Merge pull request #3092 from matrix-org/rav/response_cache_metrics

Add metrics for ResponseCache

synapse/appservice/api.py

@@ -73,7 +73,8 @@ class ApplicationServiceApi(SimpleHttpClient):
         super(ApplicationServiceApi, self).__init__(hs)
         self.clock = hs.get_clock()
 
-        self.protocol_meta_cache = ResponseCache(hs, timeout_ms=HOUR_IN_MS)
+        self.protocol_meta_cache = ResponseCache(hs, "as_protocol_meta",
+                                                 timeout_ms=HOUR_IN_MS)
 
     @defer.inlineCallbacks
     def query_user(self, service, user_id):

synapse/federation/federation_server.py

@@ -65,7 +65,7 @@ class FederationServer(FederationBase):
 
         # We cache responses to state queries, as they take a while and often
         # come in waves.
-        self._state_resp_cache = ResponseCache(hs, timeout_ms=30000)
+        self._state_resp_cache = ResponseCache(hs, "state_resp", timeout_ms=30000)
 
     @defer.inlineCallbacks
     @log_function

synapse/handlers/room_list.py

@@ -44,8 +44,9 @@ 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, "room_list")
-        self.remote_response_cache = ResponseCache(hs, timeout_ms=30 * 1000)
+        self.remote_response_cache = ResponseCache(hs, "remote_room_list",
+                                                   timeout_ms=30 * 1000)
 
     def get_local_public_room_list(self, limit=None, since_token=None,
                                    search_filter=None,

synapse/handlers/sync.py

@@ -169,7 +169,7 @@ 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, "sync")
         self.state = hs.get_state_handler()
 
     def wait_for_sync_for_user(self, sync_config, since_token=None, timeout=0,

synapse/replication/http/send_event.py

@@ -115,7 +115,7 @@ class ReplicationSendEventRestServlet(RestServlet):
         self.clock = hs.get_clock()
 
         # The responses are tiny, so we may as well cache them for a while
-        self.response_cache = ResponseCache(hs, timeout_ms=30 * 60 * 1000)
+        self.response_cache = ResponseCache(hs, "send_event", timeout_ms=30 * 60 * 1000)
 
     def on_PUT(self, request, event_id):
         result = self.response_cache.get(event_id)

synapse/util/caches/response_cache.py

@@ -14,6 +14,7 @@
 # limitations under the License.
 
 from synapse.util.async import ObservableDeferred
+from synapse.util.caches import metrics as cache_metrics
 
 
 class ResponseCache(object):
@@ -24,12 +25,21 @@ class ResponseCache(object):
     used rather than trying to compute a new response.
     """
 
-    def __init__(self, hs, timeout_ms=0):
+    def __init__(self, hs, name, timeout_ms=0):
         self.pending_result_cache = {}  # Requests that haven't finished yet.
 
         self.clock = hs.get_clock()
         self.timeout_sec = timeout_ms / 1000.
 
+        self._metrics = cache_metrics.register_cache(
+            "response_cache",
+            size_callback=lambda: self.size(),
+            cache_name=name,
+        )
+
+    def size(self):
+        return len(self.pending_result_cache)
+
     def get(self, key):
         """Look up the given key.
 
@@ -45,8 +55,10 @@ class ResponseCache(object):
         """
         result = self.pending_result_cache.get(key)
         if result is not None:
+            self._metrics.inc_hits()
             return result.observe()
         else:
+            self._metrics.inc_misses()
             return None
 
     def set(self, key, deferred):