summary refs log tree commit diff
path: root/synapse/replication/tcp/resource.py
blob: 33d2f589ac362247cfdf6423e2b49b5722b53b64 (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
# -*- coding: utf-8 -*-
# Copyright 2017 Vector Creations Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""The server side of the replication stream.
"""

import logging
import random
from typing import Dict, List

from prometheus_client import Counter

from twisted.internet.protocol import Factory

from synapse.metrics.background_process_metrics import run_as_background_process
from synapse.replication.tcp.protocol import ServerReplicationStreamProtocol
from synapse.replication.tcp.streams import STREAMS_MAP, FederationStream, Stream
from synapse.util.metrics import Measure

stream_updates_counter = Counter(
    "synapse_replication_tcp_resource_stream_updates", "", ["stream_name"]
)

logger = logging.getLogger(__name__)


class ReplicationStreamProtocolFactory(Factory):
    """Factory for new replication connections.
    """

    def __init__(self, hs):
        self.command_handler = hs.get_tcp_replication()
        self.clock = hs.get_clock()
        self.server_name = hs.config.server_name

        # If we've created a `ReplicationStreamProtocolFactory` then we're
        # almost certainly registering a replication listener, so let's ensure
        # that we've started a `ReplicationStreamer` instance to actually push
        # data.
        #
        # (This is a bit of a weird place to do this, but the alternatives such
        # as putting this in `HomeServer.setup()`, requires either passing the
        # listener config again or always starting a `ReplicationStreamer`.)
        hs.get_replication_streamer()

    def buildProtocol(self, addr):
        return ServerReplicationStreamProtocol(
            self.server_name, self.clock, self.command_handler
        )


class ReplicationStreamer(object):
    """Handles replication connections.

    This needs to be poked when new replication data may be available. When new
    data is available it will propagate to all connected clients.
    """

    def __init__(self, hs):
        self.store = hs.get_datastore()
        self.clock = hs.get_clock()
        self.notifier = hs.get_notifier()

        self._replication_torture_level = hs.config.replication_torture_level

        # Work out list of streams that this instance is the source of.
        self.streams = []  # type: List[Stream]
        if hs.config.worker_app is None:
            for stream in STREAMS_MAP.values():
                if stream == FederationStream and hs.config.send_federation:
                    # We only support federation stream if federation sending
                    # hase been disabled on the master.
                    continue

                self.streams.append(stream(hs))

        self.streams_by_name = {stream.NAME: stream for stream in self.streams}

        # Only bother registering the notifier callback if we have streams to
        # publish.
        if self.streams:
            self.notifier.add_replication_callback(self.on_notifier_poke)

        # Keeps track of whether we are currently checking for updates
        self.is_looping = False
        self.pending_updates = False

        self.command_handler = hs.get_tcp_replication()

    def get_streams(self) -> Dict[str, Stream]:
        """Get a mapp from stream name to stream instance.
        """
        return self.streams_by_name

    def on_notifier_poke(self):
        """Checks if there is actually any new data and sends it to the
        connections if there are.

        This should get called each time new data is available, even if it
        is currently being executed, so that nothing gets missed
        """
        if not self.command_handler.connected():
            # Don't bother if nothing is listening. We still need to advance
            # the stream tokens otherwise they'll fall beihind forever
            for stream in self.streams:
                stream.discard_updates_and_advance()
            return

        self.pending_updates = True

        if self.is_looping:
            logger.debug("Notifier poke loop already running")
            return

        run_as_background_process("replication_notifier", self._run_notifier_loop)

    async def _run_notifier_loop(self):
        self.is_looping = True

        try:
            # Keep looping while there have been pokes about potential updates.
            # This protects against the race where a stream we already checked
            # gets an update while we're handling other streams.
            while self.pending_updates:
                self.pending_updates = False

                with Measure(self.clock, "repl.stream.get_updates"):
                    all_streams = self.streams

                    if self._replication_torture_level is not None:
                        # there is no guarantee about ordering between the streams,
                        # so let's shuffle them around a bit when we are in torture mode.
                        all_streams = list(all_streams)
                        random.shuffle(all_streams)

                    for stream in all_streams:
                        if stream.last_token == stream.current_token():
                            continue

                        if self._replication_torture_level:
                            await self.clock.sleep(
                                self._replication_torture_level / 1000.0
                            )

                        logger.debug(
                            "Getting stream: %s: %s -> %s",
                            stream.NAME,
                            stream.last_token,
                            stream.current_token(),
                        )
                        try:
                            updates, current_token, limited = await stream.get_updates()
                            self.pending_updates |= limited
                        except Exception:
                            logger.info("Failed to handle stream %s", stream.NAME)
                            raise

                        logger.debug(
                            "Sending %d updates", len(updates),
                        )

                        if updates:
                            logger.info(
                                "Streaming: %s -> %s", stream.NAME, updates[-1][0]
                            )
                            stream_updates_counter.labels(stream.NAME).inc(len(updates))

                        # Some streams return multiple rows with the same stream IDs,
                        # we need to make sure they get sent out in batches. We do
                        # this by setting the current token to all but the last of
                        # a series of updates with the same token to have a None
                        # token. See RdataCommand for more details.
                        batched_updates = _batch_updates(updates)

                        for token, row in batched_updates:
                            try:
                                self.command_handler.stream_update(
                                    stream.NAME, token, row
                                )
                            except Exception:
                                logger.exception("Failed to replicate")

            logger.debug("No more pending updates, breaking poke loop")
        finally:
            self.pending_updates = False
            self.is_looping = False


def _batch_updates(updates):
    """Takes a list of updates of form [(token, row)] and sets the token to
    None for all rows where the next row has the same token. This is used to
    implement batching.

    For example:

        [(1, _), (1, _), (2, _), (3, _), (3, _)]

    becomes:

        [(None, _), (1, _), (2, _), (None, _), (3, _)]
    """
    if not updates:
        return []

    new_updates = []
    for i, update in enumerate(updates[:-1]):
        if update[0] == updates[i + 1][0]:
            new_updates.append((None, update[1]))
        else:
            new_updates.append(update)

    new_updates.append(updates[-1])
    return new_updates