From 191c7bef6bbb80f66f66e95387940c3bb6b5a0cf Mon Sep 17 00:00:00 2001
From: Mark Haines <mark.haines@matrix.org>
Date: Thu, 24 Mar 2016 17:47:31 +0000
Subject: [PATCH] Deduplicate identical /sync requests

---
 synapse/handlers/sync.py              | 16 +++++++++-
 synapse/rest/client/v2_alpha/sync.py  |  3 ++
 synapse/util/caches/response_cache.py | 46 +++++++++++++++++++++++++++
 3 files changed, 64 insertions(+), 1 deletion(-)
 create mode 100644 synapse/util/caches/response_cache.py

diff --git a/synapse/handlers/sync.py b/synapse/handlers/sync.py
index 1f6fde8e8a..48ab5707e1 100644
--- a/synapse/handlers/sync.py
+++ b/synapse/handlers/sync.py
@@ -20,6 +20,7 @@ from synapse.api.constants import Membership, EventTypes
 from synapse.util import unwrapFirstError
 from synapse.util.logcontext import LoggingContext, preserve_fn
 from synapse.util.metrics import Measure
+from synapse.util.caches.response_cache import ResponseCache
 from synapse.push.clientformat import format_push_rules_for_user
 
 from twisted.internet import defer
@@ -35,6 +36,7 @@ SyncConfig = collections.namedtuple("SyncConfig", [
     "user",
     "filter_collection",
     "is_guest",
+    "request_key",
 ])
 
 
@@ -136,8 +138,8 @@ class SyncHandler(BaseHandler):
         super(SyncHandler, self).__init__(hs)
         self.event_sources = hs.get_event_sources()
         self.clock = hs.get_clock()
+        self.response_cache = ResponseCache()
 
-    @defer.inlineCallbacks
     def wait_for_sync_for_user(self, sync_config, since_token=None, timeout=0,
                                full_state=False):
         """Get the sync for a client if we have new data for it now. Otherwise
@@ -146,7 +148,19 @@ class SyncHandler(BaseHandler):
         Returns:
             A Deferred SyncResult.
         """
+        result = self.response_cache.get(sync_config.request_key)
+        if not result:
+            result = self.response_cache.set(
+                sync_config.request_key,
+                self._wait_for_sync_for_user(
+                    sync_config, since_token, timeout, full_state
+                )
+            )
+        return result
 
+    @defer.inlineCallbacks
+    def _wait_for_sync_for_user(self, sync_config, since_token, timeout,
+                                full_state):
         context = LoggingContext.current_context()
         if context:
             if since_token is None:
diff --git a/synapse/rest/client/v2_alpha/sync.py b/synapse/rest/client/v2_alpha/sync.py
index de4a020ad4..c5785d7074 100644
--- a/synapse/rest/client/v2_alpha/sync.py
+++ b/synapse/rest/client/v2_alpha/sync.py
@@ -115,6 +115,8 @@ class SyncRestServlet(RestServlet):
             )
         )
 
+        request_key = (user, timeout, since, filter_id, full_state)
+
         if filter_id:
             if filter_id.startswith('{'):
                 try:
@@ -134,6 +136,7 @@ class SyncRestServlet(RestServlet):
             user=user,
             filter_collection=filter,
             is_guest=requester.is_guest,
+            request_key=request_key,
         )
 
         if since is not None:
diff --git a/synapse/util/caches/response_cache.py b/synapse/util/caches/response_cache.py
new file mode 100644
index 0000000000..1c2e344269
--- /dev/null
+++ b/synapse/util/caches/response_cache.py
@@ -0,0 +1,46 @@
+# -*- coding: utf-8 -*-
+# Copyright 2016 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.
+
+from synapse.util.async import ObservableDeferred
+
+
+class ResponseCache(object):
+    """
+    This caches a deferred response. Until the deferred completes it will be
+    returned from the cache. This means that if the client retries the request
+    while the response is still being computed, that original response will be
+    used rather than trying to compute a new response.
+    """
+
+    def __init__(self):
+        self.pending_result_cache = {}  # Request that haven't finished yet.
+
+    def get(self, key):
+        result = self.pending_result_cache.get(key)
+        if result is not None:
+            return result.observe()
+        else:
+            return None
+
+    def set(self, key, deferred):
+        result = ObservableDeferred(deferred)
+        self.pending_result_cache[key] = result
+
+        def remove(r):
+            self.pending_result_cache.pop(key, None)
+            return r
+
+        result.addBoth(remove)
+        return result.observe()