diff --git a/synapse/__init__.py b/synapse/__init__.py
index 012425fa60..8fe8df4edb 100644
--- a/synapse/__init__.py
+++ b/synapse/__init__.py
@@ -16,4 +16,4 @@
""" This is a reference implementation of a Matrix home server.
"""
-__version__ = "0.6.1d"
+__version__ = "0.6.1f"
diff --git a/synapse/config/_base.py b/synapse/config/_base.py
index 9b0f8c3c32..87cdbf1d30 100644
--- a/synapse/config/_base.py
+++ b/synapse/config/_base.py
@@ -28,6 +28,16 @@ class Config(object):
pass
@staticmethod
+ def parse_size(string):
+ sizes = {"K": 1024, "M": 1024 * 1024}
+ size = 1
+ suffix = string[-1]
+ if suffix in sizes:
+ string = string[:-1]
+ size = sizes[suffix]
+ return int(string) * size
+
+ @staticmethod
def abspath(file_path):
return os.path.abspath(file_path) if file_path else file_path
diff --git a/synapse/config/database.py b/synapse/config/database.py
index daa161c952..87efe54645 100644
--- a/synapse/config/database.py
+++ b/synapse/config/database.py
@@ -24,6 +24,7 @@ class DatabaseConfig(Config):
self.database_path = ":memory:"
else:
self.database_path = self.abspath(args.database_path)
+ self.event_cache_size = self.parse_size(args.event_cache_size)
@classmethod
def add_arguments(cls, parser):
@@ -33,6 +34,10 @@ class DatabaseConfig(Config):
"-d", "--database-path", default="homeserver.db",
help="The database name."
)
+ db_group.add_argument(
+ "--event-cache-size", default="100K",
+ help="Number of events to cache in memory."
+ )
@classmethod
def generate_config(cls, args, config_dir_path):
diff --git a/synapse/http/server.py b/synapse/http/server.py
index 6d084fa33c..589c30c199 100644
--- a/synapse/http/server.py
+++ b/synapse/http/server.py
@@ -113,6 +113,7 @@ class JsonResource(HttpServer, resource.Resource):
path.
"""
code = None
+ start = self.clock.time_msec()
try:
# Just say yes to OPTIONS.
if request.method == "OPTIONS":
@@ -137,8 +138,6 @@ class JsonResource(HttpServer, resource.Resource):
request.method, request.path
)
- start = self.clock.time_msec()
-
code, response = yield path_entry.callback(
request,
*args
diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py
index 2225be96be..4b618e0c65 100644
--- a/synapse/storage/__init__.py
+++ b/synapse/storage/__init__.py
@@ -167,6 +167,9 @@ class DataStore(RoomMemberStore, RoomStore,
stream_ordering=None, is_new_state=True,
current_state=None):
+ # Remove the any existing cache entries for the event_id
+ self._get_event_cache.pop(event.event_id)
+
# We purposefully do this first since if we include a `current_state`
# key, we *want* to update the `current_state_events` table
if current_state:
@@ -423,6 +426,8 @@ class DataStore(RoomMemberStore, RoomStore,
)
def _store_redaction(self, txn, event):
+ # invalidate the cache for the redacted event
+ self._get_event_cache.pop(event.redacts)
txn.execute(
"INSERT OR IGNORE INTO redactions "
"(event_id, redacts) VALUES (?,?)",
diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py
index 3e1ab0a159..29fc334f45 100644
--- a/synapse/storage/_base.py
+++ b/synapse/storage/_base.py
@@ -19,6 +19,7 @@ from synapse.events import FrozenEvent
from synapse.events.utils import prune_event
from synapse.util.logutils import log_function
from synapse.util.logcontext import PreserveLoggingContext, LoggingContext
+from synapse.util.lrucache import LruCache
from twisted.internet import defer
@@ -128,6 +129,8 @@ class SQLBaseStore(object):
self._txn_perf_counters = PerformanceCounters()
self._get_event_counters = PerformanceCounters()
+ self._get_event_cache = LruCache(hs.config.event_cache_size)
+
def start_profiling(self):
self._previous_loop_ts = self._clock.time_msec()
@@ -579,6 +582,19 @@ class SQLBaseStore(object):
def _get_event_txn(self, txn, event_id, check_redacted=True,
get_prev_content=False, allow_rejected=False):
+
+ start_time = time.time() * 1000
+ update_counter = self._get_event_counters.update
+
+ try:
+ cache = self._get_event_cache.setdefault(event_id, {})
+ # Separate cache entries for each way to invoke _get_event_txn
+ return cache[(check_redacted, get_prev_content, allow_rejected)]
+ except KeyError:
+ pass
+ finally:
+ start_time = update_counter("event_cache", start_time)
+
sql = (
"SELECT e.internal_metadata, e.json, r.event_id, rej.reason "
"FROM event_json as e "
@@ -588,8 +604,6 @@ class SQLBaseStore(object):
"LIMIT 1 "
)
- start_time = time.time() * 1000
-
txn.execute(sql, (event_id,))
res = txn.fetchone()
@@ -599,14 +613,16 @@ class SQLBaseStore(object):
internal_metadata, js, redacted, rejected_reason = res
- self._get_event_counters.update("select_event", start_time)
+ start_time = update_counter("select_event", start_time)
if allow_rejected or not rejected_reason:
- return self._get_event_from_row_txn(
+ result = self._get_event_from_row_txn(
txn, internal_metadata, js, redacted,
check_redacted=check_redacted,
get_prev_content=get_prev_content,
)
+ cache[(check_redacted, get_prev_content, allow_rejected)] = result
+ return result
else:
return None
diff --git a/synapse/storage/roommember.py b/synapse/storage/roommember.py
index c69dd995ce..779f9ce544 100644
--- a/synapse/storage/roommember.py
+++ b/synapse/storage/roommember.py
@@ -35,6 +35,11 @@ RoomsForUser = namedtuple(
class RoomMemberStore(SQLBaseStore):
+ def __init__(self, *args, **kw):
+ super(RoomMemberStore, self).__init__(*args, **kw)
+
+ self._user_rooms_cache = {}
+
def _store_room_member_txn(self, txn, event):
"""Store a room member in the database.
"""
@@ -98,6 +103,8 @@ class RoomMemberStore(SQLBaseStore):
txn.execute(sql, (event.room_id, domain))
+ self.invalidate_rooms_for_user(target_user_id)
+
@defer.inlineCallbacks
def get_room_member(self, user_id, room_id):
"""Retrieve the current state of a room member.
@@ -240,28 +247,53 @@ class RoomMemberStore(SQLBaseStore):
results = self._parse_events_txn(txn, rows)
return results
+ # TODO(paul): Create a nice @cached decorator to do this
+ # @cached
+ # def get_foo(...)
+ # ...
+ # invalidate_foo = get_foo.invalidator
+
+ @defer.inlineCallbacks
+ def get_rooms_for_user(self, user_id):
+ # TODO(paul): put some performance counters in here so we can easily
+ # track what impact this cache is having
+ if user_id in self._user_rooms_cache:
+ defer.returnValue(self._user_rooms_cache[user_id])
+
+ rooms = yield self.get_rooms_for_user_where_membership_is(
+ user_id, membership_list=[Membership.JOIN],
+ )
+
+ # TODO(paul): Consider applying a maximum size; just evict things at
+ # random, or consider LRU?
+
+ self._user_rooms_cache[user_id] = rooms
+ defer.returnValue(rooms)
+
+ def invalidate_rooms_for_user(self, user_id):
+ if user_id in self._user_rooms_cache:
+ del self._user_rooms_cache[user_id]
+
+ @defer.inlineCallbacks
def user_rooms_intersect(self, user_id_list):
""" Checks whether all the users whose IDs are given in a list share a
room.
+
+ This is a "hot path" function that's called a lot, e.g. by presence for
+ generating the event stream. As such, it is implemented locally by
+ wrapping logic around heavily-cached database queries.
"""
- def interaction(txn):
- user_list_clause = " OR ".join(["m.user_id = ?"] * len(user_id_list))
- sql = (
- "SELECT m.room_id FROM room_memberships as m "
- "INNER JOIN current_state_events as c "
- "ON m.event_id = c.event_id "
- "WHERE m.membership = 'join' "
- "AND (%(clause)s) "
- # TODO(paul): We've got duplicate rows in the database somewhere
- # so we have to DISTINCT m.user_id here
- "GROUP BY m.room_id HAVING COUNT(DISTINCT m.user_id) = ?"
- ) % {"clause": user_list_clause}
-
- args = list(user_id_list)
- args.append(len(user_id_list))
+ if len(user_id_list) < 2:
+ defer.returnValue(True)
- txn.execute(sql, args)
+ deferreds = [self.get_rooms_for_user(u) for u in user_id_list]
+
+ results = yield defer.DeferredList(deferreds)
+
+ # A list of sets of strings giving room IDs for each user
+ room_id_lists = [set([r.room_id for r in result[1]]) for result in results]
- return len(txn.fetchall()) > 0
+ # There isn't a setintersection(*list_of_sets)
+ ret = len(room_id_lists.pop(0).intersection(*room_id_lists)) > 0
- return self.runInteraction("user_rooms_intersect", interaction)
+ defer.returnValue(ret)
diff --git a/synapse/util/lrucache.py b/synapse/util/lrucache.py
new file mode 100644
index 0000000000..a45c673d32
--- /dev/null
+++ b/synapse/util/lrucache.py
@@ -0,0 +1,110 @@
+# -*- coding: utf-8 -*-
+# 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.
+
+
+class LruCache(object):
+ """Least-recently-used cache."""
+ # TODO(mjark) Add hit/miss counters
+ # TODO(mjark) Add mutex for linked list for thread safety.
+ def __init__(self, max_size):
+ cache = {}
+ list_root = []
+ list_root[:] = [list_root, list_root, None, None]
+
+ PREV, NEXT, KEY, VALUE = 0, 1, 2, 3
+
+ def add_node(key, value):
+ prev_node = list_root
+ next_node = prev_node[NEXT]
+ node = [prev_node, next_node, key, value]
+ prev_node[NEXT] = node
+ next_node[PREV] = node
+ cache[key] = node
+
+ def move_node_to_front(node):
+ prev_node = node[PREV]
+ next_node = node[NEXT]
+ prev_node[NEXT] = next_node
+ next_node[PREV] = prev_node
+ prev_node = list_root
+ next_node = prev_node[NEXT]
+ node[PREV] = prev_node
+ node[NEXT] = next_node
+ prev_node[NEXT] = node
+ next_node[PREV] = node
+
+ def delete_node(node):
+ prev_node = node[PREV]
+ next_node = node[NEXT]
+ prev_node[NEXT] = next_node
+ next_node[PREV] = prev_node
+ cache.pop(node[KEY], None)
+
+ def cache_get(key, default=None):
+ node = cache.get(key, None)
+ if node is not None:
+ move_node_to_front(node)
+ return node[VALUE]
+ else:
+ return default
+
+ def cache_set(key, value):
+ node = cache.get(key, None)
+ if node is not None:
+ move_node_to_front(node)
+ node[VALUE] = value
+ else:
+ add_node(key, value)
+ if len(cache) > max_size:
+ delete_node(list_root[PREV])
+
+ def cache_set_default(key, value):
+ node = cache.get(key, None)
+ if node is not None:
+ return node[VALUE]
+ else:
+ add_node(key, value)
+ if len(cache) > max_size:
+ delete_node(list_root[PREV])
+ return value
+
+ def cache_pop(key, default=None):
+ node = cache.get(key, None)
+ if node:
+ delete_node(node)
+ return node[VALUE]
+ else:
+ return default
+
+ self.sentinel = object()
+ self.get = cache_get
+ self.set = cache_set
+ self.setdefault = cache_set_default
+ self.pop = cache_pop
+
+ def __getitem__(self, key):
+ result = self.get(key, self.sentinel)
+ if result is self.sentinel:
+ raise KeyError()
+ else:
+ return result
+
+ def __setitem__(self, key, value):
+ self.set(key, value)
+
+ def __delitem__(self, key, value):
+ result = self.pop(key, self.sentinel)
+ if result is self.sentinel:
+ raise KeyError()
|
