Update all the workers and master to use TCP replication

synapse/app/appservice.py

@@ -26,8 +26,8 @@ from synapse.replication.slave.storage.directory import DirectoryStore
 from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
@@ -36,7 +36,7 @@ from synapse.util.versionstring import get_version_string
 
 from synapse import events
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -120,30 +120,23 @@ class AppserviceServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        appservice_handler = self.get_application_service_handler()
 
-        @defer.inlineCallbacks
+    def build_tcp_replication(self):
-        def replicate(results):
+        return ASReplicationHandler(self)
-            stream = results.get("events")
-            if stream:
-                max_stream_id = stream["position"]
-                yield appservice_handler.notify_interested_services(max_stream_id)
 
-        while True:
+
-            try:
+class ASReplicationHandler(ReplicationClientHandler):
-                args = store.stream_positions()
+    def __init__(self, hs):
-                args["timeout"] = 30000
+        super(ASReplicationHandler, self).__init__(hs.get_datastore())
-                result = yield http_client.get_json(replication_url, args=args)
+        self.appservice_handler = hs.get_application_service_handler()
-                yield store.process_replication(result)
+
-                replicate(result)
+    def on_rdata(self, stream_name, token, rows):
-            except:
+        super(ASReplicationHandler, self).on_rdata(stream_name, token, rows)
-                logger.exception("Error replicating from %r", replication_url)
+
-                yield sleep(30)
+        if stream_name == "events":
+            max_stream_id = self.store.get_room_max_stream_ordering()
+            self.appservice_handler.notify_interested_services(max_stream_id)
 
 
 def start(config_options):
@@ -199,7 +192,6 @@ def start(config_options):
             reactor.run()
 
     def start():
-        ps.replicate()
         ps.get_datastore().start_profiling()
         ps.get_state_handler().start_caching()
 

synapse/app/client_reader.py

@@ -30,11 +30,11 @@ from synapse.replication.slave.storage.room import RoomStore
 from synapse.replication.slave.storage.directory import DirectoryStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
 from synapse.replication.slave.storage.transactions import TransactionStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.rest.client.v1.room import PublicRoomListRestServlet
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
@@ -45,7 +45,7 @@ from synapse.crypto import context_factory
 from synapse import events
 
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -145,21 +145,10 @@ class ClientReaderServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
 
-        while True:
+    def build_tcp_replication(self):
-            try:
+        return ReplicationClientHandler(self.get_datastore())
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
 
 
 def start(config_options):
@@ -209,7 +198,6 @@ def start(config_options):
     def start():
         ss.get_state_handler().start_caching()
         ss.get_datastore().start_profiling()
-        ss.replicate()
 
     reactor.callWhenRunning(start)
 

synapse/app/federation_reader.py

@@ -27,9 +27,9 @@ from synapse.replication.slave.storage.keys import SlavedKeyStore
 from synapse.replication.slave.storage.room import RoomStore
 from synapse.replication.slave.storage.transactions import TransactionStore
 from synapse.replication.slave.storage.directory import DirectoryStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.server import HomeServer
 from synapse.storage.engines import create_engine
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
@@ -42,7 +42,7 @@ from synapse.crypto import context_factory
 from synapse import events
 
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -134,21 +134,10 @@ class FederationReaderServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
 
-        while True:
+    def build_tcp_replication(self):
-            try:
+        return ReplicationClientHandler(self.get_datastore())
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
 
 
 def start(config_options):
@@ -198,7 +187,6 @@ def start(config_options):
     def start():
         ss.get_state_handler().start_caching()
         ss.get_datastore().start_profiling()
-        ss.replicate()
 
     reactor.callWhenRunning(start)
 

synapse/app/federation_sender.py

@@ -31,9 +31,10 @@ from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
 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 UserPresenceState
-from synapse.util.async import sleep
+from synapse.util.async import Linearizer
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
@@ -59,7 +60,23 @@ class FederationSenderSlaveStore(
     SlavedDeviceInboxStore, TransactionStore, SlavedReceiptsStore, SlavedEventStore,
     SlavedRegistrationStore, SlavedDeviceStore,
 ):
-    pass
+    def __init__(self, db_conn, hs):
+        super(FederationSenderSlaveStore, self).__init__(db_conn, hs)
+        self.federation_out_pos_startup = self._get_federation_out_pos(db_conn)
+
+    def _get_federation_out_pos(self, db_conn):
+        sql = (
+            "SELECT stream_id FROM federation_stream_position"
+            " WHERE type = ?"
+        )
+        sql = self.database_engine.convert_param_style(sql)
+
+        txn = db_conn.cursor()
+        txn.execute(sql, ("federation",))
+        rows = txn.fetchall()
+        txn.close()
+
+        return rows[0][0] if rows else -1
 
 
 class FederationSenderServer(HomeServer):
@@ -127,26 +144,29 @@ class FederationSenderServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        send_handler = FederationSenderHandler(self)
 
-        send_handler.on_start()
+    def build_tcp_replication(self):
+        return FederationSenderReplicationHandler(self)
 
-        while True:
+
-            try:
+class FederationSenderReplicationHandler(ReplicationClientHandler):
-                args = store.stream_positions()
+    def __init__(self, hs):
-                args.update((yield send_handler.stream_positions()))
+        super(FederationSenderReplicationHandler, self).__init__(hs.get_datastore())
-                args["timeout"] = 30000
+        self.send_handler = FederationSenderHandler(hs)
-                result = yield http_client.get_json(replication_url, args=args)
+
-                yield store.process_replication(result)
+    def on_rdata(self, stream_name, token, rows):
-                yield send_handler.process_replication(result)
+        super(FederationSenderReplicationHandler, self).on_rdata(
-            except:
+            stream_name, token, rows
-                logger.exception("Error replicating from %r", replication_url)
+        )
-                yield sleep(30)
+        self.send_handler.process_replication_rows(stream_name, token, rows)
+        if stream_name == "federation":
+            self.send_federation_ack(token)
+
+    def get_streams_to_replicate(self):
+        args = super(FederationSenderReplicationHandler, self).get_streams_to_replicate()
+        args.update(self.send_handler.stream_positions())
+        return args
 
 
 def start(config_options):
@@ -205,7 +225,6 @@ def start(config_options):
             reactor.run()
 
     def start():
-        ps.replicate()
         ps.get_datastore().start_profiling()
         ps.get_state_handler().start_caching()
 
@@ -233,6 +252,9 @@ class FederationSenderHandler(object):
         self.store = hs.get_datastore()
         self.federation_sender = hs.get_federation_sender()
 
+        self.federation_position = self.store.federation_out_pos_startup
+        self._fed_position_linearizer = Linearizer(name="_fed_position_linearizer")
+
         self._room_serials = {}
         self._room_typing = {}
 
@@ -243,25 +265,13 @@ class FederationSenderHandler(object):
             self.store.get_room_max_stream_ordering()
         )
 
-    @defer.inlineCallbacks
     def stream_positions(self):
-        stream_id = yield self.store.get_federation_out_pos("federation")
+        return {"federation": self.federation_position}
-        defer.returnValue({
-            "federation": stream_id,
 
-            # Ack stuff we've "processed", this should only be called from
+    def process_replication_rows(self, stream_name, token, rows):
-            # one process.
-            "federation_ack": stream_id,
-        })
-
-    @defer.inlineCallbacks
-    def process_replication(self, result):
         # The federation stream contains things that we want to send out, e.g.
         # presence, typing, etc.
-        fed_stream = result.get("federation")
+        if stream_name == "federation":
-        if fed_stream:
-            latest_id = int(fed_stream["position"])
-
             # The federation stream containis a bunch of different types of
             # rows that need to be handled differently. We parse the rows, put
             # them into the appropriate collection and then send them off.
@@ -272,8 +282,9 @@ class FederationSenderHandler(object):
             device_destinations = set()
 
             # Parse the rows in the stream
-            for row in fed_stream["rows"]:
+            for row in rows:
-                position, typ, content_js = row
+                typ = row.type
+                content_js = row.data
                 content = json.loads(content_js)
 
                 if typ == send_queue.PRESENCE_TYPE:
@@ -325,16 +336,19 @@ class FederationSenderHandler(object):
             for destination in device_destinations:
                 self.federation_sender.send_device_messages(destination)
 
-            # Record where we are in the stream.
+            self.update_token(token)
-            yield self.store.update_federation_out_pos(
-                "federation", latest_id
-            )
 
         # We also need to poke the federation sender when new events happen
-        event_stream = result.get("events")
+        elif stream_name == "events":
-        if event_stream:
+            self.federation_sender.notify_new_events(token)
-            latest_pos = event_stream["position"]
+
-            self.federation_sender.notify_new_events(latest_pos)
+    @defer.inlineCallbacks
+    def update_token(self, token):
+        self.federation_position = token
+        with (yield self._fed_position_linearizer.queue(None)):
+            yield self.store.update_federation_out_pos(
+                "federation", self.federation_position
+            )
 
 
 if __name__ == '__main__':

synapse/app/media_repository.py

@@ -25,13 +25,13 @@ from synapse.replication.slave.storage._base import BaseSlavedStore
 from synapse.replication.slave.storage.appservice import SlavedApplicationServiceStore
 from synapse.replication.slave.storage.registration import SlavedRegistrationStore
 from synapse.replication.slave.storage.transactions import TransactionStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.rest.media.v0.content_repository import ContentRepoResource
 from synapse.rest.media.v1.media_repository import MediaRepositoryResource
 from synapse.server import HomeServer
 from synapse.storage.client_ips import ClientIpStore
 from synapse.storage.engines import create_engine
 from synapse.storage.media_repository import MediaRepositoryStore
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
 from synapse.util.manhole import manhole
@@ -45,7 +45,7 @@ from synapse.crypto import context_factory
 from synapse import events
 
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -142,21 +142,10 @@ class MediaRepositoryServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
 
-        while True:
+    def build_tcp_replication(self):
-            try:
+        return ReplicationClientHandler(self.get_datastore())
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(5)
 
 
 def start(config_options):
@@ -206,7 +195,6 @@ def start(config_options):
     def start():
         ss.get_state_handler().start_caching()
         ss.get_datastore().start_profiling()
-        ss.replicate()
 
     reactor.callWhenRunning(start)
 

synapse/app/pusher.py

@@ -27,9 +27,9 @@ from synapse.replication.slave.storage.events import SlavedEventStore
 from synapse.replication.slave.storage.pushers import SlavedPusherStore
 from synapse.replication.slave.storage.receipts import SlavedReceiptsStore
 from synapse.replication.slave.storage.account_data import SlavedAccountDataStore
+from synapse.replication.tcp.client import ReplicationClientHandler
 from synapse.storage.engines import create_engine
 from synapse.storage import DataStore
-from synapse.util.async import sleep
 from synapse.util.httpresourcetree import create_resource_tree
 from synapse.util.logcontext import LoggingContext, preserve_fn, \
     PreserveLoggingContext
@@ -39,7 +39,7 @@ from synapse.util.versionstring import get_version_string
 
 from synapse import events
 
-from twisted.internet import reactor, defer
+from twisted.internet import reactor
 from twisted.web.resource import Resource
 
 from daemonize import Daemonize
@@ -89,7 +89,6 @@ class PusherSlaveStore(
 
 
 class PusherServer(HomeServer):
-
     def get_db_conn(self, run_new_connection=True):
         # Any param beginning with cp_ is a parameter for adbapi, and should
         # not be passed to the database engine.
@@ -109,16 +108,7 @@ class PusherServer(HomeServer):
         logger.info("Finished setting up.")
 
     def remove_pusher(self, app_id, push_key, user_id):
-        http_client = self.get_simple_http_client()
+        self.get_tcp_replication().send_remove_pusher(app_id, push_key, user_id)
-        replication_url = self.config.worker_replication_url
-        url = replication_url + "/remove_pushers"
-        return http_client.post_json_get_json(url, {
-            "remove": [{
-                "app_id": app_id,
-                "push_key": push_key,
-                "user_id": user_id,
-            }]
-        })
 
     def _listen_http(self, listener_config):
         port = listener_config["port"]
@@ -166,73 +156,51 @@ class PusherServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    def replicate(self):
-        http_client = self.get_simple_http_client()
-        store = self.get_datastore()
-        replication_url = self.config.worker_replication_url
-        pusher_pool = self.get_pusherpool()
 
-        def stop_pusher(user_id, app_id, pushkey):
+    def build_tcp_replication(self):
-            key = "%s:%s" % (app_id, pushkey)
+        return PusherReplicationHandler(self)
-            pushers_for_user = pusher_pool.pushers.get(user_id, {})
-            pusher = pushers_for_user.pop(key, None)
-            if pusher is None:
-                return
-            logger.info("Stopping pusher %r / %r", user_id, key)
-            pusher.on_stop()
 
-        def start_pusher(user_id, app_id, pushkey):
-            key = "%s:%s" % (app_id, pushkey)
-            logger.info("Starting pusher %r / %r", user_id, key)
-            return pusher_pool._refresh_pusher(app_id, pushkey, user_id)
 
-        @defer.inlineCallbacks
+class PusherReplicationHandler(ReplicationClientHandler):
-        def poke_pushers(results):
+    def __init__(self, hs):
-            pushers_rows = set(
+        super(PusherReplicationHandler, self).__init__(hs.get_datastore())
-                map(tuple, results.get("pushers", {}).get("rows", []))
+
+        self.pusher_pool = hs.get_pusherpool()
+
+    def on_rdata(self, stream_name, token, rows):
+        super(PusherReplicationHandler, self).on_rdata(stream_name, token, rows)
+        self.poke_pushers(stream_name, token, rows)
+
+    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)
+                else:
+                    self.start_pusher(row.user_id, row.app_id, row.pushkey)
+        elif stream_name == "events":
+            preserve_fn(self.pusher_pool.on_new_notifications)(
+                token, token,
             )
-            deleted_pushers_rows = set(
+        elif stream_name == "receipts":
-                map(tuple, results.get("deleted_pushers", {}).get("rows", []))
+            preserve_fn(self.pusher_pool.on_new_receipts)(
+                token, token, set(row.room_id for row in rows)
             )
-            for row in sorted(pushers_rows | deleted_pushers_rows):
-                if row in deleted_pushers_rows:
-                    user_id, app_id, pushkey = row[1:4]
-                    stop_pusher(user_id, app_id, pushkey)
-                elif row in pushers_rows:
-                    user_id = row[1]
-                    app_id = row[5]
-                    pushkey = row[8]
-                    yield start_pusher(user_id, app_id, pushkey)
 
-            stream = results.get("events")
+    def stop_pusher(self, user_id, app_id, pushkey):
-            if stream and stream["rows"]:
+        key = "%s:%s" % (app_id, pushkey)
-                min_stream_id = stream["rows"][0][0]
+        pushers_for_user = self.pusher_pool.pushers.get(user_id, {})
-                max_stream_id = stream["position"]
+        pusher = pushers_for_user.pop(key, None)
-                preserve_fn(pusher_pool.on_new_notifications)(
+        if pusher is None:
-                    min_stream_id, max_stream_id
+            return
-                )
+        logger.info("Stopping pusher %r / %r", user_id, key)
+        pusher.on_stop()
 
-            stream = results.get("receipts")
+    def start_pusher(self, user_id, app_id, pushkey):
-            if stream and stream["rows"]:
+        key = "%s:%s" % (app_id, pushkey)
-                rows = stream["rows"]
+        logger.info("Starting pusher %r / %r", user_id, key)
-                affected_room_ids = set(row[1] for row in rows)
+        return self.pusher_pool._refresh_pusher(app_id, pushkey, user_id)
-                min_stream_id = rows[0][0]
-                max_stream_id = stream["position"]
-                preserve_fn(pusher_pool.on_new_receipts)(
-                    min_stream_id, max_stream_id, affected_room_ids
-                )
-
-        while True:
-            try:
-                args = store.stream_positions()
-                args["timeout"] = 30000
-                result = yield http_client.get_json(replication_url, args=args)
-                yield store.process_replication(result)
-                poke_pushers(result)
-            except:
-                logger.exception("Error replicating from %r", replication_url)
-                yield sleep(30)
 
 
 def start(config_options):
@@ -288,7 +256,6 @@ def start(config_options):
             reactor.run()
 
     def start():
-        ps.replicate()
         ps.get_pusherpool().start()
         ps.get_datastore().start_profiling()
         ps.get_state_handler().start_caching()

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, \
+from synapse.util.logcontext import LoggingContext, PreserveLoggingContext
-    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.sync_callback = None
-        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)
 
     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:
+            if self.sync_callback:
-                # TODO: Don't block the sync request on this HTTP hit.
+                if self.user_to_num_current_syncs[user_id] == 1:
-                yield self._send_syncing_users_now()
+                    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())
+        return defer.succeed(_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()
 
     @defer.inlineCallbacks
     def notify_from_replication(self, states, stream_id):
@@ -223,26 +177,24 @@ class SynchrotronPresence(object):
         )
 
     @defer.inlineCallbacks
-    def process_replication(self, result):
+    def process_replication_rows(self, token, rows):
-        stream = result.get("presence", {"rows": []})
+        states = [UserPresenceState(
-        states = []
+            row.user_id, row.state, row.last_active_ts,
-        for row in stream["rows"]:
+            row.last_federation_update_ts, row.last_user_sync_ts, row.status_msg,
-            (
+            row.currently_active
-                position, user_id, state, last_active_ts,
+        ) for row in rows]
-                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)
 
-        if states and "position" in stream:
+        for state in states:
-            stream_id = int(stream["position"])
+            self.user_to_current_state[row.user_id] = state
-            yield self.notify_from_replication(states, stream_id)
+
+        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):
+    def process_replication_rows(self, token, rows):
-        stream = result.get("typing")
+        self._latest_room_serial = token
-        if stream:
-            self._latest_room_serial = int(stream["position"])
 
-            for row in stream["rows"]:
+        for row in rows:
-                position, room_id, typing_json = row
+            typing = json.loads(row.user_ids)
-                typing = json.loads(typing_json)
+            self._room_serials[row.room_id] = token
-                self._room_serials[room_id] = position
+            self._room_typing[row.room_id] = typing
-                self._room_typing[room_id] = typing
 
 
 class SynchrotronApplicationService(object):
@@ -351,118 +300,10 @@ class SynchrotronServer(HomeServer):
             else:
                 logger.warn("Unrecognized listener type: %s", listener["type"])
 
-    @defer.inlineCallbacks
+        self.get_tcp_replication().start_replication(self)
-    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(
+    def build_tcp_replication(self):
-            result, stream_name, stream_key, room=None, user=None
+        return SyncReplicationHandler(self)
-        ):
-            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
-                    )
-
-        @defer.inlineCallbacks
-        def notify_device_list_update(result):
-            stream = result.get("device_lists")
-            if not stream:
-                return
-
-            position_index = stream["field_names"].index("position")
-            user_index = stream["field_names"].index("user_id")
-
-            for row in stream["rows"]:
-                position = row[position_index]
-                user_id = row[user_index]
-
-                room_ids = yield store.get_rooms_for_user(user_id)
-
-                notifier.on_new_event(
-                    "device_list_key", position, rooms=room_ids,
-                )
-
-        @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
-                    )
-
-            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"
-            )
-            notify_from_stream(
-                result, "tag_account_data", "account_data_key", user="user_id"
-            )
-            notify_from_stream(
-                result, "receipts", "receipt_key", room="room_id"
-            )
-            notify_from_stream(
-                result, "typing", "typing_key", room="room_id"
-            )
-            notify_from_stream(
-                result, "to_device", "to_device_key", user="user_id"
-            )
-            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)
@@ -471,6 +312,80 @@ class SynchrotronServer(HomeServer):
         return SynchrotronTyping(self)
 
 
+class SyncReplicationHandler(ReplicationClientHandler):
+    def __init__(self, hs):
+        super(SyncReplicationHandler, self).__init__(hs.get_datastore())
+
+        self.store = hs.get_datastore()
+        self.typing_handler = hs.get_typing_handler()
+        self.presence_handler = hs.get_presence_handler()
+        self.notifier = hs.get_notifier()
+
+        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],
+            )
+        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],
+            )
+        elif stream_name == "receipts":
+            self.notifier.on_new_event(
+                "receipt_key", token, rooms=[row.room_id for row in rows],
+            )
+        elif stream_name == "typing":
+            self.notifier.on_new_event(
+                "typing_key", token, rooms=[row.room_id for row in rows],
+            )
+        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,
+            )
+
+
 def start(config_options):
     try:
         config = HomeServerConfig.load_config(
@@ -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)

synapse/config/workers.py

@@ -29,6 +29,9 @@ class WorkerConfig(Config):
         self.worker_log_file = config.get("worker_log_file")
         self.worker_log_config = config.get("worker_log_config")
         self.worker_replication_url = config.get("worker_replication_url")
+        self.worker_replication_host = config.get("worker_replication_host", None)
+        self.worker_replication_port = config.get("worker_replication_port", None)
+        self.worker_name = config.get("worker_name", self.worker_app)
 
         if self.worker_listeners:
             for listener in self.worker_listeners:

synapse/server.py

@@ -132,6 +132,7 @@ class HomeServer(object):
         'federation_sender',
         'receipts_handler',
         'macaroon_generator',
+        'tcp_replication',
     ]
 
     def __init__(self, hostname, **kwargs):
@@ -290,6 +291,9 @@ class HomeServer(object):
     def build_receipts_handler(self):
         return ReceiptsHandler(self)
 
+    def build_tcp_replication(self):
+        raise NotImplementedError()
+
     def remove_pusher(self, app_id, push_key, user_id):
         return self.get_pusherpool().remove_pusher(app_id, push_key, user_id)