From 36c28bc467e53cf7318964a4dc986428644f27ed Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Mon, 27 Mar 2017 16:33:44 +0100
Subject: Update all the workers and master to use TCP replication

---
 synapse/app/synchrotron.py | 306 ++++++++++++++++-----------------------------
 1 file changed, 110 insertions(+), 196 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index 34e34e5580..c9c48f324f 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -16,7 +16,7 @@
 
 import synapse
 
-from synapse.api.constants import EventTypes, PresenceState
+from synapse.api.constants import EventTypes
 from synapse.config._base import ConfigError
 from synapse.config.homeserver import HomeServerConfig
 from synapse.config.logger import setup_logging
@@ -40,15 +40,14 @@ from synapse.replication.slave.storage.presence import SlavedPresenceStore
 from synapse.replication.slave.storage.deviceinbox import SlavedDeviceInboxStore
 from synapse.replication.slave.storage.devices import SlavedDeviceStore
 from synapse.replication.slave.storage.room import RoomStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
 from synapse.storage.presence import PresenceStore, UserPresenceState
 from synapse.storage.roommember import RoomMemberStore
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, preserve_fn, \
-    PreserveLoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.stringutils import random_string
@@ -111,7 +110,6 @@ class SynchrotronPresence(object):
         self.http_client = hs.get_simple_http_client()
         self.store = hs.get_datastore()
         self.user_to_num_current_syncs = {}
-        self.syncing_users_url = hs.config.worker_replication_url + "/syncing_users"
         self.clock = hs.get_clock()
         self.notifier = hs.get_notifier()
 
@@ -124,14 +122,7 @@ class SynchrotronPresence(object):
         self.process_id = random_string(16)
         logger.info("Presence process_id is %r", self.process_id)
 
-        self._sending_sync = False
-        self._need_to_send_sync = False
-        self.clock.looping_call(
-            self._send_syncing_users_regularly,
-            UPDATE_SYNCING_USERS_MS,
-        )
-
-        reactor.addSystemEventTrigger("before", "shutdown", self._on_shutdown)
+        self.sync_callback = None
 
     def set_state(self, user, state, ignore_status_msg=False):
         # TODO Hows this supposed to work?
@@ -142,15 +133,15 @@ class SynchrotronPresence(object):
     _get_interested_parties = PresenceHandler._get_interested_parties.__func__
     current_state_for_users = PresenceHandler.current_state_for_users.__func__
 
-    @defer.inlineCallbacks
     def user_syncing(self, user_id, affect_presence):
         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
-            prev_states = yield self.current_state_for_users([user_id])
-            if prev_states[user_id].state == PresenceState.OFFLINE:
-                # TODO: Don't block the sync request on this HTTP hit.
-                yield self._send_syncing_users_now()
+
+            if self.sync_callback:
+                if self.user_to_num_current_syncs[user_id] == 1:
+                    now = self.clock.time_msec()
+                    self.sync_callback(user_id, True, now)
 
         def _end():
             # We check that the user_id is in user_to_num_current_syncs because
@@ -159,6 +150,11 @@ class SynchrotronPresence(object):
             if affect_presence and user_id in self.user_to_num_current_syncs:
                 self.user_to_num_current_syncs[user_id] -= 1
 
+                if self.sync_callback:
+                    if self.user_to_num_current_syncs[user_id] == 0:
+                        now = self.clock.time_msec()
+                        self.sync_callback(user_id, False, now)
+
         @contextlib.contextmanager
         def _user_syncing():
             try:
@@ -166,49 +162,7 @@ class SynchrotronPresence(object):
             finally:
                 _end()
 
-        defer.returnValue(_user_syncing())
-
-    @defer.inlineCallbacks
-    def _on_shutdown(self):
-        # When the synchrotron is shutdown tell the master to clear the in
-        # progress syncs for this process
-        self.user_to_num_current_syncs.clear()
-        yield self._send_syncing_users_now()
-
-    def _send_syncing_users_regularly(self):
-        # Only send an update if we aren't in the middle of sending one.
-        if not self._sending_sync:
-            preserve_fn(self._send_syncing_users_now)()
-
-    @defer.inlineCallbacks
-    def _send_syncing_users_now(self):
-        if self._sending_sync:
-            # We don't want to race with sending another update.
-            # Instead we wait for that update to finish and send another
-            # update afterwards.
-            self._need_to_send_sync = True
-            return
-
-        # Flag that we are sending an update.
-        self._sending_sync = True
-
-        yield self.http_client.post_json_get_json(self.syncing_users_url, {
-            "process_id": self.process_id,
-            "syncing_users": [
-                user_id for user_id, count in self.user_to_num_current_syncs.items()
-                if count > 0
-            ],
-        })
-
-        # Unset the flag as we are no longer sending an update.
-        self._sending_sync = False
-        if self._need_to_send_sync:
-            # If something happened while we were sending the update then
-            # we might need to send another update.
-            # TODO: Check if the update that was sent matches the current state
-            # as we only need to send an update if they are different.
-            self._need_to_send_sync = False
-            yield self._send_syncing_users_now()
+        return defer.succeed(_user_syncing())
 
     @defer.inlineCallbacks
     def notify_from_replication(self, states, stream_id):
@@ -223,26 +177,24 @@ class SynchrotronPresence(object):
         )
 
     @defer.inlineCallbacks
-    def process_replication(self, result):
-        stream = result.get("presence", {"rows": []})
-        states = []
-        for row in stream["rows"]:
-            (
-                position, user_id, state, last_active_ts,
-                last_federation_update_ts, last_user_sync_ts, status_msg,
-                currently_active
-            ) = row
-            state = UserPresenceState(
-                user_id, state, last_active_ts,
-                last_federation_update_ts, last_user_sync_ts, status_msg,
-                currently_active
-            )
-            self.user_to_current_state[user_id] = state
-            states.append(state)
+    def process_replication_rows(self, token, rows):
+        states = [UserPresenceState(
+            row.user_id, row.state, row.last_active_ts,
+            row.last_federation_update_ts, row.last_user_sync_ts, row.status_msg,
+            row.currently_active
+        ) for row in rows]
 
-        if states and "position" in stream:
-            stream_id = int(stream["position"])
-            yield self.notify_from_replication(states, stream_id)
+        for state in states:
+            self.user_to_current_state[row.user_id] = state
+
+        stream_id = token
+        yield self.notify_from_replication(states, stream_id)
+
+    def get_currently_syncing_users(self):
+        return [
+            user_id for user_id, count in self.user_to_num_current_syncs.iteritems()
+            if count > 0
+        ]
 
 
 class SynchrotronTyping(object):
@@ -257,16 +209,13 @@ class SynchrotronTyping(object):
         # value which we *must* use for the next replication request.
         return {"typing": self._latest_room_serial}
 
-    def process_replication(self, result):
-        stream = result.get("typing")
-        if stream:
-            self._latest_room_serial = int(stream["position"])
+    def process_replication_rows(self, token, rows):
+        self._latest_room_serial = token
 
-            for row in stream["rows"]:
-                position, room_id, typing_json = row
-                typing = json.loads(typing_json)
-                self._room_serials[room_id] = position
-                self._room_typing[room_id] = typing
+        for row in rows:
+            typing = json.loads(row.user_ids)
+            self._room_serials[row.room_id] = token
+            self._room_typing[row.room_id] = typing
 
 
 class SynchrotronApplicationService(object):
@@ -351,124 +300,90 @@ class SynchrotronServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        notifier = self.get_notifier()
-        presence_handler = self.get_presence_handler()
-        typing_handler = self.get_typing_handler()
-
-        def notify_from_stream(
-            result, stream_name, stream_key, room=None, user=None
-        ):
-            stream = result.get(stream_name)
-            if stream:
-                position_index = stream["field_names"].index("position")
-                if room:
-                    room_index = stream["field_names"].index(room)
-                if user:
-                    user_index = stream["field_names"].index(user)
-
-                users = ()
-                rooms = ()
-                for row in stream["rows"]:
-                    position = row[position_index]
-
-                    if user:
-                        users = (row[user_index],)
-
-                    if room:
-                        rooms = (row[room_index],)
-
-                    notifier.on_new_event(
-                        stream_key, position, users=users, rooms=rooms
-                    )
+        self.get_tcp_replication().start_replication(self)
 
-        @defer.inlineCallbacks
-        def notify_device_list_update(result):
-            stream = result.get("device_lists")
-            if not stream:
-                return
+    def build_tcp_replication(self):
+        return SyncReplicationHandler(self)
 
-            position_index = stream["field_names"].index("position")
-            user_index = stream["field_names"].index("user_id")
+    def build_presence_handler(self):
+        return SynchrotronPresence(self)
 
-            for row in stream["rows"]:
-                position = row[position_index]
-                user_id = row[user_index]
+    def build_typing_handler(self):
+        return SynchrotronTyping(self)
 
-                room_ids = yield store.get_rooms_for_user(user_id)
 
-                notifier.on_new_event(
-                    "device_list_key", position, rooms=room_ids,
-                )
+class SyncReplicationHandler(ReplicationClientHandler):
+    def __init__(self, hs):
+        super(SyncReplicationHandler, self).__init__(hs.get_datastore())
 
-        @defer.inlineCallbacks
-        def notify(result):
-            stream = result.get("events")
-            if stream:
-                max_position = stream["position"]
-
-                event_map = yield store.get_events([row[1] for row in stream["rows"]])
-
-                for row in stream["rows"]:
-                    position = row[0]
-                    event_id = row[1]
-                    event = event_map.get(event_id, None)
-                    if not event:
-                        continue
-
-                    extra_users = ()
-                    if event.type == EventTypes.Member:
-                        extra_users = (event.state_key,)
-                    notifier.on_new_room_event(
-                        event, position, max_position, extra_users
-                    )
+        self.store = hs.get_datastore()
+        self.typing_handler = hs.get_typing_handler()
+        self.presence_handler = hs.get_presence_handler()
+        self.notifier = hs.get_notifier()
 
-            notify_from_stream(
-                result, "push_rules", "push_rules_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "user_account_data", "account_data_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "room_account_data", "account_data_key", user="user_id"
+        self.presence_handler.sync_callback = self.send_user_sync
+
+    def on_rdata(self, stream_name, token, rows):
+        super(SyncReplicationHandler, self).on_rdata(stream_name, token, rows)
+
+        if stream_name == "typing":
+            self.typing_handler.process_replication_rows(token, rows)
+        elif stream_name == "presence":
+            self.presence_handler.process_replication_rows(token, rows)
+        self.notify(stream_name, token, rows)
+
+    def get_streams_to_replicate(self):
+        args = super(SyncReplicationHandler, self).get_streams_to_replicate()
+        args.update(self.typing_handler.stream_positions())
+        return args
+
+    def get_currently_syncing_users(self):
+        return self.presence_handler.get_currently_syncing_users()
+
+    @defer.inlineCallbacks
+    def notify(self, stream_name, token, rows):
+        if stream_name == "events":
+            # We shouldn't get multiple rows per token for events stream, so
+            # we don't need to optimise this for multiple rows.
+            for row in rows:
+                event = yield self.store.get_event(row.event_id)
+                extra_users = ()
+                if event.type == EventTypes.Member:
+                    extra_users = (event.state_key,)
+                max_token = self.store.get_room_max_stream_ordering()
+                self.notifier.on_new_room_event(
+                    event, token, max_token, extra_users
+                )
+        elif stream_name == "push_rules":
+            self.notifier.on_new_event(
+                "push_rules_key", token, users=[row.user_id for row in rows],
             )
-            notify_from_stream(
-                result, "tag_account_data", "account_data_key", user="user_id"
+        elif stream_name in ("account_data", "tag_account_data",):
+            self.notifier.on_new_event(
+                "account_data_key", token, users=[row.user_id for row in rows],
             )
-            notify_from_stream(
-                result, "receipts", "receipt_key", room="room_id"
+        elif stream_name == "receipts":
+            self.notifier.on_new_event(
+                "receipt_key", token, rooms=[row.room_id for row in rows],
             )
-            notify_from_stream(
-                result, "typing", "typing_key", room="room_id"
+        elif stream_name == "typing":
+            self.notifier.on_new_event(
+                "typing_key", token, rooms=[row.room_id for row in rows],
             )
-            notify_from_stream(
-                result, "to_device", "to_device_key", user="user_id"
+        elif stream_name == "to_device":
+            entities = [row.entity for row in rows if row.entity.startswith("@")]
+            if entities:
+                self.notifier.on_new_event(
+                    "to_device_key", token, users=entities,
+                )
+        elif stream_name == "device_lists":
+            all_room_ids = set()
+            for row in rows:
+                room_ids = yield self.store.get_rooms_for_user(row.user_id)
+                all_room_ids.update(room_ids)
+            self.notifier.on_new_event(
+                "device_list_key", token, rooms=all_room_ids,
             )
-            yield notify_device_list_update(result)
-
-        while True:
-            try:
-                args = store.stream_positions()
-                args.update(typing_handler.stream_positions())
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-                typing_handler.process_replication(result)
-                yield presence_handler.process_replication(result)
-                yield notify(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
-
-    def build_presence_handler(self):
-        return SynchrotronPresence(self)
-
-    def build_typing_handler(self):
-        return SynchrotronTyping(self)
 
 
 def start(config_options):
@@ -514,7 +429,6 @@ def start(config_options):
 
     def start():
         ss.get_datastore().start_profiling()
-        ss.replicate()
         ss.get_state_handler().start_caching()
 
     reactor.callWhenRunning(start)
-- 
cgit 1.5.1


From 3376f160124ff2f8915341c90a24ba83495fb8d9 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Tue, 4 Apr 2017 11:09:26 +0100
Subject: Shuffle and comment synchrotron presence

---
 synapse/app/synchrotron.py | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index c9c48f324f..7341970a81 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -106,6 +106,7 @@ UPDATE_SYNCING_USERS_MS = 10 * 1000
 
 class SynchrotronPresence(object):
     def __init__(self, hs):
+        self.hs = hs
         self.is_mine_id = hs.is_mine_id
         self.http_client = hs.get_simple_http_client()
         self.store = hs.get_datastore()
@@ -122,7 +123,8 @@ class SynchrotronPresence(object):
         self.process_id = random_string(16)
         logger.info("Presence process_id is %r", self.process_id)
 
-        self.sync_callback = None
+    def send_user_sync(self, user_id, is_syncing, last_sync_ms):
+        self.hs.get_tcp_replication().send_user_sync(user_id, is_syncing, last_sync_ms)
 
     def set_state(self, user, state, ignore_status_msg=False):
         # TODO Hows this supposed to work?
@@ -138,10 +140,10 @@ class SynchrotronPresence(object):
             curr_sync = self.user_to_num_current_syncs.get(user_id, 0)
             self.user_to_num_current_syncs[user_id] = curr_sync + 1
 
-            if self.sync_callback:
-                if self.user_to_num_current_syncs[user_id] == 1:
-                    now = self.clock.time_msec()
-                    self.sync_callback(user_id, True, now)
+            # If we went from no in flight sync to some, notify replication
+            if self.user_to_num_current_syncs[user_id] == 1:
+                now = self.clock.time_msec()
+                self.send_user_sync(user_id, True, now)
 
         def _end():
             # We check that the user_id is in user_to_num_current_syncs because
@@ -150,10 +152,10 @@ class SynchrotronPresence(object):
             if affect_presence and user_id in self.user_to_num_current_syncs:
                 self.user_to_num_current_syncs[user_id] -= 1
 
-                if self.sync_callback:
-                    if self.user_to_num_current_syncs[user_id] == 0:
-                        now = self.clock.time_msec()
-                        self.sync_callback(user_id, False, now)
+                # If we went from one in flight sync to non, notify replication
+                if self.user_to_num_current_syncs[user_id] == 0:
+                    now = self.clock.time_msec()
+                    self.send_user_sync(user_id, False, now)
 
         @contextlib.contextmanager
         def _user_syncing():
-- 
cgit 1.5.1


From ac66e11f2b1235f801195ad0065008cdca2f1b0b Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Tue, 4 Apr 2017 15:22:54 +0100
Subject: Add the appropriate amount of preserve_fn

---
 synapse/app/appservice.py        |  6 ++++--
 synapse/app/federation_sender.py |  4 ++--
 synapse/app/pusher.py            | 13 +++++++------
 synapse/app/synchrotron.py       | 13 ++++++-------
 4 files changed, 19 insertions(+), 17 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/appservice.py b/synapse/app/appservice.py
index 13b69d8dee..9a476efa63 100644
--- a/synapse/app/appservice.py
+++ b/synapse/app/appservice.py
@@ -29,7 +29,7 @@ from synapse.replication.slave.storage.registration import SlavedRegistrationSto
 from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -136,7 +136,9 @@ class ASReplicationHandler(ReplicationClientHandler):
 
         if stream_name == "events":
             max_stream_id = self.store.get_room_max_stream_ordering()
-            self.appservice_handler.notify_interested_services(max_stream_id)
+            preserve_fn(
+                self.appservice_handler.notify_interested_services
+            )(max_stream_id)
 
 
 def start(config_options):
diff --git a/synapse/app/federation_sender.py b/synapse/app/federation_sender.py
index 705fe5e3df..fa7b19e53a 100644
--- a/synapse/app/federation_sender.py
+++ b/synapse/app/federation_sender.py
@@ -36,7 +36,7 @@ from synapse.storage.engines import create_engine
 from synapse.storage.presence import UserPresenceState
 from synapse.util.async import Linearizer
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.versionstring import get_version_string
@@ -337,7 +337,7 @@ class FederationSenderHandler(object):
             for destination in device_destinations:
                 self.federation_sender.send_device_messages(destination)
 
-            self.update_token(token)
+            preserve_fn(self.update_token)(token)
 
         # We also need to poke the federation sender when new events happen
         elif stream_name == "events":
diff --git a/synapse/app/pusher.py b/synapse/app/pusher.py
index cb76f058b0..f9114acfcb 100644
--- a/synapse/app/pusher.py
+++ b/synapse/app/pusher.py
@@ -39,7 +39,7 @@ from synapse.util.versionstring import get_version_string
 
 from synapse import events
 
-from twisted.internet import reactor
+from twisted.internet import reactor, defer
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -170,21 +170,22 @@ class PusherReplicationHandler(ReplicationClientHandler):
 
     def on_rdata(self, stream_name, token, rows):
         super(PusherReplicationHandler, self).on_rdata(stream_name, token, rows)
-        self.poke_pushers(stream_name, token, rows)
+        preserve_fn(self.poke_pushers)(stream_name, token, rows)
 
+    @defer.inlineCallbacks
     def poke_pushers(self, stream_name, token, rows):
         if stream_name == "pushers":
             for row in rows:
                 if row.deleted:
-                    self.stop_pusher(row.user_id, row.app_id, row.pushkey)
+                    yield self.stop_pusher(row.user_id, row.app_id, row.pushkey)
                 else:
-                    self.start_pusher(row.user_id, row.app_id, row.pushkey)
+                    yield self.start_pusher(row.user_id, row.app_id, row.pushkey)
         elif stream_name == "events":
-            preserve_fn(self.pusher_pool.on_new_notifications)(
+            yield self.pusher_pool.on_new_notifications(
                 token, token,
             )
         elif stream_name == "receipts":
-            preserve_fn(self.pusher_pool.on_new_receipts)(
+            yield self.pusher_pool.on_new_receipts(
                 token, token, set(row.room_id for row in rows)
             )
 
diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index 7341970a81..a1ef5dfa77 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -47,7 +47,7 @@ from synapse.storage.engines import create_engine
 from synapse.storage.presence import PresenceStore, UserPresenceState
 from synapse.storage.roommember import RoomMemberStore
 from synapse.util.httpresourcetree import create_resource_tree
-from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
 from synapse.util.manhole import manhole
 from synapse.util.rlimit import change_resource_limit
 from synapse.util.stringutils import random_string
@@ -328,11 +328,7 @@ class SyncReplicationHandler(ReplicationClientHandler):
     def on_rdata(self, stream_name, token, rows):
         super(SyncReplicationHandler, self).on_rdata(stream_name, token, rows)
 
-        if stream_name == "typing":
-            self.typing_handler.process_replication_rows(token, rows)
-        elif stream_name == "presence":
-            self.presence_handler.process_replication_rows(token, rows)
-        self.notify(stream_name, token, rows)
+        preserve_fn(self.process_and_notify)(stream_name, token, rows)
 
     def get_streams_to_replicate(self):
         args = super(SyncReplicationHandler, self).get_streams_to_replicate()
@@ -343,7 +339,7 @@ class SyncReplicationHandler(ReplicationClientHandler):
         return self.presence_handler.get_currently_syncing_users()
 
     @defer.inlineCallbacks
-    def notify(self, stream_name, token, rows):
+    def process_and_notify(self, stream_name, token, rows):
         if stream_name == "events":
             # We shouldn't get multiple rows per token for events stream, so
             # we don't need to optimise this for multiple rows.
@@ -369,6 +365,7 @@ class SyncReplicationHandler(ReplicationClientHandler):
                 "receipt_key", token, rooms=[row.room_id for row in rows],
             )
         elif stream_name == "typing":
+            self.typing_handler.process_replication_rows(token, rows)
             self.notifier.on_new_event(
                 "typing_key", token, rooms=[row.room_id for row in rows],
             )
@@ -386,6 +383,8 @@ class SyncReplicationHandler(ReplicationClientHandler):
             self.notifier.on_new_event(
                 "device_list_key", token, rooms=all_room_ids,
             )
+        elif stream_name == "presence":
+            yield self.presence_handler.process_replication_rows(token, rows)
 
 
 def start(config_options):
-- 
cgit 1.5.1


From 96b9b6c1275087acf57faf1306bcc392dbd9f842 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Wed, 5 Apr 2017 11:34:20 +0100
Subject: Don't double json encode typing replication data

---
 synapse/app/synchrotron.py | 4 +---
 synapse/handlers/typing.py | 4 +---
 2 files changed, 2 insertions(+), 6 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index a1ef5dfa77..1fac021ea9 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -62,7 +62,6 @@ import sys
 import logging
 import contextlib
 import gc
-import ujson as json
 
 logger = logging.getLogger("synapse.app.synchrotron")
 
@@ -215,9 +214,8 @@ class SynchrotronTyping(object):
         self._latest_room_serial = token
 
         for row in rows:
-            typing = json.loads(row.user_ids)
             self._room_serials[row.room_id] = token
-            self._room_typing[row.room_id] = typing
+            self._room_typing[row.room_id] = row.user_ids
 
 
 class SynchrotronApplicationService(object):
diff --git a/synapse/handlers/typing.py b/synapse/handlers/typing.py
index d6809862e0..3b7818af5c 100644
--- a/synapse/handlers/typing.py
+++ b/synapse/handlers/typing.py
@@ -24,7 +24,6 @@ from synapse.types import UserID, get_domain_from_id
 import logging
 
 from collections import namedtuple
-import ujson as json
 
 logger = logging.getLogger(__name__)
 
@@ -288,8 +287,7 @@ class TypingHandler(object):
         for room_id, serial in self._room_serials.items():
             if last_id < serial and serial <= current_id:
                 typing = self._room_typing[room_id]
-                typing_bytes = json.dumps(list(typing), ensure_ascii=False)
-                rows.append((serial, room_id, typing_bytes))
+                rows.append((serial, room_id, list(typing)))
         rows.sort()
         return rows
 
-- 
cgit 1.5.1


From b5cb6347a4d9b9fb5c21a2a6e3a2852c14b837fb Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Wed, 5 Apr 2017 13:25:40 +0100
Subject: Don't immediately notify the master about users whose syncs have gone
 away

---
 synapse/app/synchrotron.py | 40 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 36 insertions(+), 4 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index a1ef5dfa77..a5dfbef8da 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -120,12 +120,46 @@ class SynchrotronPresence(object):
             for state in active_presence
         }
 
+        # user_id -> last_sync_ms. Lists the users that have stopped syncing
+        # but we haven't notified the master of that yet
+        self.users_going_offline = {}
+
+        self._send_stop_syncing_loop = self.clock.looping_call(
+            self.send_stop_syncing, 10 * 1000
+        )
+
         self.process_id = random_string(16)
         logger.info("Presence process_id is %r", self.process_id)
 
     def send_user_sync(self, user_id, is_syncing, last_sync_ms):
         self.hs.get_tcp_replication().send_user_sync(user_id, is_syncing, last_sync_ms)
 
+    def mark_as_coming_online(self, user_id):
+        """A user has started syncing. Send a UserSync to the master, unless they
+        had recently stopped syncing.
+        """
+        going_offline = self.users_going_offline.pop(user_id, None)
+        if not going_offline:
+            self.send_user_sync(user_id, True, self.clock.time_msec())
+
+    def mark_as_going_offline(self, user_id):
+        """A user has stopped syncing. We wait before notifying the master as
+        its likely they'll come back soon. This allows us to avoid sending
+        a stopped syncing immediately followed by a started syncing notification
+        to the master
+        """
+        self.users_going_offline[user_id] = self.clock.time_msec()
+
+    def send_stop_syncing(self):
+        """Check if there are any users who have stopped syncing a while ago
+        and haven't come back yet. If there are poke the master about them.
+        """
+        now = self.clock.time_msec()
+        for user_id, last_sync_ms in self.users_going_offline.items():
+            if now - last_sync_ms > 10 * 1000:
+                self.users_going_offline.pop(user_id, None)
+                self.send_user_sync(user_id, False, last_sync_ms)
+
     def set_state(self, user, state, ignore_status_msg=False):
         # TODO Hows this supposed to work?
         pass
@@ -142,8 +176,7 @@ class SynchrotronPresence(object):
 
             # If we went from no in flight sync to some, notify replication
             if self.user_to_num_current_syncs[user_id] == 1:
-                now = self.clock.time_msec()
-                self.send_user_sync(user_id, True, now)
+                self.mark_as_coming_online(user_id)
 
         def _end():
             # We check that the user_id is in user_to_num_current_syncs because
@@ -154,8 +187,7 @@ class SynchrotronPresence(object):
 
                 # If we went from one in flight sync to non, notify replication
                 if self.user_to_num_current_syncs[user_id] == 0:
-                    now = self.clock.time_msec()
-                    self.send_user_sync(user_id, False, now)
+                    self.mark_as_going_offline(user_id)
 
         @contextlib.contextmanager
         def _user_syncing():
-- 
cgit 1.5.1


From dbf87282d39b9ee503f07d68c99ccc1624fc6e17 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Thu, 6 Apr 2017 13:11:21 +0100
Subject: Docs

---
 synapse/app/synchrotron.py | 7 +++++++
 1 file changed, 7 insertions(+)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index a5dfbef8da..e0716a6226 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -137,9 +137,13 @@ class SynchrotronPresence(object):
     def mark_as_coming_online(self, user_id):
         """A user has started syncing. Send a UserSync to the master, unless they
         had recently stopped syncing.
+
+        Args:
+            user_id (str)
         """
         going_offline = self.users_going_offline.pop(user_id, None)
         if not going_offline:
+            # Safe to skip if we haven't yet told the master they were offline
             self.send_user_sync(user_id, True, self.clock.time_msec())
 
     def mark_as_going_offline(self, user_id):
@@ -147,6 +151,9 @@ class SynchrotronPresence(object):
         its likely they'll come back soon. This allows us to avoid sending
         a stopped syncing immediately followed by a started syncing notification
         to the master
+
+        Args:
+            user_id (str)
         """
         self.users_going_offline[user_id] = self.clock.time_msec()
 
-- 
cgit 1.5.1


From 391712a4f9c43359c9f6c36a6ed51800b370ffc0 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Thu, 6 Apr 2017 13:35:00 +0100
Subject: Comment

---
 synapse/app/synchrotron.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index e0716a6226..67d9210f2a 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -143,7 +143,7 @@ class SynchrotronPresence(object):
         """
         going_offline = self.users_going_offline.pop(user_id, None)
         if not going_offline:
-            # Safe to skip if we haven't yet told the master they were offline
+            # Safe to skip because we haven't yet told the master they were offline
             self.send_user_sync(user_id, True, self.clock.time_msec())
 
     def mark_as_going_offline(self, user_id):
-- 
cgit 1.5.1


From 29574fd5b3537cc272a4d792669b8d5be2a92b6f Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Mon, 10 Apr 2017 16:48:30 +0100
Subject: Reduce federation presence replication traffic

This is mainly done by moving the calculation of where to send presence
updates from the presence handler to the transaction queue, so we only
need to send the presence event (and not the destinations) across the
replication connection. Before we were duplicating by sending the full
state across once per destination.
---
 synapse/app/federation_sender.py            | 12 ++++
 synapse/app/synchrotron.py                  |  6 +-
 synapse/federation/send_queue.py            | 47 ++++++--------
 synapse/federation/transaction_queue.py     | 99 ++++++++++++++++++++++++++---
 synapse/handlers/presence.py                | 54 ++++------------
 synapse/replication/slave/storage/events.py |  1 +
 6 files changed, 139 insertions(+), 80 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/federation_sender.py b/synapse/app/federation_sender.py
index 477e16e0fa..49efb602bc 100644
--- a/synapse/app/federation_sender.py
+++ b/synapse/app/federation_sender.py
@@ -32,6 +32,7 @@ from synapse.replication.slave.storage.transactions import TransactionStore
 from synapse.replication.slave.storage.devices import SlavedDeviceStore
 from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
+from synapse.storage.presence import PresenceStore
 from synapse.util.async import Linearizer
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
@@ -80,6 +81,17 @@ class FederationSenderSlaveStore(
 
         return rows[0][0] if rows else -1
 
+    # XXX: This is a bit broken because we don't persist the accepted list in a
+    # way that can be replicated. This means that we don't have a way to
+    # invalidate the cache correctly.
+    # This is fine since in practice nobody uses the presence list stuff...
+    get_presence_list_accepted = PresenceStore.__dict__[
+        "get_presence_list_accepted"
+    ]
+    get_presence_list_observers_accepted = PresenceStore.__dict__[
+        "get_presence_list_observers_accepted"
+    ]
+
 
 class FederationSenderServer(HomeServer):
     def get_db_conn(self, run_new_connection=True):
diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index d39e3161fe..7b6f82abdc 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -206,10 +206,8 @@ class SynchrotronPresence(object):
 
     @defer.inlineCallbacks
     def notify_from_replication(self, states, stream_id):
-        parties = yield self._get_interested_parties(
-            states, calculate_remote_hosts=False
-        )
-        room_ids_to_states, users_to_states, _ = parties
+        parties = yield self._get_interested_parties(states)
+        room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
             "presence_key", stream_id, rooms=room_ids_to_states.keys(),
diff --git a/synapse/federation/send_queue.py b/synapse/federation/send_queue.py
index 748548bbe2..a12c18f4df 100644
--- a/synapse/federation/send_queue.py
+++ b/synapse/federation/send_queue.py
@@ -53,6 +53,7 @@ class FederationRemoteSendQueue(object):
         self.server_name = hs.hostname
         self.clock = hs.get_clock()
         self.notifier = hs.get_notifier()
+        self.is_mine_id = hs.is_mine_id
 
         self.presence_map = {}
         self.presence_changed = sorteddict()
@@ -120,7 +121,9 @@ class FederationRemoteSendQueue(object):
                 del self.presence_changed[key]
 
             user_ids = set(
-                user_id for uids in self.presence_changed.values() for _, user_id in uids
+                user_id
+                for uids in self.presence_changed.itervalues()
+                for user_id in uids
             )
 
             to_del = [
@@ -187,18 +190,14 @@ class FederationRemoteSendQueue(object):
 
         self.notifier.on_new_replication_data()
 
-    def send_presence(self, destination, states):
+    def send_presence(self, states):
         """As per TransactionQueue"""
         pos = self._next_pos()
 
-        self.presence_map.update({
-            state.user_id: state
-            for state in states
-        })
+        local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
 
-        self.presence_changed[pos] = [
-            (destination, state.user_id) for state in states
-        ]
+        self.presence_map.update({state.user_id: state for state in local_states})
+        self.presence_changed[pos] = [state.user_id for state in local_states]
 
         self.notifier.on_new_replication_data()
 
@@ -251,15 +250,14 @@ class FederationRemoteSendQueue(object):
         keys = self.presence_changed.keys()
         i = keys.bisect_right(from_token)
         j = keys.bisect_right(to_token) + 1
-        dest_user_ids = set(
-            (pos, dest_user_id)
+        dest_user_ids = [
+            (pos, user_id)
             for pos in keys[i:j]
-            for dest_user_id in self.presence_changed[pos]
-        )
+            for user_id in self.presence_changed[pos]
+        ]
 
-        for (key, (dest, user_id)) in dest_user_ids:
+        for (key, user_id) in dest_user_ids:
             rows.append((key, PresenceRow(
-                destination=dest,
                 state=self.presence_map[user_id],
             )))
 
@@ -354,7 +352,6 @@ class BaseFederationRow(object):
 
 
 class PresenceRow(BaseFederationRow, namedtuple("PresenceRow", (
-    "destination",  # str
     "state",  # UserPresenceState
 ))):
     TypeId = "p"
@@ -362,18 +359,14 @@ class PresenceRow(BaseFederationRow, namedtuple("PresenceRow", (
     @staticmethod
     def from_data(data):
         return PresenceRow(
-            destination=data["destination"],
-            state=UserPresenceState.from_dict(data["state"])
+            state=UserPresenceState.from_dict(data)
         )
 
     def to_data(self):
-        return {
-            "destination": self.destination,
-            "state": self.state.as_dict()
-        }
+        return self.state.as_dict()
 
     def add_to_buffer(self, buff):
-        buff.presence.setdefault(self.destination, []).append(self.state)
+        buff.presence.append(self.state)
 
 
 class KeyedEduRow(BaseFederationRow, namedtuple("KeyedEduRow", (
@@ -469,7 +462,7 @@ TypeToRow = {
 
 
 ParsedFederationStreamData = namedtuple("ParsedFederationStreamData", (
-    "presence",  # dict of destination -> [UserPresenceState]
+    "presence",  # list(UserPresenceState)
     "keyed_edus",  # dict of destination -> { key -> Edu }
     "edus",  # dict of destination -> [Edu]
     "failures",  # dict of destination -> [failures]
@@ -491,7 +484,7 @@ def process_rows_for_federation(transaction_queue, rows):
     # them into the appropriate collection and then send them off.
 
     buff = ParsedFederationStreamData(
-        presence={},
+        presence=[],
         keyed_edus={},
         edus={},
         failures={},
@@ -508,8 +501,8 @@ def process_rows_for_federation(transaction_queue, rows):
         parsed_row = RowType.from_data(row.data)
         parsed_row.add_to_buffer(buff)
 
-    for destination, states in buff.presence.iteritems():
-        transaction_queue.send_presence(destination, states)
+    if buff.presence:
+        transaction_queue.send_presence(buff.presence)
 
     for destination, edu_map in buff.keyed_edus.iteritems():
         for key, edu in edu_map.items():
diff --git a/synapse/federation/transaction_queue.py b/synapse/federation/transaction_queue.py
index c27ce7c5f3..fd9e1fa01c 100644
--- a/synapse/federation/transaction_queue.py
+++ b/synapse/federation/transaction_queue.py
@@ -21,7 +21,7 @@ from .units import Transaction, Edu
 
 from synapse.api.errors import HttpResponseException
 from synapse.util.async import run_on_reactor
-from synapse.util.logcontext import preserve_context_over_fn
+from synapse.util.logcontext import preserve_context_over_fn, preserve_fn
 from synapse.util.retryutils import NotRetryingDestination, get_retry_limiter
 from synapse.util.metrics import measure_func
 from synapse.types import get_domain_from_id
@@ -78,6 +78,7 @@ class TransactionQueue(object):
         self.pending_edus_by_dest = edus = {}
 
         # Presence needs to be separate as we send single aggragate EDUs
+        self.pending_presence = {}
         self.pending_presence_by_dest = presence = {}
         self.pending_edus_keyed_by_dest = edus_keyed = {}
 
@@ -113,6 +114,8 @@ class TransactionQueue(object):
         self._is_processing = False
         self._last_poked_id = -1
 
+        self._processing_pending_presence = False
+
     def can_send_to(self, destination):
         """Can we send messages to the given server?
 
@@ -224,17 +227,95 @@ class TransactionQueue(object):
                 self._attempt_new_transaction, destination
             )
 
-    def send_presence(self, destination, states):
-        if not self.can_send_to(destination):
+    @preserve_fn
+    @defer.inlineCallbacks
+    def send_presence(self, states):
+        """Send the new presence states to the appropriate destinations.
+
+        Args:
+            states (list(UserPresenceState))
+        """
+
+        # First we queue up the new presence by user ID, so multiple presence
+        # updates in quick successtion are correctly handled
+        self.pending_presence.update({state.user_id: state for state in states})
+
+        # We then handle the new pending presence in batches, first figuring
+        # out the destinations we need to send each state to and then poking it
+        # to attempt a new transaction. We linearize this so that we don't
+        # accidentally mess up the ordering and send multiple presence updates
+        # in the wrong order
+        if self._processing_pending_presence:
             return
 
-        self.pending_presence_by_dest.setdefault(destination, {}).update({
-            state.user_id: state for state in states
-        })
+        self._processing_pending_presence = True
+        try:
+            while True:
+                states = self.pending_presence
+                self.pending_presence = {}
 
-        preserve_context_over_fn(
-            self._attempt_new_transaction, destination
-        )
+                if not states:
+                    break
+
+                yield self._process_presence_inner(states)
+        finally:
+            self._processing_pending_presence = False
+
+    @measure_func("txnqueue._process_presence")
+    @defer.inlineCallbacks
+    def _process_presence_inner(self, states):
+        """Given a list of states populate self.pending_presence_by_dest and
+        poke to send a new transaction to each destination
+
+        Args:
+            states (list(UserPresenceState))
+        """
+        # First we look up the rooms each user is in (as well as any explicit
+        # subscriptions), then for each distinct room we look up the remote
+        # hosts in those rooms.
+        room_ids_to_states = {}
+        users_to_states = {}
+        for state in states.itervalues():
+            room_ids = yield self.store.get_rooms_for_user(state.user_id)
+            for room_id in room_ids:
+                room_ids_to_states.setdefault(room_id, []).append(state)
+
+            plist = yield self.store.get_presence_list_observers_accepted(
+                state.user_id,
+            )
+            for u in plist:
+                users_to_states.setdefault(u, []).append(state)
+
+        hosts_and_states = []
+        for room_id, states in room_ids_to_states.items():
+            local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
+            if not local_states:
+                continue
+
+            hosts = yield self.store.get_hosts_in_room(room_id)
+            hosts_and_states.append((hosts, local_states))
+
+        for user_id, states in users_to_states.items():
+            local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
+            if not local_states:
+                continue
+
+            host = get_domain_from_id(user_id)
+            hosts_and_states.append(([host], local_states))
+
+        # And now finally queue up new transactions
+        for destinations, states in hosts_and_states:
+            for destination in destinations:
+                if not self.can_send_to(destination):
+                    continue
+
+                self.pending_presence_by_dest.setdefault(
+                    destination, {}
+                ).update({
+                    state.user_id: state for state in states
+                })
+
+                preserve_fn(self._attempt_new_transaction)(destination)
 
     def send_edu(self, destination, edu_type, content, key=None):
         edu = Edu(
diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py
index 9ed5af3cb4..c1c0dd4d3d 100644
--- a/synapse/handlers/presence.py
+++ b/synapse/handlers/presence.py
@@ -318,11 +318,7 @@ class PresenceHandler(object):
             if to_federation_ping:
                 federation_presence_out_counter.inc_by(len(to_federation_ping))
 
-                _, _, hosts_to_states = yield self._get_interested_parties(
-                    to_federation_ping.values()
-                )
-
-                self._push_to_remotes(hosts_to_states)
+                self._push_to_remotes(to_federation_ping.values())
 
     def _handle_timeouts(self):
         """Checks the presence of users that have timed out and updates as
@@ -615,12 +611,12 @@ class PresenceHandler(object):
         defer.returnValue(states)
 
     @defer.inlineCallbacks
-    def _get_interested_parties(self, states, calculate_remote_hosts=True):
+    def _get_interested_parties(self, states):
         """Given a list of states return which entities (rooms, users, servers)
         are interested in the given states.
 
         Returns:
-            3-tuple: `(room_ids_to_states, users_to_states, hosts_to_states)`,
+            2-tuple: `(room_ids_to_states, users_to_states)`,
             with each item being a dict of `entity_name` -> `[UserPresenceState]`
         """
         room_ids_to_states = {}
@@ -637,30 +633,10 @@ class PresenceHandler(object):
             # Always notify self
             users_to_states.setdefault(state.user_id, []).append(state)
 
-        hosts_to_states = {}
-        if calculate_remote_hosts:
-            for room_id, states in room_ids_to_states.items():
-                local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
-                if not local_states:
-                    continue
-
-                hosts = yield self.store.get_hosts_in_room(room_id)
-
-                for host in hosts:
-                    hosts_to_states.setdefault(host, []).extend(local_states)
-
-        for user_id, states in users_to_states.items():
-            local_states = filter(lambda s: self.is_mine_id(s.user_id), states)
-            if not local_states:
-                continue
-
-            host = get_domain_from_id(user_id)
-            hosts_to_states.setdefault(host, []).extend(local_states)
-
         # TODO: de-dup hosts_to_states, as a single host might have multiple
         # of same presence
 
-        defer.returnValue((room_ids_to_states, users_to_states, hosts_to_states))
+        defer.returnValue((room_ids_to_states, users_to_states))
 
     @defer.inlineCallbacks
     def _persist_and_notify(self, states):
@@ -670,33 +646,32 @@ class PresenceHandler(object):
         stream_id, max_token = yield self.store.update_presence(states)
 
         parties = yield self._get_interested_parties(states)
-        room_ids_to_states, users_to_states, hosts_to_states = parties
+        room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
             "presence_key", stream_id, rooms=room_ids_to_states.keys(),
-            users=[UserID.from_string(u) for u in users_to_states.keys()]
+            users=[UserID.from_string(u) for u in users_to_states]
         )
 
-        self._push_to_remotes(hosts_to_states)
+        self._push_to_remotes(states)
 
     @defer.inlineCallbacks
     def notify_for_states(self, state, stream_id):
         parties = yield self._get_interested_parties([state])
-        room_ids_to_states, users_to_states, hosts_to_states = parties
+        room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
             "presence_key", stream_id, rooms=room_ids_to_states.keys(),
-            users=[UserID.from_string(u) for u in users_to_states.keys()]
+            users=[UserID.from_string(u) for u in users_to_states]
         )
 
-    def _push_to_remotes(self, hosts_to_states):
+    def _push_to_remotes(self, states):
         """Sends state updates to remote servers.
 
         Args:
-            hosts_to_states (dict): Mapping `server_name` -> `[UserPresenceState]`
+            hosts_to_states (list): list(state)
         """
-        for host, states in hosts_to_states.items():
-            self.federation.send_presence(host, states)
+        self.federation.send_presence(states)
 
     @defer.inlineCallbacks
     def incoming_presence(self, origin, content):
@@ -837,14 +812,13 @@ class PresenceHandler(object):
         if self.is_mine(user):
             state = yield self.current_state_for_user(user.to_string())
 
-            hosts = set(get_domain_from_id(u) for u in user_ids)
-            self._push_to_remotes({host: (state,) for host in hosts})
+            self._push_to_remotes([state])
         else:
             user_ids = filter(self.is_mine_id, user_ids)
 
             states = yield self.current_state_for_users(user_ids)
 
-            self._push_to_remotes({user.domain: states.values()})
+            self._push_to_remotes(states.values())
 
     @defer.inlineCallbacks
     def get_presence_list(self, observer_user, accepted=None):
diff --git a/synapse/replication/slave/storage/events.py b/synapse/replication/slave/storage/events.py
index 5fd47706ef..4ca1e5aa8c 100644
--- a/synapse/replication/slave/storage/events.py
+++ b/synapse/replication/slave/storage/events.py
@@ -71,6 +71,7 @@ class SlavedEventStore(BaseSlavedStore):
     # to reach inside the __dict__ to extract them.
     get_rooms_for_user = RoomMemberStore.__dict__["get_rooms_for_user"]
     get_users_in_room = RoomMemberStore.__dict__["get_users_in_room"]
+    get_hosts_in_room = RoomMemberStore.__dict__["get_hosts_in_room"]
     get_users_who_share_room_with_user = (
         RoomMemberStore.__dict__["get_users_who_share_room_with_user"]
     )
-- 
cgit 1.5.1


From 414522aed59aca9b711253aba98d15e2d59e14f9 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Tue, 11 Apr 2017 15:30:02 +0100
Subject: Move get_interested_parties

---
 synapse/app/synchrotron.py   |  5 ++--
 synapse/handlers/presence.py | 69 ++++++++++++++++++++++----------------------
 2 files changed, 36 insertions(+), 38 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index 7b6f82abdc..e3fbf02c9c 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -20,7 +20,7 @@ from synapse.api.constants import EventTypes
 from synapse.config._base import ConfigError
 from synapse.config.homeserver import HomeServerConfig
 from synapse.config.logger import setup_logging
-from synapse.handlers.presence import PresenceHandler
+from synapse.handlers.presence import PresenceHandler, get_interested_parties
 from synapse.http.site import SynapseSite
 from synapse.http.server import JsonResource
 from synapse.metrics.resource import MetricsResource, METRICS_PREFIX
@@ -172,7 +172,6 @@ class SynchrotronPresence(object):
 
     get_states = PresenceHandler.get_states.__func__
     get_state = PresenceHandler.get_state.__func__
-    _get_interested_parties = PresenceHandler._get_interested_parties.__func__
     current_state_for_users = PresenceHandler.current_state_for_users.__func__
 
     def user_syncing(self, user_id, affect_presence):
@@ -206,7 +205,7 @@ class SynchrotronPresence(object):
 
     @defer.inlineCallbacks
     def notify_from_replication(self, states, stream_id):
-        parties = yield self._get_interested_parties(states)
+        parties = yield get_interested_parties(self.store, states)
         room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
diff --git a/synapse/handlers/presence.py b/synapse/handlers/presence.py
index 98e736be5b..b9ce997a94 100644
--- a/synapse/handlers/presence.py
+++ b/synapse/handlers/presence.py
@@ -610,31 +610,6 @@ class PresenceHandler(object):
 
         defer.returnValue(states)
 
-    @defer.inlineCallbacks
-    def _get_interested_parties(self, states):
-        """Given a list of states return which entities (rooms, users, servers)
-        are interested in the given states.
-
-        Returns:
-            2-tuple: `(room_ids_to_states, users_to_states)`,
-            with each item being a dict of `entity_name` -> `[UserPresenceState]`
-        """
-        room_ids_to_states = {}
-        users_to_states = {}
-        for state in states:
-            room_ids = yield self.store.get_rooms_for_user(state.user_id)
-            for room_id in room_ids:
-                room_ids_to_states.setdefault(room_id, []).append(state)
-
-            plist = yield self.store.get_presence_list_observers_accepted(state.user_id)
-            for u in plist:
-                users_to_states.setdefault(u, []).append(state)
-
-            # Always notify self
-            users_to_states.setdefault(state.user_id, []).append(state)
-
-        defer.returnValue((room_ids_to_states, users_to_states))
-
     @defer.inlineCallbacks
     def _persist_and_notify(self, states):
         """Persist states in the database, poke the notifier and send to
@@ -642,7 +617,7 @@ class PresenceHandler(object):
         """
         stream_id, max_token = yield self.store.update_presence(states)
 
-        parties = yield self._get_interested_parties(states)
+        parties = yield get_interested_parties(self.store, states)
         room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
@@ -654,7 +629,7 @@ class PresenceHandler(object):
 
     @defer.inlineCallbacks
     def notify_for_states(self, state, stream_id):
-        parties = yield self._get_interested_parties([state])
+        parties = yield get_interested_parties(self.store, [state])
         room_ids_to_states, users_to_states = parties
 
         self.notifier.on_new_event(
@@ -1316,6 +1291,36 @@ def handle_update(prev_state, new_state, is_mine, wheel_timer, now):
 
     return new_state, persist_and_notify, federation_ping
 
+
+@defer.inlineCallbacks
+def get_interested_parties(store, states):
+    """Given a list of states return which entities (rooms, users)
+    are interested in the given states.
+
+    Args:
+        states (list(UserPresenceState))
+
+    Returns:
+        2-tuple: `(room_ids_to_states, users_to_states)`,
+        with each item being a dict of `entity_name` -> `[UserPresenceState]`
+    """
+    room_ids_to_states = {}
+    users_to_states = {}
+    for state in states:
+        room_ids = yield store.get_rooms_for_user(state.user_id)
+        for room_id in room_ids:
+            room_ids_to_states.setdefault(room_id, []).append(state)
+
+        plist = yield store.get_presence_list_observers_accepted(state.user_id)
+        for u in plist:
+            users_to_states.setdefault(u, []).append(state)
+
+        # Always notify self
+        users_to_states.setdefault(state.user_id, []).append(state)
+
+    defer.returnValue((room_ids_to_states, users_to_states))
+
+
 @defer.inlineCallbacks
 def get_interested_remotes(store, states):
     """Given a list of presence states figure out which remote servers
@@ -1351,17 +1356,11 @@ def get_interested_remotes(store, states):
             users_to_states.setdefault(u, []).append(state)
 
     for room_id, states in room_ids_to_states.items():
-        if not local_states:
-            continue
-
         hosts = yield store.get_hosts_in_room(room_id)
-        hosts_and_states.append((hosts, local_states))
+        hosts_and_states.append((hosts, states))
 
     for user_id, states in users_to_states.items():
-        if not local_states:
-            continue
-
         host = get_domain_from_id(user_id)
-        hosts_and_states.append(([host], local_states))
+        hosts_and_states.append(([host], states))
 
     defer.returnValue(hosts_and_states)
-- 
cgit 1.5.1


From 9c712a366fbbba90dbe18f7246c279d3cbb1cf10 Mon Sep 17 00:00:00 2001
From: Erik Johnston <erik@matrix.org>
Date: Tue, 11 Apr 2017 16:07:33 +0100
Subject: Move get_presence_list_* to SlaveStore

---
 synapse/app/federation_sender.py              | 15 ++-------------
 synapse/app/synchrotron.py                    | 12 +-----------
 synapse/replication/slave/storage/presence.py | 10 ++++++++++
 3 files changed, 13 insertions(+), 24 deletions(-)

(limited to 'synapse/app/synchrotron.py')

diff --git a/synapse/app/federation_sender.py b/synapse/app/federation_sender.py
index 49efb602bc..e51a69074d 100644
--- a/synapse/app/federation_sender.py
+++ b/synapse/app/federation_sender.py
@@ -28,11 +28,11 @@ from synapse.replication.slave.storage.deviceinbox import SlavedDeviceInboxStore
 from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.slave.storage.presence import SlavedPresenceStore
 from synapse.replication.slave.storage.transactions import TransactionStore
 from synapse.replication.slave.storage.devices import SlavedDeviceStore
 from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
-from synapse.storage.presence import PresenceStore
 from synapse.util.async import Linearizer
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
@@ -56,7 +56,7 @@ logger = logging.getLogger("synapse.app.appservice")
 
 class FederationSenderSlaveStore(
     SlavedDeviceInboxStore, TransactionStore, SlavedReceiptsStore, SlavedEventStore,
-    SlavedRegistrationStore, SlavedDeviceStore,
+    SlavedRegistrationStore, SlavedDeviceStore, SlavedPresenceStore,
 ):
     def __init__(self, db_conn, hs):
         super(FederationSenderSlaveStore, self).__init__(db_conn, hs)
@@ -81,17 +81,6 @@ class FederationSenderSlaveStore(
 
         return rows[0][0] if rows else -1
 
-    # XXX: This is a bit broken because we don't persist the accepted list in a
-    # way that can be replicated. This means that we don't have a way to
-    # invalidate the cache correctly.
-    # This is fine since in practice nobody uses the presence list stuff...
-    get_presence_list_accepted = PresenceStore.__dict__[
-        "get_presence_list_accepted"
-    ]
-    get_presence_list_observers_accepted = PresenceStore.__dict__[
-        "get_presence_list_observers_accepted"
-    ]
-
 
 class FederationSenderServer(HomeServer):
     def get_db_conn(self, run_new_connection=True):
diff --git a/synapse/app/synchrotron.py b/synapse/app/synchrotron.py
index e3fbf02c9c..13c00ef2ba 100644
--- a/synapse/app/synchrotron.py
+++ b/synapse/app/synchrotron.py
@@ -44,7 +44,7 @@ from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
-from synapse.storage.presence import PresenceStore, UserPresenceState
+from synapse.storage.presence import UserPresenceState
 from synapse.storage.roommember import RoomMemberStore
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext, preserve_fn
@@ -89,16 +89,6 @@ class SynchrotronSlavedStore(
         RoomMemberStore.__dict__["did_forget"]
     )
 
-    # XXX: This is a bit broken because we don't persist the accepted list in a
-    # way that can be replicated. This means that we don't have a way to
-    # invalidate the cache correctly.
-    get_presence_list_accepted = PresenceStore.__dict__[
-        "get_presence_list_accepted"
-    ]
-    get_presence_list_observers_accepted = PresenceStore.__dict__[
-        "get_presence_list_observers_accepted"
-    ]
-
 
 UPDATE_SYNCING_USERS_MS = 10 * 1000
 
diff --git a/synapse/replication/slave/storage/presence.py b/synapse/replication/slave/storage/presence.py
index dffc80adc3..cfb9280181 100644
--- a/synapse/replication/slave/storage/presence.py
+++ b/synapse/replication/slave/storage/presence.py
@@ -39,6 +39,16 @@ class SlavedPresenceStore(BaseSlavedStore):
     _get_presence_for_user = PresenceStore.__dict__["_get_presence_for_user"]
     get_presence_for_users = PresenceStore.__dict__["get_presence_for_users"]
 
+    # XXX: This is a bit broken because we don't persist the accepted list in a
+    # way that can be replicated. This means that we don't have a way to
+    # invalidate the cache correctly.
+    get_presence_list_accepted = PresenceStore.__dict__[
+        "get_presence_list_accepted"
+    ]
+    get_presence_list_observers_accepted = PresenceStore.__dict__[
+        "get_presence_list_observers_accepted"
+    ]
+
     def get_current_presence_token(self):
         return self._presence_id_gen.get_current_token()
 
-- 
cgit 1.5.1