont_notify' in actions:
+            logger.debug(
+                "%s for %s: dont_notify",
+                single_event['event_id'], self.user_name
+            )
+            processed = True
+        else:
+            rejected = yield self.dispatch_push(single_event, tweaks)
+            self.has_unread = True
+            if isinstance(rejected, list) or isinstance(rejected, tuple):
                 processed = True
-            else:
-                rejected = yield self.dispatch_push(single_event, tweaks)
-                self.has_unread = True
-                if isinstance(rejected, list) or isinstance(rejected, tuple):
-                    processed = True
-                    for pk in rejected:
-                        if pk != self.pushkey:
-                            # for sanity, we only remove the pushkey if it
-                            # was the one we actually sent...
-                            logger.warn(
-                                ("Ignoring rejected pushkey %s because we"
-                                 " didn't send it"), pk
-                            )
-                        else:
-                            logger.info(
-                                "Pushkey %s was rejected: removing",
-                                pk
-                            )
-                            yield self.hs.get_pusherpool().remove_pusher(
-                                self.app_id, pk, self.user_name
-                            )
+                for pk in rejected:
+                    if pk != self.pushkey:
+                        # for sanity, we only remove the pushkey if it
+                        # was the one we actually sent...
+                        logger.warn(
+                            ("Ignoring rejected pushkey %s because we"
+                             " didn't send it"), pk
+                        )
+                    else:
+                        logger.info(
+                            "Pushkey %s was rejected: removing",
+                            pk
+                        )
+                        yield self.hs.get_pusherpool().remove_pusher(
+                            self.app_id, pk, self.user_name
+                        )
 
-            if not self.alive:
-                continue
+        if not self.alive:
+            return
 
-            if processed:
-                self.backoff_delay = Pusher.INITIAL_BACKOFF
-                self.last_token = chunk['end']
-                self.store.update_pusher_last_token_and_success(
+        if processed:
+            self.backoff_delay = Pusher.INITIAL_BACKOFF
+            self.last_token = chunk['end']
+            self.store.update_pusher_last_token_and_success(
+                self.app_id,
+                self.pushkey,
+                self.user_name,
+                self.last_token,
+                self.clock.time_msec()
+            )
+            if self.failing_since:
+                self.failing_since = None
+                self.store.update_pusher_failing_since(
                     self.app_id,
                     self.pushkey,
                     self.user_name,
-                    self.last_token,
-                    self.clock.time_msec()
+                    self.failing_since)
+        else:
+            if not self.failing_since:
+                self.failing_since = self.clock.time_msec()
+                self.store.update_pusher_failing_since(
+                    self.app_id,
+                    self.pushkey,
+                    self.user_name,
+                    self.failing_since
+                )
+
+            if (self.failing_since and
+               self.failing_since <
+               self.clock.time_msec() - Pusher.GIVE_UP_AFTER):
+                # we really only give up so that if the URL gets
+                # fixed, we don't suddenly deliver a load
+                # of old notifications.
+                logger.warn("Giving up on a notification to user %s, "
+                            "pushkey %s",
+                            self.user_name, self.pushkey)
+                self.backoff_delay = Pusher.INITIAL_BACKOFF
+                self.last_token = chunk['end']
+                self.store.update_pusher_last_token(
+                    self.app_id,
+                    self.pushkey,
+                    self.user_name,
+                    self.last_token
+                )
+
+                self.failing_since = None
+                self.store.update_pusher_failing_since(
+                    self.app_id,
+                    self.pushkey,
+                    self.user_name,
+                    self.failing_since
                 )
-                if self.failing_since:
-                    self.failing_since = None
-                    self.store.update_pusher_failing_since(
-                        self.app_id,
-                        self.pushkey,
-                        self.user_name,
-                        self.failing_since)
             else:
-                if not self.failing_since:
-                    self.failing_since = self.clock.time_msec()
-                    self.store.update_pusher_failing_since(
-                        self.app_id,
-                        self.pushkey,
-                        self.user_name,
-                        self.failing_since
-                    )
-
-                if (self.failing_since and
-                   self.failing_since <
-                   self.clock.time_msec() - Pusher.GIVE_UP_AFTER):
-                    # we really only give up so that if the URL gets
-                    # fixed, we don't suddenly deliver a load
-                    # of old notifications.
-                    logger.warn("Giving up on a notification to user %s, "
-                                "pushkey %s",
-                                self.user_name, self.pushkey)
-                    self.backoff_delay = Pusher.INITIAL_BACKOFF
-                    self.last_token = chunk['end']
-                    self.store.update_pusher_last_token(
-                        self.app_id,
-                        self.pushkey,
-                        self.user_name,
-                        self.last_token
-                    )
-
-                    self.failing_since = None
-                    self.store.update_pusher_failing_since(
-                        self.app_id,
-                        self.pushkey,
-                        self.user_name,
-                        self.failing_since
-                    )
-                else:
-                    logger.warn("Failed to dispatch push for user %s "
-                                "(failing for %dms)."
-                                "Trying again in %dms",
-                                self.user_name,
-                                self.clock.time_msec() - self.failing_since,
-                                self.backoff_delay)
-                    yield synapse.util.async.sleep(self.backoff_delay / 1000.0)
-                    self.backoff_delay *= 2
-                    if self.backoff_delay > Pusher.MAX_BACKOFF:
-                        self.backoff_delay = Pusher.MAX_BACKOFF
+                logger.warn("Failed to dispatch push for user %s "
+                            "(failing for %dms)."
+                            "Trying again in %dms",
+                            self.user_name,
+                            self.clock.time_msec() - self.failing_since,
+                            self.backoff_delay)
+                yield synapse.util.async.sleep(self.backoff_delay / 1000.0)
+                self.backoff_delay *= 2
+                if self.backoff_delay > Pusher.MAX_BACKOFF:
+                    self.backoff_delay = Pusher.MAX_BACKOFF
 
     def stop(self):
         self.alive = False

synapse/rest/media/v1/base_resource.py

@@ -15,13 +15,14 @@
 
 from .thumbnailer import Thumbnailer
 
+from synapse.http.matrixfederationclient import MatrixFederationHttpClient
 from synapse.http.server import respond_with_json
 from synapse.util.stringutils import random_string
 from synapse.api.errors import (
     cs_error, Codes, SynapseError
 )
 
-from twisted.internet import defer
+from twisted.internet import defer, threads
 from twisted.web.resource import Resource
 from twisted.protocols.basic import FileSender
 
@@ -52,7 +53,7 @@ class BaseMediaResource(Resource):
     def __init__(self, hs, filepaths):
         Resource.__init__(self)
         self.auth = hs.get_auth()
-        self.client = hs.get_http_client()
+        self.client = MatrixFederationHttpClient(hs)
         self.clock = hs.get_clock()
         self.server_name = hs.hostname
         self.store = hs.get_datastore()
@@ -273,57 +274,65 @@ class BaseMediaResource(Resource):
         if not requirements:
             return
 
+        remote_thumbnails = []
+
         input_path = self.filepaths.remote_media_filepath(server_name, file_id)
         thumbnailer = Thumbnailer(input_path)
         m_width = thumbnailer.width
         m_height = thumbnailer.height
 
-        if m_width * m_height >= self.max_image_pixels:
-            logger.info(
-                "Image too large to thumbnail %r x %r > %r",
-                m_width, m_height, self.max_image_pixels
-            )
-            return
+        def generate_thumbnails():
+            if m_width * m_height >= self.max_image_pixels:
+                logger.info(
+                    "Image too large to thumbnail %r x %r > %r",
+                    m_width, m_height, self.max_image_pixels
+                )
+                return
 
-        scales = set()
-        crops = set()
-        for r_width, r_height, r_method, r_type in requirements:
-            if r_method == "scale":
-                t_width, t_height = thumbnailer.aspect(r_width, r_height)
-                scales.add((
-                    min(m_width, t_width), min(m_height, t_height), r_type,
-                ))
-            elif r_method == "crop":
-                crops.add((r_width, r_height, r_type))
+            scales = set()
+            crops = set()
+            for r_width, r_height, r_method, r_type in requirements:
+                if r_method == "scale":
+                    t_width, t_height = thumbnailer.aspect(r_width, r_height)
+                    scales.add((
+                        min(m_width, t_width), min(m_height, t_height), r_type,
+                    ))
+                elif r_method == "crop":
+                    crops.add((r_width, r_height, r_type))
 
-        for t_width, t_height, t_type in scales:
-            t_method = "scale"
-            t_path = self.filepaths.remote_media_thumbnail(
-                server_name, file_id, t_width, t_height, t_type, t_method
-            )
-            self._makedirs(t_path)
-            t_len = thumbnailer.scale(t_path, t_width, t_height, t_type)
-            yield self.store.store_remote_media_thumbnail(
-                server_name, media_id, file_id,
-                t_width, t_height, t_type, t_method, t_len
-            )
+            for t_width, t_height, t_type in scales:
+                t_method = "scale"
+                t_path = self.filepaths.remote_media_thumbnail(
+                    server_name, file_id, t_width, t_height, t_type, t_method
+                )
+                self._makedirs(t_path)
+                t_len = thumbnailer.scale(t_path, t_width, t_height, t_type)
+                remote_thumbnails.append([
+                    server_name, media_id, file_id,
+                    t_width, t_height, t_type, t_method, t_len
+                ])
 
-        for t_width, t_height, t_type in crops:
-            if (t_width, t_height, t_type) in scales:
-                # If the aspect ratio of the cropped thumbnail matches a purely
-                # scaled one then there is no point in calculating a separate
-                # thumbnail.
-                continue
-            t_method = "crop"
-            t_path = self.filepaths.remote_media_thumbnail(
-                server_name, file_id, t_width, t_height, t_type, t_method
-            )
-            self._makedirs(t_path)
-            t_len = thumbnailer.crop(t_path, t_width, t_height, t_type)
-            yield self.store.store_remote_media_thumbnail(
-                server_name, media_id, file_id,
-                t_width, t_height, t_type, t_method, t_len
-            )
+            for t_width, t_height, t_type in crops:
+                if (t_width, t_height, t_type) in scales:
+                    # If the aspect ratio of the cropped thumbnail matches a purely
+                    # scaled one then there is no point in calculating a separate
+                    # thumbnail.
+                    continue
+                t_method = "crop"
+                t_path = self.filepaths.remote_media_thumbnail(
+                    server_name, file_id, t_width, t_height, t_type, t_method
+                )
+                self._makedirs(t_path)
+                t_len = thumbnailer.crop(t_path, t_width, t_height, t_type)
+                remote_thumbnails.append([
+                    server_name, media_id, file_id,
+                    t_width, t_height, t_type, t_method, t_len
+                ])
+
+        yield threads.deferToThread(generate_thumbnails)
+
+        for r in remote_thumbnails:
+            yield self.store.store_remote_media_thumbnail(*r)
 
         defer.returnValue({
             "width": m_width,

synapse/state.py

@@ -106,7 +106,7 @@ class StateHandler(object):
         defer.returnValue(state)
 
     @defer.inlineCallbacks
-    def compute_event_context(self, event, old_state=None):
+    def compute_event_context(self, event, old_state=None, outlier=False):
         """ Fills out the context with the `current state` of the graph. The
         `current state` here is defined to be the state of the event graph
         just before the event - i.e. it never includes `event`
@@ -119,9 +119,23 @@ class StateHandler(object):
         Returns:
             an EventContext
         """
+        yield run_on_reactor()
+
         context = EventContext()
 
-        yield run_on_reactor()
+        if outlier:
+            # If this is an outlier, then we know it shouldn't have any current
+            # state. Certainly store.get_current_state won't return any, and
+            # persisting the event won't store the state group.
+            if old_state:
+                context.current_state = {
+                    (s.type, s.state_key): s for s in old_state
+                }
+            else:
+                context.current_state = {}
+            context.prev_state_events = []
+            context.state_group = None
+            defer.returnValue(context)
 
         if old_state:
             context.current_state = {
@@ -155,10 +169,6 @@ class StateHandler(object):
         context.current_state = curr_state
         context.state_group = group if not event.is_state() else None
 
-        prev_state = yield self.store.add_event_hashes(
-            prev_state
-        )
-
         if event.is_state():
             key = (event.type, event.state_key)
             if key in context.current_state:

synapse/storage/__init__.py

@@ -51,7 +51,7 @@ logger = logging.getLogger(__name__)
 
 # Remember to update this number every time a change is made to database
 # schema files, so the users will be informed on server restarts.
-SCHEMA_VERSION = 19
+SCHEMA_VERSION = 20
 
 dir_path = os.path.abspath(os.path.dirname(__file__))
 
@@ -348,7 +348,7 @@ def _upgrade_existing_database(cur, current_version, applied_delta_files,
                         module_name, absolute_path, python_file
                     )
                 logger.debug("Running script %s", relative_path)
-                module.run_upgrade(cur)
+                module.run_upgrade(cur, database_engine)
             elif ext == ".sql":
                 # A plain old .sql file, just read and execute it
                 logger.debug("Applying schema %s", relative_path)

synapse/storage/_base.py

@@ -127,7 +127,7 @@ class Cache(object):
         self.cache.clear()
 
 
-def cached(max_entries=1000, num_args=1, lru=False):
+class CacheDescriptor(object):
     """ A method decorator that applies a memoizing cache around the function.
 
     The function is presumed to take zero or more arguments, which are used in
@@ -141,25 +141,32 @@ def cached(max_entries=1000, num_args=1, lru=False):
     which can be used to insert values into the cache specifically, without
     calling the calculation function.
     """
-    def wrap(orig):
+    def __init__(self, orig, max_entries=1000, num_args=1, lru=False):
+        self.orig = orig
+
+        self.max_entries = max_entries
+        self.num_args = num_args
+        self.lru = lru
+
+    def __get__(self, obj, objtype=None):
         cache = Cache(
-            name=orig.__name__,
-            max_entries=max_entries,
-            keylen=num_args,
-            lru=lru,
+            name=self.orig.__name__,
+            max_entries=self.max_entries,
+            keylen=self.num_args,
+            lru=self.lru,
         )
 
-        @functools.wraps(orig)
+        @functools.wraps(self.orig)
         @defer.inlineCallbacks
-        def wrapped(self, *keyargs):
+        def wrapped(*keyargs):
             try:
-                cached_result = cache.get(*keyargs)
+                cached_result = cache.get(*keyargs[:self.num_args])
                 if DEBUG_CACHES:
-                    actual_result = yield orig(self, *keyargs)
+                    actual_result = yield self.orig(obj, *keyargs)
                     if actual_result != cached_result:
                         logger.error(
                             "Stale cache entry %s%r: cached: %r, actual %r",
-                            orig.__name__, keyargs,
+                            self.orig.__name__, keyargs,
                             cached_result, actual_result,
                         )
                         raise ValueError("Stale cache entry")
@@ -170,18 +177,28 @@ def cached(max_entries=1000, num_args=1, lru=False):
                 # while the SELECT is executing (SYN-369)
                 sequence = cache.sequence
 
-                ret = yield orig(self, *keyargs)
+                ret = yield self.orig(obj, *keyargs)
 
-                cache.update(sequence, *keyargs + (ret,))
+                cache.update(sequence, *keyargs[:self.num_args] + (ret,))
 
                 defer.returnValue(ret)
 
         wrapped.invalidate = cache.invalidate
         wrapped.invalidate_all = cache.invalidate_all
         wrapped.prefill = cache.prefill
+
+        obj.__dict__[self.orig.__name__] = wrapped
+
         return wrapped
 
-    return wrap
+
+def cached(max_entries=1000, num_args=1, lru=False):
+    return lambda orig: CacheDescriptor(
+        orig,
+        max_entries=max_entries,
+        num_args=num_args,
+        lru=lru
+    )
 
 
 class LoggingTransaction(object):

synapse/storage/events.py

@@ -736,7 +736,8 @@ class EventsStore(SQLBaseStore):
 
             because = yield self.get_event(
                 redaction_id,
-                check_redacted=False
+                check_redacted=False,
+                allow_none=True,
             )
 
             if because:
@@ -746,6 +747,7 @@ class EventsStore(SQLBaseStore):
             prev = yield self.get_event(
                 ev.unsigned["replaces_state"],
                 get_prev_content=False,
+                allow_none=True,
             )
             if prev:
                 ev.unsigned["prev_content"] = prev.get_dict()["content"]

synapse/storage/schema/delta/14/upgrade_appservice_db.py

@@ -18,7 +18,7 @@ import logging
 logger = logging.getLogger(__name__)
 
 
-def run_upgrade(cur):
+def run_upgrade(cur, *args, **kwargs):
     cur.execute("SELECT id, regex FROM application_services_regex")
     for row in cur.fetchall():
         try:

synapse/storage/schema/delta/20/pushers.py

@@ -0,0 +1,76 @@
+# Copyright 2015 OpenMarket Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# 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.
+
+
+"""
+Main purpose of this upgrade is to change the unique key on the
+pushers table again (it was missed when the v16 full schema was
+made) but this also changes the pushkey and data columns to text.
+When selecting a bytea column into a text column, postgres inserts
+the hex encoded data, and there's no portable way of getting the
+UTF-8 bytes, so we have to do it in Python.
+"""
+
+import logging
+
+logger = logging.getLogger(__name__)
+
+
+def run_upgrade(cur, database_engine, *args, **kwargs):
+    logger.info("Porting pushers table...")
+    cur.execute("""
+        CREATE TABLE IF NOT EXISTS pushers2 (
+          id BIGINT PRIMARY KEY,
+          user_name TEXT NOT NULL,
+          access_token BIGINT DEFAULT NULL,
+          profile_tag VARCHAR(32) NOT NULL,
+          kind VARCHAR(8) NOT NULL,
+          app_id VARCHAR(64) NOT NULL,
+          app_display_name VARCHAR(64) NOT NULL,
+          device_display_name VARCHAR(128) NOT NULL,
+          pushkey TEXT NOT NULL,
+          ts BIGINT NOT NULL,
+          lang VARCHAR(8),
+          data TEXT,
+          last_token TEXT,
+          last_success BIGINT,
+          failing_since BIGINT,
+          UNIQUE (app_id, pushkey, user_name)
+        )
+    """)
+    cur.execute("""SELECT
+        id, user_name, access_token, profile_tag, kind,
+        app_id, app_display_name, device_display_name,
+        pushkey, ts, lang, data, last_token, last_success,
+        failing_since
+        FROM pushers
+    """)
+    count = 0
+    for row in cur.fetchall():
+        row = list(row)
+        row[8] = bytes(row[8]).decode("utf-8")
+        row[11] = bytes(row[11]).decode("utf-8")
+        cur.execute(database_engine.convert_param_style("""
+            INSERT into pushers2 (
+            id, user_name, access_token, profile_tag, kind,
+            app_id, app_display_name, device_display_name,
+            pushkey, ts, lang, data, last_token, last_success,
+            failing_since
+            ) values (%s)""" % (','.join(['?' for _ in range(len(row))]))),
+            row
+        )
+        count += 1
+    cur.execute("DROP TABLE pushers")
+    cur.execute("ALTER TABLE pushers2 RENAME TO pushers")
+    logger.info("Moved %d pushers to new table", count)

synapse/storage/state.py

@@ -81,19 +81,23 @@ class StateStore(SQLBaseStore):
             f,
         )
 
-        @defer.inlineCallbacks
-        def c(vals):
-            vals[:] = yield self._get_events(vals, get_prev_content=False)
-
-        yield defer.gatherResults(
+        state_list = yield defer.gatherResults(
             [
-                c(vals)
-                for vals in states.values()
+                self._fetch_events_for_group(group, vals)
+                for group, vals in states.items()
             ],
             consumeErrors=True,
         )
 
-        defer.returnValue(states)
+        defer.returnValue(dict(state_list))
+
+    @cached(num_args=1)
+    def _fetch_events_for_group(self, state_group, events):
+        return self._get_events(
+            events, get_prev_content=False
+        ).addCallback(
+            lambda evs: (state_group, evs)
+        )
 
     def _store_state_groups_txn(self, txn, event, context):
         if context.current_state is None:

tests/handlers/test_federation.py

@@ -100,7 +100,7 @@ class FederationTestCase(unittest.TestCase):
             return defer.succeed({})
         self.datastore.have_events.side_effect = have_events
 
-        def annotate(ev, old_state=None):
+        def annotate(ev, old_state=None, outlier=False):
             context = Mock()
             context.current_state = {}
             context.auth_events = {}
@@ -120,7 +120,7 @@ class FederationTestCase(unittest.TestCase):
         )
 
         self.state_handler.compute_event_context.assert_called_once_with(
-            ANY, old_state=None,
+            ANY, old_state=None, outlier=False
         )
 
         self.auth.check.assert_called_once_with(ANY, auth_events={})

tests/handlers/test_room.py

@@ -42,6 +42,7 @@ class RoomMemberHandlerTestCase(unittest.TestCase):
                 "get_room",
                 "store_room",
                 "get_latest_events_in_room",
+                "add_event_hashes",
             ]),
             resource_for_federation=NonCallableMock(),
             http_client=NonCallableMock(spec_set=[]),
@@ -88,6 +89,7 @@ class RoomMemberHandlerTestCase(unittest.TestCase):
         self.ratelimiter.send_message.return_value = (True, 0)
 
         self.datastore.persist_event.return_value = (1,1)
+        self.datastore.add_event_hashes.return_value = []
 
     @defer.inlineCallbacks
     def test_invite(self):

tests/storage/test__base.py

@@ -96,73 +96,84 @@ class CacheDecoratorTestCase(unittest.TestCase):
 
     @defer.inlineCallbacks
     def test_passthrough(self):
-        @cached()
-        def func(self, key):
-            return key
+        class A(object):
+            @cached()
+            def func(self, key):
+                return key
 
-        self.assertEquals((yield func(self, "foo")), "foo")
-        self.assertEquals((yield func(self, "bar")), "bar")
+        a = A()
+
+        self.assertEquals((yield a.func("foo")), "foo")
+        self.assertEquals((yield a.func("bar")), "bar")
 
     @defer.inlineCallbacks
     def test_hit(self):
         callcount = [0]
 
-        @cached()
-        def func(self, key):
-            callcount[0] += 1
-            return key
+        class A(object):
+            @cached()
+            def func(self, key):
+                callcount[0] += 1
+                return key
 
-        yield func(self, "foo")
+        a = A()
+        yield a.func("foo")
 
         self.assertEquals(callcount[0], 1)
 
-        self.assertEquals((yield func(self, "foo")), "foo")
+        self.assertEquals((yield a.func("foo")), "foo")
         self.assertEquals(callcount[0], 1)
 
     @defer.inlineCallbacks
     def test_invalidate(self):
         callcount = [0]
 
-        @cached()
-        def func(self, key):
-            callcount[0] += 1
-            return key
+        class A(object):
+            @cached()
+            def func(self, key):
+                callcount[0] += 1
+                return key
 
-        yield func(self, "foo")
+        a = A()
+        yield a.func("foo")
 
         self.assertEquals(callcount[0], 1)
 
-        func.invalidate("foo")
+        a.func.invalidate("foo")
 
-        yield func(self, "foo")
+        yield a.func("foo")
 
         self.assertEquals(callcount[0], 2)
 
     def test_invalidate_missing(self):
-        @cached()
-        def func(self, key):
-            return key
+        class A(object):
+            @cached()
+            def func(self, key):
+                return key
 
-        func.invalidate("what")
+        A().func.invalidate("what")
 
     @defer.inlineCallbacks
     def test_max_entries(self):
         callcount = [0]
 
-        @cached(max_entries=10)
-        def func(self, key):
-            callcount[0] += 1
-            return key
+        class A(object):
+            @cached(max_entries=10)
+            def func(self, key):
+                callcount[0] += 1
+                return key
 
-        for k in range(0,12):
-            yield func(self, k)
+        a = A()
+
+        for k in range(0, 12):
+            yield a.func(k)
 
         self.assertEquals(callcount[0], 12)
 
         # There must have been at least 2 evictions, meaning if we calculate
         # all 12 values again, we must get called at least 2 more times
         for k in range(0,12):
-            yield func(self, k)
+            yield a.func(k)
 
         self.assertTrue(callcount[0] >= 14,
             msg="Expected callcount >= 14, got %d" % (callcount[0]))
@@ -171,12 +182,15 @@ class CacheDecoratorTestCase(unittest.TestCase):
     def test_prefill(self):
         callcount = [0]
 
-        @cached()
-        def func(self, key):
-            callcount[0] += 1
-            return key
+        class A(object):
+            @cached()
+            def func(self, key):
+                callcount[0] += 1
+                return key
 
-        func.prefill("foo", 123)
+        a = A()
 
-        self.assertEquals((yield func(self, "foo")), 123)
+        a.func.prefill("foo", 123)
+
+        self.assertEquals((yield a.func("foo")), 123)
         self.assertEquals(callcount[0], 0)

tests/storage/test_registration.py

@@ -46,7 +46,7 @@ class RegistrationStoreTestCase(unittest.TestCase):
             (yield self.store.get_user_by_id(self.user_id))
         )
 
-        result = yield self.store.get_user_by_token(self.tokens[1])
+        result = yield self.store.get_user_by_token(self.tokens[0])
 
         self.assertDictContainsSubset(
             {