# # This file is licensed under the Affero General Public License (AGPL) version 3. # # Copyright (C) 2023 New Vector, Ltd # # This program is free software: you can redistribute it and/or modify # it under the terms of the GNU Affero General Public License as # published by the Free Software Foundation, either version 3 of the # License, or (at your option) any later version. # # See the GNU Affero General Public License for more details: # . # # Originally licensed under the Apache License, Version 2.0: # . # # [This file includes modifications made by New Vector Limited] # # """ This module is responsible for getting events from the DB for pagination and event streaming. The order it returns events in depend on whether we are streaming forwards or are paginating backwards. We do this because we want to handle out of order messages nicely, while still returning them in the correct order when we paginate bacwards. This is implemented by keeping two ordering columns: stream_ordering and topological_ordering. Stream ordering is basically insertion/received order (except for events from backfill requests). The topological_ordering is a weak ordering of events based on the pdu graph. This means that we have to have two different types of tokens, depending on what sort order was used: - stream tokens are of the form: "s%d", which maps directly to the column - topological tokems: "t%d-%d", where the integers map to the topological and stream ordering columns respectively. """ import logging from typing import ( TYPE_CHECKING, Any, Collection, Dict, Iterable, List, Optional, Set, Tuple, cast, overload, ) import attr from immutabledict import immutabledict from typing_extensions import Literal from twisted.internet import defer from synapse.api.constants import Direction from synapse.api.filtering import Filter from synapse.events import EventBase from synapse.logging.context import make_deferred_yieldable, run_in_background from synapse.logging.opentracing import trace from synapse.storage._base import SQLBaseStore from synapse.storage.database import ( DatabasePool, LoggingDatabaseConnection, LoggingTransaction, make_in_list_sql_clause, ) from synapse.storage.databases.main.events_worker import EventsWorkerStore from synapse.storage.engines import BaseDatabaseEngine, PostgresEngine, Sqlite3Engine from synapse.storage.util.id_generators import MultiWriterIdGenerator from synapse.types import PersistedEventPosition, RoomStreamToken from synapse.util.caches.descriptors import cached from synapse.util.caches.stream_change_cache import StreamChangeCache from synapse.util.cancellation import cancellable if TYPE_CHECKING: from synapse.server import HomeServer logger = logging.getLogger(__name__) MAX_STREAM_SIZE = 1000 _STREAM_TOKEN = "stream" _TOPOLOGICAL_TOKEN = "topological" # Used as return values for pagination APIs @attr.s(slots=True, frozen=True, auto_attribs=True) class _EventDictReturn: event_id: str topological_ordering: Optional[int] stream_ordering: int @attr.s(slots=True, frozen=True, auto_attribs=True) class _EventsAround: events_before: List[EventBase] events_after: List[EventBase] start: RoomStreamToken end: RoomStreamToken def generate_pagination_where_clause( direction: Direction, column_names: Tuple[str, str], from_token: Optional[Tuple[Optional[int], int]], to_token: Optional[Tuple[Optional[int], int]], engine: BaseDatabaseEngine, ) -> str: """Creates an SQL expression to bound the columns by the pagination tokens. For example creates an SQL expression like: (6, 7) >= (topological_ordering, stream_ordering) AND (5, 3) < (topological_ordering, stream_ordering) would be generated for dir=b, from_token=(6, 7) and to_token=(5, 3). Note that tokens are considered to be after the row they are in, e.g. if a row A has a token T, then we consider A to be before T. This convention is important when figuring out inequalities for the generated SQL, and produces the following result: - If paginating forwards then we exclude any rows matching the from token, but include those that match the to token. - If paginating backwards then we include any rows matching the from token, but include those that match the to token. Args: direction: Whether we're paginating backwards or forwards. column_names: The column names to bound. Must *not* be user defined as these get inserted directly into the SQL statement without escapes. from_token: The start point for the pagination. This is an exclusive minimum bound if direction is forwards, and an inclusive maximum bound if direction is backwards. to_token: The endpoint point for the pagination. This is an inclusive maximum bound if direction is forwards, and an exclusive minimum bound if direction is backwards. engine: The database engine to generate the clauses for Returns: The sql expression """ where_clause = [] if from_token: where_clause.append( _make_generic_sql_bound( bound=">=" if direction == Direction.BACKWARDS else "<", column_names=column_names, values=from_token, engine=engine, ) ) if to_token: where_clause.append( _make_generic_sql_bound( bound="<" if direction == Direction.BACKWARDS else ">=", column_names=column_names, values=to_token, engine=engine, ) ) return " AND ".join(where_clause) def generate_pagination_bounds( direction: Direction, from_token: Optional[RoomStreamToken], to_token: Optional[RoomStreamToken], ) -> Tuple[ str, Optional[Tuple[Optional[int], int]], Optional[Tuple[Optional[int], int]] ]: """ Generate a start and end point for this page of events. Args: direction: Whether pagination is going forwards or backwards. from_token: The token to start pagination at, or None to start at the first value. to_token: The token to end pagination at, or None to not limit the end point. Returns: A three tuple of: ASC or DESC for sorting of the query. The starting position as a tuple of ints representing (topological position, stream position) or None if no from_token was provided. The topological position may be None for live tokens. The end position in the same format as the starting position, or None if no to_token was provided. """ # Tokens really represent positions between elements, but we use # the convention of pointing to the event before the gap. Hence # we have a bit of asymmetry when it comes to equalities. if direction == Direction.BACKWARDS: order = "DESC" else: order = "ASC" # The bounds for the stream tokens are complicated by the fact # that we need to handle the instance_map part of the tokens. We do this # by fetching all events between the min stream token and the maximum # stream token (as returned by `RoomStreamToken.get_max_stream_pos`) and # then filtering the results. from_bound: Optional[Tuple[Optional[int], int]] = None if from_token: if from_token.topological is not None: from_bound = from_token.as_historical_tuple() elif direction == Direction.BACKWARDS: from_bound = ( None, from_token.get_max_stream_pos(), ) else: from_bound = ( None, from_token.stream, ) to_bound: Optional[Tuple[Optional[int], int]] = None if to_token: if to_token.topological is not None: to_bound = to_token.as_historical_tuple() elif direction == Direction.BACKWARDS: to_bound = ( None, to_token.stream, ) else: to_bound = ( None, to_token.get_max_stream_pos(), ) return order, from_bound, to_bound def generate_next_token( direction: Direction, last_topo_ordering: int, last_stream_ordering: int ) -> RoomStreamToken: """ Generate the next room stream token based on the currently returned data. Args: direction: Whether pagination is going forwards or backwards. last_topo_ordering: The last topological ordering being returned. last_stream_ordering: The last stream ordering being returned. Returns: A new RoomStreamToken to return to the client. """ if direction == Direction.BACKWARDS: # Tokens are positions between events. # This token points *after* the last event in the chunk. # We need it to point to the event before it in the chunk # when we are going backwards so we subtract one from the # stream part. last_stream_ordering -= 1 return RoomStreamToken(topological=last_topo_ordering, stream=last_stream_ordering) def _make_generic_sql_bound( bound: str, column_names: Tuple[str, str], values: Tuple[Optional[int], int], engine: BaseDatabaseEngine, ) -> str: """Create an SQL expression that bounds the given column names by the values, e.g. create the equivalent of `(1, 2) < (col1, col2)`. Only works with two columns. Older versions of SQLite don't support that syntax so we have to expand it out manually. Args: bound: The comparison operator to use. One of ">", "<", ">=", "<=", where the values are on the left and columns on the right. names: The column names. Must *not* be user defined as these get inserted directly into the SQL statement without escapes. values: The values to bound the columns by. If the first value is None then only creates a bound on the second column. engine: The database engine to generate the SQL for Returns: The SQL statement """ assert bound in (">", "<", ">=", "<=") name1, name2 = column_names val1, val2 = values if val1 is None: val2 = int(val2) return "(%d %s %s)" % (val2, bound, name2) val1 = int(val1) val2 = int(val2) if isinstance(engine, PostgresEngine): # Postgres doesn't optimise ``(x < a) OR (x=a AND y bool: """Returns True if the event persisted by the given instance at the given topological/stream_ordering falls between the two tokens (taking a None token to mean unbounded). Used to filter results from fetching events in the DB against the given tokens. This is necessary to handle the case where the tokens include position maps, which we handle by fetching more than necessary from the DB and then filtering (rather than attempting to construct a complicated SQL query). """ event_historical_tuple = ( topological_ordering, stream_ordering, ) if lower_token: if lower_token.topological is not None: # If these are historical tokens we compare the `(topological, stream)` # tuples. if event_historical_tuple <= lower_token.as_historical_tuple(): return False else: # If these are live tokens we compare the stream ordering against the # writers stream position. if stream_ordering <= lower_token.get_stream_pos_for_instance( instance_name ): return False if upper_token: if upper_token.topological is not None: if upper_token.as_historical_tuple() < event_historical_tuple: return False else: if upper_token.get_stream_pos_for_instance(instance_name) < stream_ordering: return False return True def filter_to_clause(event_filter: Optional[Filter]) -> Tuple[str, List[str]]: # NB: This may create SQL clauses that don't optimise well (and we don't # have indices on all possible clauses). E.g. it may create # "room_id == X AND room_id != X", which postgres doesn't optimise. if not event_filter: return "", [] clauses = [] args = [] # Handle event types with potential wildcard characters if event_filter.types: type_clauses = [] for typ in event_filter.types: if "*" in typ: type_clauses.append("event.type LIKE ?") typ = typ.replace("*", "%") # Replace * with % for SQL LIKE pattern else: type_clauses.append("event.type = ?") args.append(typ) clauses.append("(%s)" % " OR ".join(type_clauses)) # Handle event types to exclude with potential wildcard characters for typ in event_filter.not_types: if "*" in typ: clauses.append("event.type NOT LIKE ?") typ = typ.replace("*", "%") else: clauses.append("event.type != ?") args.append(typ) if event_filter.senders: clauses.append( "(%s)" % " OR ".join("event.sender = ?" for _ in event_filter.senders) ) args.extend(event_filter.senders) for sender in event_filter.not_senders: clauses.append("event.sender != ?") args.append(sender) if event_filter.rooms: clauses.append( "(%s)" % " OR ".join("event.room_id = ?" for _ in event_filter.rooms) ) args.extend(event_filter.rooms) for room_id in event_filter.not_rooms: clauses.append("event.room_id != ?") args.append(room_id) if event_filter.contains_url: clauses.append("event.contains_url = ?") args.append(event_filter.contains_url) # We're only applying the "labels" filter on the database query, because applying the # "not_labels" filter via a SQL query is non-trivial. Instead, we let # event_filter.check_fields apply it, which is not as efficient but makes the # implementation simpler. if event_filter.labels: clauses.append("(%s)" % " OR ".join("label = ?" for _ in event_filter.labels)) args.extend(event_filter.labels) # Filter on relation_senders / relation types from the joined tables. if event_filter.related_by_senders: clauses.append( "(%s)" % " OR ".join( "related_event.sender = ?" for _ in event_filter.related_by_senders ) ) args.extend(event_filter.related_by_senders) if event_filter.related_by_rel_types: clauses.append( "(%s)" % " OR ".join( "relation_type = ?" for _ in event_filter.related_by_rel_types ) ) args.extend(event_filter.related_by_rel_types) if event_filter.rel_types: clauses.append( "(%s)" % " OR ".join( "event_relation.relation_type = ?" for _ in event_filter.rel_types ) ) args.extend(event_filter.rel_types) if event_filter.not_rel_types: clauses.append( "((%s) OR event_relation.relation_type IS NULL)" % " AND ".join( "event_relation.relation_type != ?" for _ in event_filter.not_rel_types ) ) args.extend(event_filter.not_rel_types) return " AND ".join(clauses), args class StreamWorkerStore(EventsWorkerStore, SQLBaseStore): def __init__( self, database: DatabasePool, db_conn: LoggingDatabaseConnection, hs: "HomeServer", ): super().__init__(database, db_conn, hs) self._instance_name = hs.get_instance_name() self._send_federation = hs.should_send_federation() self._federation_shard_config = hs.config.worker.federation_shard_config # If we're a process that sends federation we may need to reset the # `federation_stream_position` table to match the current sharding # config. We don't do this now as otherwise two processes could conflict # during startup which would cause one to die. self._need_to_reset_federation_stream_positions = self._send_federation events_max = self.get_room_max_stream_ordering() event_cache_prefill, min_event_val = self.db_pool.get_cache_dict( db_conn, "events", entity_column="room_id", stream_column="stream_ordering", max_value=events_max, ) self._events_stream_cache = StreamChangeCache( "EventsRoomStreamChangeCache", min_event_val, prefilled_cache=event_cache_prefill, ) self._membership_stream_cache = StreamChangeCache( "MembershipStreamChangeCache", events_max ) self._stream_order_on_start = self.get_room_max_stream_ordering() self._min_stream_order_on_start = self.get_room_min_stream_ordering() def get_room_max_stream_ordering(self) -> int: """Get the stream_ordering of regular events that we have committed up to Returns the maximum stream id such that all stream ids less than or equal to it have been successfully persisted. """ return self._stream_id_gen.get_current_token() def get_room_min_stream_ordering(self) -> int: """Get the stream_ordering of backfilled events that we have committed up to Backfilled events use *negative* stream orderings, so this returns the minimum negative stream id such that all stream ids greater than or equal to it have been successfully persisted. """ return self._backfill_id_gen.get_current_token() def get_room_max_token(self) -> RoomStreamToken: """Get a `RoomStreamToken` that marks the current maximum persisted position of the events stream. Useful to get a token that represents "now". The token returned is a "live" token that may have an instance_map component. """ min_pos = self._stream_id_gen.get_current_token() positions = {} if isinstance(self._stream_id_gen, MultiWriterIdGenerator): # The `min_pos` is the minimum position that we know all instances # have finished persisting to, so we only care about instances whose # positions are ahead of that. (Instance positions can be behind the # min position as there are times we can work out that the minimum # position is ahead of the naive minimum across all current # positions. See MultiWriterIdGenerator for details) positions = { i: p for i, p in self._stream_id_gen.get_positions().items() if p > min_pos } return RoomStreamToken(stream=min_pos, instance_map=immutabledict(positions)) async def get_room_events_stream_for_rooms( self, room_ids: Collection[str], from_key: RoomStreamToken, to_key: RoomStreamToken, limit: int = 0, order: str = "DESC", ) -> Dict[str, Tuple[List[EventBase], RoomStreamToken]]: """Get new room events in stream ordering since `from_key`. Args: room_ids from_key: Token from which no events are returned before to_key: Token from which no events are returned after. (This is typically the current stream token) limit: Maximum number of events to return order: Either "DESC" or "ASC". Determines which events are returned when the result is limited. If "DESC" then the most recent `limit` events are returned, otherwise returns the oldest `limit` events. Returns: A map from room id to a tuple containing: - list of recent events in the room - stream ordering key for the start of the chunk of events returned. """ room_ids = self._events_stream_cache.get_entities_changed( room_ids, from_key.stream ) if not room_ids: return {} results = {} room_ids = list(room_ids) for rm_ids in (room_ids[i : i + 20] for i in range(0, len(room_ids), 20)): res = await make_deferred_yieldable( defer.gatherResults( [ run_in_background( self.get_room_events_stream_for_room, room_id, from_key, to_key, limit, order=order, ) for room_id in rm_ids ], consumeErrors=True, ) ) results.update(dict(zip(rm_ids, res))) return results def get_rooms_that_changed( self, room_ids: Collection[str], from_key: RoomStreamToken ) -> Set[str]: """Given a list of rooms and a token, return rooms where there may have been changes. """ from_id = from_key.stream return { room_id for room_id in room_ids if self._events_stream_cache.has_entity_changed(room_id, from_id) } async def get_room_events_stream_for_room( self, room_id: str, from_key: RoomStreamToken, to_key: RoomStreamToken, limit: int = 0, order: str = "DESC", ) -> Tuple[List[EventBase], RoomStreamToken]: """Get new room events in stream ordering since `from_key`. Args: room_id from_key: Token from which no events are returned before to_key: Token from which no events are returned after. (This is typically the current stream token) limit: Maximum number of events to return order: Either "DESC" or "ASC". Determines which events are returned when the result is limited. If "DESC" then the most recent `limit` events are returned, otherwise returns the oldest `limit` events. Returns: The list of events (in ascending stream order) and the token from the start of the chunk of events returned. """ if from_key == to_key: return [], from_key has_changed = self._events_stream_cache.has_entity_changed( room_id, from_key.stream ) if not has_changed: return [], from_key def f(txn: LoggingTransaction) -> List[_EventDictReturn]: # To handle tokens with a non-empty instance_map we fetch more # results than necessary and then filter down min_from_id = from_key.stream max_to_id = to_key.get_max_stream_pos() sql = """ SELECT event_id, instance_name, topological_ordering, stream_ordering FROM events WHERE room_id = ? AND not outlier AND stream_ordering > ? AND stream_ordering <= ? ORDER BY stream_ordering %s LIMIT ? """ % ( order, ) txn.execute(sql, (room_id, min_from_id, max_to_id, 2 * limit)) rows = [ _EventDictReturn(event_id, None, stream_ordering) for event_id, instance_name, topological_ordering, stream_ordering in txn if _filter_results( from_key, to_key, instance_name, topological_ordering, stream_ordering, ) ][:limit] return rows rows = await self.db_pool.runInteraction("get_room_events_stream_for_room", f) ret = await self.get_events_as_list( [r.event_id for r in rows], get_prev_content=True ) if order.lower() == "desc": ret.reverse() if rows: key = RoomStreamToken(stream=min(r.stream_ordering for r in rows)) else: # Assume we didn't get anything because there was nothing to # get. key = from_key return ret, key @cancellable async def get_membership_changes_for_user( self, user_id: str, from_key: RoomStreamToken, to_key: RoomStreamToken, excluded_rooms: Optional[List[str]] = None, ) -> List[EventBase]: """Fetch membership events for a given user. All such events whose stream ordering `s` lies in the range `from_key < s <= to_key` are returned. Events are ordered by ascending stream order. """ # Start by ruling out cases where a DB query is not necessary. if from_key == to_key: return [] if from_key: has_changed = self._membership_stream_cache.has_entity_changed( user_id, int(from_key.stream) ) if not has_changed: return [] def f(txn: LoggingTransaction) -> List[_EventDictReturn]: # To handle tokens with a non-empty instance_map we fetch more # results than necessary and then filter down min_from_id = from_key.stream max_to_id = to_key.get_max_stream_pos() args: List[Any] = [user_id, min_from_id, max_to_id] ignore_room_clause = "" if excluded_rooms is not None and len(excluded_rooms) > 0: ignore_room_clause = "AND e.room_id NOT IN (%s)" % ",".join( "?" for _ in excluded_rooms ) args = args + excluded_rooms sql = """ SELECT m.event_id, instance_name, topological_ordering, stream_ordering FROM events AS e, room_memberships AS m WHERE e.event_id = m.event_id AND m.user_id = ? AND e.stream_ordering > ? AND e.stream_ordering <= ? %s ORDER BY e.stream_ordering ASC """ % ( ignore_room_clause, ) txn.execute(sql, args) rows = [ _EventDictReturn(event_id, None, stream_ordering) for event_id, instance_name, topological_ordering, stream_ordering in txn if _filter_results( from_key, to_key, instance_name, topological_ordering, stream_ordering, ) ] return rows rows = await self.db_pool.runInteraction("get_membership_changes_for_user", f) ret = await self.get_events_as_list( [r.event_id for r in rows], get_prev_content=True ) return ret async def get_recent_events_for_room( self, room_id: str, limit: int, end_token: RoomStreamToken ) -> Tuple[List[EventBase], RoomStreamToken]: """Get the most recent events in the room in topological ordering. Args: room_id limit end_token: The stream token representing now. Returns: A list of events and a token pointing to the start of the returned events. The events returned are in ascending topological order. """ rows, token = await self.get_recent_event_ids_for_room( room_id, limit, end_token ) events = await self.get_events_as_list( [r.event_id for r in rows], get_prev_content=True ) return events, token async def get_recent_event_ids_for_room( self, room_id: str, limit: int, end_token: RoomStreamToken ) -> Tuple[List[_EventDictReturn], RoomStreamToken]: """Get the most recent events in the room in topological ordering. Args: room_id limit end_token: The stream token representing now. Returns: A list of _EventDictReturn and a token pointing to the start of the returned events. The events returned are in ascending order. """ # Allow a zero limit here, and no-op. if limit == 0: return [], end_token rows, token = await self.db_pool.runInteraction( "get_recent_event_ids_for_room", self._paginate_room_events_txn, room_id, from_token=end_token, limit=limit, ) # We want to return the results in ascending order. rows.reverse() return rows, token async def get_room_event_before_stream_ordering( self, room_id: str, stream_ordering: int ) -> Optional[Tuple[int, int, str]]: """Gets details of the first event in a room at or before a stream ordering Args: room_id: stream_ordering: Returns: A tuple of (stream ordering, topological ordering, event_id) """ def _f(txn: LoggingTransaction) -> Optional[Tuple[int, int, str]]: sql = """ SELECT stream_ordering, topological_ordering, event_id FROM events LEFT JOIN rejections USING (event_id) WHERE room_id = ? AND stream_ordering <= ? AND NOT outlier AND rejections.event_id IS NULL ORDER BY stream_ordering DESC LIMIT 1 """ txn.execute(sql, (room_id, stream_ordering)) return cast(Optional[Tuple[int, int, str]], txn.fetchone()) return await self.db_pool.runInteraction( "get_room_event_before_stream_ordering", _f ) async def get_last_event_in_room_before_stream_ordering( self, room_id: str, end_token: RoomStreamToken, ) -> Optional[str]: """Returns the ID of the last event in a room at or before a stream ordering Args: room_id end_token: The token used to stream from Returns: The ID of the most recent event, or None if there are no events in the room before this stream ordering. """ def get_last_event_in_room_before_stream_ordering_txn( txn: LoggingTransaction, ) -> Optional[str]: # We need to handle the fact that the stream tokens can be vector # clocks. We do this by getting all rows between the minimum and # maximum stream ordering in the token, plus one row less than the # minimum stream ordering. We then filter the results against the # token and return the first row that matches. sql = """ SELECT * FROM ( SELECT instance_name, stream_ordering, topological_ordering, event_id FROM events LEFT JOIN rejections USING (event_id) WHERE room_id = ? AND ? < stream_ordering AND stream_ordering <= ? AND NOT outlier AND rejections.event_id IS NULL ORDER BY stream_ordering DESC ) AS a UNION SELECT * FROM ( SELECT instance_name, stream_ordering, topological_ordering, event_id FROM events LEFT JOIN rejections USING (event_id) WHERE room_id = ? AND stream_ordering <= ? AND NOT outlier AND rejections.event_id IS NULL ORDER BY stream_ordering DESC LIMIT 1 ) AS b """ txn.execute( sql, ( room_id, end_token.stream, end_token.get_max_stream_pos(), room_id, end_token.stream, ), ) for instance_name, stream_ordering, topological_ordering, event_id in txn: if _filter_results( lower_token=None, upper_token=end_token, instance_name=instance_name, topological_ordering=topological_ordering, stream_ordering=stream_ordering, ): return event_id return None return await self.db_pool.runInteraction( "get_last_event_in_room_before_stream_ordering", get_last_event_in_room_before_stream_ordering_txn, ) async def get_current_room_stream_token_for_room_id( self, room_id: str ) -> RoomStreamToken: """Returns the current position of the rooms stream (historic token).""" stream_ordering = self.get_room_max_stream_ordering() topo = await self.db_pool.runInteraction( "_get_max_topological_txn", self._get_max_topological_txn, room_id ) return RoomStreamToken(topological=topo, stream=stream_ordering) @overload def get_stream_id_for_event_txn( self, txn: LoggingTransaction, event_id: str, allow_none: Literal[False] = False, ) -> int: ... @overload def get_stream_id_for_event_txn( self, txn: LoggingTransaction, event_id: str, allow_none: bool = False, ) -> Optional[int]: ... def get_stream_id_for_event_txn( self, txn: LoggingTransaction, event_id: str, allow_none: bool = False, ) -> Optional[int]: # Type ignore: we pass keyvalues a Dict[str, str]; the function wants # Dict[str, Any]. I think mypy is unhappy because Dict is invariant? return self.db_pool.simple_select_one_onecol_txn( # type: ignore[call-overload] txn=txn, table="events", keyvalues={"event_id": event_id}, retcol="stream_ordering", allow_none=allow_none, ) async def get_position_for_event(self, event_id: str) -> PersistedEventPosition: """Get the persisted position for an event""" row = await self.db_pool.simple_select_one( table="events", keyvalues={"event_id": event_id}, retcols=("stream_ordering", "instance_name"), desc="get_position_for_event", ) return PersistedEventPosition(row[1] or "master", row[0]) async def get_topological_token_for_event(self, event_id: str) -> RoomStreamToken: """The stream token for an event Args: event_id: The id of the event to look up a stream token for. Raises: StoreError if the event wasn't in the database. Returns: A `RoomStreamToken` topological token. """ row = await self.db_pool.simple_select_one( table="events", keyvalues={"event_id": event_id}, retcols=("stream_ordering", "topological_ordering"), desc="get_topological_token_for_event", ) return RoomStreamToken(topological=row[1], stream=row[0]) async def get_current_topological_token(self, room_id: str, stream_key: int) -> int: """Gets the topological token in a room after or at the given stream ordering. Args: room_id stream_key """ if isinstance(self.database_engine, PostgresEngine): min_function = "LEAST" elif isinstance(self.database_engine, Sqlite3Engine): min_function = "MIN" else: raise RuntimeError(f"Unknown database engine {self.database_engine}") # This query used to be # SELECT COALESCE(MIN(topological_ordering), 0) FROM events # WHERE room_id = ? and events.stream_ordering >= {stream_key} # which returns 0 if the stream_key is newer than any event in # the room. That's not wrong, but it seems to interact oddly with backfill, # requiring a second call to /messages to actually backfill from a remote # homeserver. # # Instead, rollback the stream ordering to that after the most recent event in # this room. sql = f""" WITH fallback(max_stream_ordering) AS ( SELECT MAX(stream_ordering) FROM events WHERE room_id = ? ) SELECT COALESCE(MIN(topological_ordering), 0) FROM events WHERE room_id = ? AND events.stream_ordering >= {min_function}( ?, (SELECT max_stream_ordering FROM fallback) ) """ row = await self.db_pool.execute( "get_current_topological_token", sql, room_id, room_id, stream_key ) return row[0][0] if row else 0 def _get_max_topological_txn(self, txn: LoggingTransaction, room_id: str) -> int: txn.execute( "SELECT MAX(topological_ordering) FROM events WHERE room_id = ?", (room_id,), ) rows = txn.fetchall() # An aggregate function like MAX() will always return one row per group # so we can safely rely on the lookup here. For example, when a we # lookup a `room_id` which does not exist, `rows` will look like # `[(None,)]` return rows[0][0] if rows[0][0] is not None else 0 async def get_events_around( self, room_id: str, event_id: str, before_limit: int, after_limit: int, event_filter: Optional[Filter] = None, ) -> _EventsAround: """Retrieve events and pagination tokens around a given event in a room. """ results = await self.db_pool.runInteraction( "get_events_around", self._get_events_around_txn, room_id, event_id, before_limit, after_limit, event_filter, ) events_before = await self.get_events_as_list( list(results["before"]["event_ids"]), get_prev_content=True ) events_after = await self.get_events_as_list( list(results["after"]["event_ids"]), get_prev_content=True ) return _EventsAround( events_before=events_before, events_after=events_after, start=results["before"]["token"], end=results["after"]["token"], ) def _get_events_around_txn( self, txn: LoggingTransaction, room_id: str, event_id: str, before_limit: int, after_limit: int, event_filter: Optional[Filter], ) -> dict: """Retrieves event_ids and pagination tokens around a given event in a room. Args: room_id event_id before_limit after_limit event_filter Returns: dict """ stream_ordering, topological_ordering = cast( Tuple[int, int], self.db_pool.simple_select_one_txn( txn, "events", keyvalues={"event_id": event_id, "room_id": room_id}, retcols=["stream_ordering", "topological_ordering"], ), ) # Paginating backwards includes the event at the token, but paginating # forward doesn't. before_token = RoomStreamToken( topological=topological_ordering - 1, stream=stream_ordering ) after_token = RoomStreamToken( topological=topological_ordering, stream=stream_ordering ) rows, start_token = self._paginate_room_events_txn( txn, room_id, before_token, direction=Direction.BACKWARDS, limit=before_limit, event_filter=event_filter, ) events_before = [r.event_id for r in rows] rows, end_token = self._paginate_room_events_txn( txn, room_id, after_token, direction=Direction.FORWARDS, limit=after_limit, event_filter=event_filter, ) events_after = [r.event_id for r in rows] return { "before": {"event_ids": events_before, "token": start_token}, "after": {"event_ids": events_after, "token": end_token}, } async def get_all_new_event_ids_stream( self, from_id: int, current_id: int, limit: int, ) -> Tuple[int, Dict[str, Optional[int]]]: """Get all new events Returns all event ids with from_id < stream_ordering <= current_id. Args: from_id: the stream_ordering of the last event we processed current_id: the stream_ordering of the most recently processed event limit: the maximum number of events to return Returns: A tuple of (next_id, event_to_received_ts), where `next_id` is the next value to pass as `from_id` (it will either be the stream_ordering of the last returned event, or, if fewer than `limit` events were found, the `current_id`). The `event_to_received_ts` is a dictionary mapping event ID to the event `received_ts`, sorted by ascending stream_ordering. """ def get_all_new_event_ids_stream_txn( txn: LoggingTransaction, ) -> Tuple[int, Dict[str, Optional[int]]]: sql = ( "SELECT e.stream_ordering, e.event_id, e.received_ts" " FROM events AS e" " WHERE" " ? < e.stream_ordering AND e.stream_ordering <= ?" " ORDER BY e.stream_ordering ASC" " LIMIT ?" ) txn.execute(sql, (from_id, current_id, limit)) rows = txn.fetchall() upper_bound = current_id if len(rows) == limit: upper_bound = rows[-1][0] event_to_received_ts: Dict[str, Optional[int]] = { row[1]: row[2] for row in rows } return upper_bound, event_to_received_ts upper_bound, event_to_received_ts = await self.db_pool.runInteraction( "get_all_new_event_ids_stream", get_all_new_event_ids_stream_txn ) return upper_bound, event_to_received_ts async def get_federation_out_pos(self, typ: str) -> int: if self._need_to_reset_federation_stream_positions: await self.db_pool.runInteraction( "_reset_federation_positions_txn", self._reset_federation_positions_txn ) self._need_to_reset_federation_stream_positions = False return await self.db_pool.simple_select_one_onecol( table="federation_stream_position", retcol="stream_id", keyvalues={"type": typ, "instance_name": self._instance_name}, desc="get_federation_out_pos", ) async def update_federation_out_pos(self, typ: str, stream_id: int) -> None: if self._need_to_reset_federation_stream_positions: await self.db_pool.runInteraction( "_reset_federation_positions_txn", self._reset_federation_positions_txn ) self._need_to_reset_federation_stream_positions = False await self.db_pool.simple_update_one( table="federation_stream_position", keyvalues={"type": typ, "instance_name": self._instance_name}, updatevalues={"stream_id": stream_id}, desc="update_federation_out_pos", ) def _reset_federation_positions_txn(self, txn: LoggingTransaction) -> None: """Fiddles with the `federation_stream_position` table to make it match the configured federation sender instances during start up. """ # The federation sender instances may have changed, so we need to # massage the `federation_stream_position` table to have a row per type # per instance sending federation. If there is a mismatch we update the # table with the correct rows using the *minimum* stream ID seen. This # may result in resending of events/EDUs to remote servers, but that is # preferable to dropping them. if not self._send_federation: return # Pull out the configured instances. If we don't have a shard config then # we assume that we're the only instance sending. configured_instances = self._federation_shard_config.instances if not configured_instances: configured_instances = [self._instance_name] elif self._instance_name not in configured_instances: return instances_in_table = self.db_pool.simple_select_onecol_txn( txn, table="federation_stream_position", keyvalues={}, retcol="instance_name", ) if set(instances_in_table) == set(configured_instances): # Nothing to do return sql = """ SELECT type, MIN(stream_id) FROM federation_stream_position GROUP BY type """ txn.execute(sql) min_positions = dict( cast(Iterable[Tuple[str, int]], txn) ) # Map from type -> min position # Ensure we do actually have some values here assert set(min_positions) == {"federation", "events"} sql = """ DELETE FROM federation_stream_position WHERE NOT (%s) """ clause, args = make_in_list_sql_clause( txn.database_engine, "instance_name", configured_instances ) txn.execute(sql % (clause,), args) for typ, stream_id in min_positions.items(): self.db_pool.simple_upsert_txn( txn, table="federation_stream_position", keyvalues={"type": typ, "instance_name": self._instance_name}, values={"stream_id": stream_id}, ) def has_room_changed_since(self, room_id: str, stream_id: int) -> bool: return self._events_stream_cache.has_entity_changed(room_id, stream_id) def _paginate_room_events_txn( self, txn: LoggingTransaction, room_id: str, from_token: RoomStreamToken, to_token: Optional[RoomStreamToken] = None, direction: Direction = Direction.BACKWARDS, limit: int = -1, event_filter: Optional[Filter] = None, ) -> Tuple[List[_EventDictReturn], RoomStreamToken]: """Returns list of events before or after a given token. Args: txn room_id from_token: The token used to stream from to_token: A token which if given limits the results to only those before direction: Indicates whether we are paginating forwards or backwards from `from_key`. limit: The maximum number of events to return. event_filter: If provided filters the events to those that match the filter. Returns: A list of _EventDictReturn and a token that points to the end of the result set. If no events are returned then the end of the stream has been reached (i.e. there are no events between `from_token` and `to_token`), or `limit` is zero. """ args: List[Any] = [room_id] order, from_bound, to_bound = generate_pagination_bounds( direction, from_token, to_token ) bounds = generate_pagination_where_clause( direction=direction, column_names=("event.topological_ordering", "event.stream_ordering"), from_token=from_bound, to_token=to_bound, engine=self.database_engine, ) filter_clause, filter_args = filter_to_clause(event_filter) if filter_clause: bounds += " AND " + filter_clause args.extend(filter_args) # We fetch more events as we'll filter the result set args.append(int(limit) * 2) select_keywords = "SELECT" join_clause = "" # Using DISTINCT in this SELECT query is quite expensive, because it # requires the engine to sort on the entire (not limited) result set, # i.e. the entire events table. Only use it in scenarios that could result # in the same event ID occurring multiple times in the results. needs_distinct = False if event_filter and event_filter.labels: # If we're not filtering on a label, then joining on event_labels will # return as many row for a single event as the number of labels it has. To # avoid this, only join if we're filtering on at least one label. join_clause += """ LEFT JOIN event_labels USING (event_id, room_id, topological_ordering) """ if len(event_filter.labels) > 1: # Multiple labels could cause the same event to appear multiple times. needs_distinct = True # If there is a relation_senders and relation_types filter join to the # relations table to get events related to the current event. if event_filter and ( event_filter.related_by_senders or event_filter.related_by_rel_types ): # Filtering by relations could cause the same event to appear multiple # times (since there's no limit on the number of relations to an event). needs_distinct = True join_clause += """ LEFT JOIN event_relations AS relation ON (event.event_id = relation.relates_to_id) """ if event_filter.related_by_senders: join_clause += """ LEFT JOIN events AS related_event ON (relation.event_id = related_event.event_id) """ # If there is a not_rel_types filter join to the relations table to get # the event's relation information. if event_filter and (event_filter.rel_types or event_filter.not_rel_types): join_clause += """ LEFT JOIN event_relations AS event_relation USING (event_id) """ if needs_distinct: select_keywords += " DISTINCT" sql = """ %(select_keywords)s event.event_id, event.instance_name, event.topological_ordering, event.stream_ordering FROM events AS event %(join_clause)s WHERE event.outlier = FALSE AND event.room_id = ? AND %(bounds)s ORDER BY event.topological_ordering %(order)s, event.stream_ordering %(order)s LIMIT ? """ % { "select_keywords": select_keywords, "join_clause": join_clause, "bounds": bounds, "order": order, } txn.execute(sql, args) # Filter the result set. rows = [ _EventDictReturn(event_id, topological_ordering, stream_ordering) for event_id, instance_name, topological_ordering, stream_ordering in txn if _filter_results( lower_token=to_token if direction == Direction.BACKWARDS else from_token, upper_token=from_token if direction == Direction.BACKWARDS else to_token, instance_name=instance_name, topological_ordering=topological_ordering, stream_ordering=stream_ordering, ) ][:limit] if rows: assert rows[-1].topological_ordering is not None next_token = generate_next_token( direction, rows[-1].topological_ordering, rows[-1].stream_ordering ) else: # TODO (erikj): We should work out what to do here instead. next_token = to_token if to_token else from_token return rows, next_token @trace async def paginate_room_events( self, room_id: str, from_key: RoomStreamToken, to_key: Optional[RoomStreamToken] = None, direction: Direction = Direction.BACKWARDS, limit: int = -1, event_filter: Optional[Filter] = None, ) -> Tuple[List[EventBase], RoomStreamToken]: """Returns list of events before or after a given token. Args: room_id from_key: The token used to stream from to_key: A token which if given limits the results to only those before direction: Indicates whether we are paginating forwards or backwards from `from_key`. limit: The maximum number of events to return. event_filter: If provided filters the events to those that match the filter. Returns: The results as a list of events and a token that points to the end of the result set. If no events are returned then the end of the stream has been reached (i.e. there are no events between `from_key` and `to_key`). """ rows, token = await self.db_pool.runInteraction( "paginate_room_events", self._paginate_room_events_txn, room_id, from_key, to_key, direction, limit, event_filter, ) events = await self.get_events_as_list( [r.event_id for r in rows], get_prev_content=True ) return events, token @cached() async def get_id_for_instance(self, instance_name: str) -> int: """Get a unique, immutable ID that corresponds to the given Synapse worker instance.""" def _get_id_for_instance_txn(txn: LoggingTransaction) -> int: instance_id = self.db_pool.simple_select_one_onecol_txn( txn, table="instance_map", keyvalues={"instance_name": instance_name}, retcol="instance_id", allow_none=True, ) if instance_id is not None: return instance_id # If we don't have an entry upsert one. # # We could do this before the first check, and rely on the cache for # efficiency, but each UPSERT causes the next ID to increment which # can quickly bloat the size of the generated IDs for new instances. self.db_pool.simple_upsert_txn( txn, table="instance_map", keyvalues={"instance_name": instance_name}, values={}, ) return self.db_pool.simple_select_one_onecol_txn( txn, table="instance_map", keyvalues={"instance_name": instance_name}, retcol="instance_id", ) return await self.db_pool.runInteraction( "get_id_for_instance", _get_id_for_instance_txn ) @cached() async def get_name_from_instance_id(self, instance_id: int) -> str: """Get the instance name from an ID previously returned by `get_id_for_instance`. """ return await self.db_pool.simple_select_one_onecol( table="instance_map", keyvalues={"instance_id": instance_id}, retcol="instance_name", desc="get_name_from_instance_id", ) async def get_timeline_gaps( self, room_id: str, from_token: Optional[RoomStreamToken], to_token: RoomStreamToken, ) -> Optional[RoomStreamToken]: """Check if there is a gap, and return a token that marks the position of the gap in the stream. """ sql = """ SELECT instance_name, stream_ordering FROM timeline_gaps WHERE room_id = ? AND ? < stream_ordering AND stream_ordering <= ? ORDER BY stream_ordering """ rows = await self.db_pool.execute( "get_timeline_gaps", sql, room_id, from_token.stream if from_token else 0, to_token.get_max_stream_pos(), ) if not rows: return None positions = [ PersistedEventPosition(instance_name, stream_ordering) for instance_name, stream_ordering in rows ] if from_token: positions = [p for p in positions if p.persisted_after(from_token)] positions = [p for p in positions if not p.persisted_after(to_token)] if positions: # We return a stream token that ensures the event *at* the position # of the gap is included (as the gap is *before* the persisted # event). last_position = positions[-1] return RoomStreamToken(stream=last_position.stream - 1) return None