diff options
Diffstat (limited to 'synapse/storage/stream.py')
| -rw-r--r-- | synapse/storage/stream.py | 204 |
1 files changed, 150 insertions, 54 deletions
diff --git a/synapse/storage/stream.py b/synapse/storage/stream.py index b9ad965fd6..862c5c3ea1 100644 --- a/synapse/storage/stream.py +++ b/synapse/storage/stream.py @@ -40,6 +40,7 @@ from synapse.util.caches.descriptors import cached from synapse.api.constants import EventTypes from synapse.types import RoomStreamToken from synapse.util.logcontext import preserve_fn +from synapse.storage.engines import PostgresEngine, Sqlite3Engine import logging @@ -54,26 +55,92 @@ _STREAM_TOKEN = "stream" _TOPOLOGICAL_TOKEN = "topological" -def lower_bound(token): +def lower_bound(token, engine, inclusive=False): + inclusive = "=" if inclusive else "" if token.topological is None: - return "(%d < %s)" % (token.stream, "stream_ordering") + return "(%d <%s %s)" % (token.stream, inclusive, "stream_ordering") else: - return "(%d < %s OR (%d = %s AND %d < %s))" % ( + if isinstance(engine, PostgresEngine): + # Postgres doesn't optimise ``(x < a) OR (x=a AND y<b)`` as well + # as it optimises ``(x,y) < (a,b)`` on multicolumn indexes. So we + # use the later form when running against postgres. + return "((%d,%d) <%s (%s,%s))" % ( + token.topological, token.stream, inclusive, + "topological_ordering", "stream_ordering", + ) + return "(%d < %s OR (%d = %s AND %d <%s %s))" % ( token.topological, "topological_ordering", token.topological, "topological_ordering", - token.stream, "stream_ordering", + token.stream, inclusive, "stream_ordering", ) -def upper_bound(token): +def upper_bound(token, engine, inclusive=True): + inclusive = "=" if inclusive else "" if token.topological is None: - return "(%d >= %s)" % (token.stream, "stream_ordering") + return "(%d >%s %s)" % (token.stream, inclusive, "stream_ordering") else: - return "(%d > %s OR (%d = %s AND %d >= %s))" % ( + if isinstance(engine, PostgresEngine): + # Postgres doesn't optimise ``(x > a) OR (x=a AND y>b)`` as well + # as it optimises ``(x,y) > (a,b)`` on multicolumn indexes. So we + # use the later form when running against postgres. + return "((%d,%d) >%s (%s,%s))" % ( + token.topological, token.stream, inclusive, + "topological_ordering", "stream_ordering", + ) + return "(%d > %s OR (%d = %s AND %d >%s %s))" % ( token.topological, "topological_ordering", token.topological, "topological_ordering", - token.stream, "stream_ordering", + token.stream, inclusive, "stream_ordering", + ) + + +def filter_to_clause(event_filter): + # 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 = [] + + if event_filter.types: + clauses.append( + "(%s)" % " OR ".join("type = ?" for _ in event_filter.types) + ) + args.extend(event_filter.types) + + for typ in event_filter.not_types: + clauses.append("type != ?") + args.append(typ) + + if event_filter.senders: + clauses.append( + "(%s)" % " OR ".join("sender = ?" for _ in event_filter.senders) ) + args.extend(event_filter.senders) + + for sender in event_filter.not_senders: + clauses.append("sender != ?") + args.append(sender) + + if event_filter.rooms: + clauses.append( + "(%s)" % " OR ".join("room_id = ?" for _ in event_filter.rooms) + ) + args.extend(event_filter.rooms) + + for room_id in event_filter.not_rooms: + clauses.append("room_id != ?") + args.append(room_id) + + if event_filter.contains_url: + clauses.append("contains_url = ?") + args.append(event_filter.contains_url) + + return " AND ".join(clauses), args class StreamStore(SQLBaseStore): @@ -301,25 +368,35 @@ class StreamStore(SQLBaseStore): @defer.inlineCallbacks def paginate_room_events(self, room_id, from_key, to_key=None, - direction='b', limit=-1): + direction='b', limit=-1, event_filter=None): # 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. args = [False, room_id] if direction == 'b': order = "DESC" - bounds = upper_bound(RoomStreamToken.parse(from_key)) + bounds = upper_bound( + RoomStreamToken.parse(from_key), self.database_engine + ) if to_key: - bounds = "%s AND %s" % ( - bounds, lower_bound(RoomStreamToken.parse(to_key)) - ) + bounds = "%s AND %s" % (bounds, lower_bound( + RoomStreamToken.parse(to_key), self.database_engine + )) else: order = "ASC" - bounds = lower_bound(RoomStreamToken.parse(from_key)) + bounds = lower_bound( + RoomStreamToken.parse(from_key), self.database_engine + ) if to_key: - bounds = "%s AND %s" % ( - bounds, upper_bound(RoomStreamToken.parse(to_key)) - ) + bounds = "%s AND %s" % (bounds, upper_bound( + RoomStreamToken.parse(to_key), self.database_engine + )) + + filter_clause, filter_args = filter_to_clause(event_filter) + + if filter_clause: + bounds += " AND " + filter_clause + args.extend(filter_args) if int(limit) > 0: args.append(int(limit)) @@ -487,13 +564,13 @@ class StreamStore(SQLBaseStore): row["topological_ordering"], row["stream_ordering"],) ) - def get_max_topological_token_for_stream_and_room(self, room_id, stream_key): + def get_max_topological_token(self, room_id, stream_key): sql = ( "SELECT max(topological_ordering) FROM events" " WHERE room_id = ? AND stream_ordering < ?" ) return self._execute( - "get_max_topological_token_for_stream_and_room", None, + "get_max_topological_token", None, sql, room_id, stream_key, ).addCallback( lambda r: r[0][0] if r else 0 @@ -586,32 +663,60 @@ class StreamStore(SQLBaseStore): retcols=["stream_ordering", "topological_ordering"], ) - stream_ordering = results["stream_ordering"] - topological_ordering = results["topological_ordering"] - - query_before = ( - "SELECT topological_ordering, stream_ordering, event_id FROM events" - " WHERE room_id = ? AND (topological_ordering < ?" - " OR (topological_ordering = ? AND stream_ordering < ?))" - " ORDER BY topological_ordering DESC, stream_ordering DESC" - " LIMIT ?" + token = RoomStreamToken( + results["topological_ordering"], + results["stream_ordering"], ) - query_after = ( - "SELECT topological_ordering, stream_ordering, event_id FROM events" - " WHERE room_id = ? AND (topological_ordering > ?" - " OR (topological_ordering = ? AND stream_ordering > ?))" - " ORDER BY topological_ordering ASC, stream_ordering ASC" - " LIMIT ?" - ) + if isinstance(self.database_engine, Sqlite3Engine): + # SQLite3 doesn't optimise ``(x < a) OR (x = a AND y < b)`` + # So we give pass it to SQLite3 as the UNION ALL of the two queries. + + query_before = ( + "SELECT topological_ordering, stream_ordering, event_id FROM events" + " WHERE room_id = ? AND topological_ordering < ?" + " UNION ALL" + " SELECT topological_ordering, stream_ordering, event_id FROM events" + " WHERE room_id = ? AND topological_ordering = ? AND stream_ordering < ?" + " ORDER BY topological_ordering DESC, stream_ordering DESC LIMIT ?" + ) + before_args = ( + room_id, token.topological, + room_id, token.topological, token.stream, + before_limit, + ) - txn.execute( - query_before, - ( - room_id, topological_ordering, topological_ordering, - stream_ordering, before_limit, + query_after = ( + "SELECT topological_ordering, stream_ordering, event_id FROM events" + " WHERE room_id = ? AND topological_ordering > ?" + " UNION ALL" + " SELECT topological_ordering, stream_ordering, event_id FROM events" + " WHERE room_id = ? AND topological_ordering = ? AND stream_ordering > ?" + " ORDER BY topological_ordering ASC, stream_ordering ASC LIMIT ?" ) - ) + after_args = ( + room_id, token.topological, + room_id, token.topological, token.stream, + after_limit, + ) + else: + query_before = ( + "SELECT topological_ordering, stream_ordering, event_id FROM events" + " WHERE room_id = ? AND %s" + " ORDER BY topological_ordering DESC, stream_ordering DESC LIMIT ?" + ) % (upper_bound(token, self.database_engine, inclusive=False),) + + before_args = (room_id, before_limit) + + query_after = ( + "SELECT topological_ordering, stream_ordering, event_id FROM events" + " WHERE room_id = ? AND %s" + " ORDER BY topological_ordering ASC, stream_ordering ASC LIMIT ?" + ) % (lower_bound(token, self.database_engine, inclusive=False),) + + after_args = (room_id, after_limit) + + txn.execute(query_before, before_args) rows = self.cursor_to_dict(txn) events_before = [r["event_id"] for r in rows] @@ -623,17 +728,11 @@ class StreamStore(SQLBaseStore): )) else: start_token = str(RoomStreamToken( - topological_ordering, - stream_ordering - 1, + token.topological, + token.stream - 1, )) - txn.execute( - query_after, - ( - room_id, topological_ordering, topological_ordering, - stream_ordering, after_limit, - ) - ) + txn.execute(query_after, after_args) rows = self.cursor_to_dict(txn) events_after = [r["event_id"] for r in rows] @@ -644,10 +743,7 @@ class StreamStore(SQLBaseStore): rows[-1]["stream_ordering"], )) else: - end_token = str(RoomStreamToken( - topological_ordering, - stream_ordering, - )) + end_token = str(token) return { "before": { |