diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py
index c432041b4e..56c434d4e8 100644
--- a/synapse/storage/__init__.py
+++ b/synapse/storage/__init__.py
@@ -36,6 +36,7 @@ from .engines import PostgresEngine
from .event_federation import EventFederationStore
from .event_push_actions import EventPushActionsStore
from .events import EventsStore
+from .events_bg_updates import EventsBackgroundUpdatesStore
from .filtering import FilteringStore
from .group_server import GroupServerStore
from .keys import KeyStore
@@ -49,6 +50,7 @@ from .pusher import PusherStore
from .receipts import ReceiptsStore
from .registration import RegistrationStore
from .rejections import RejectionsStore
+from .relations import RelationsStore
from .room import RoomStore
from .roommember import RoomMemberStore
from .search import SearchStore
@@ -64,6 +66,7 @@ logger = logging.getLogger(__name__)
class DataStore(
+ EventsBackgroundUpdatesStore,
RoomMemberStore,
RoomStore,
RegistrationStore,
@@ -99,6 +102,7 @@ class DataStore(
GroupServerStore,
UserErasureStore,
MonthlyActiveUsersStore,
+ RelationsStore,
):
def __init__(self, db_conn, hs):
self.hs = hs
diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py
index 086318a530..52891bb9eb 100644
--- a/synapse/storage/_base.py
+++ b/synapse/storage/_base.py
@@ -1279,7 +1279,8 @@ class SQLBaseStore(object):
" AND ".join("%s = ?" % (k,) for k in keyvalues),
)
- return txn.execute(sql, list(keyvalues.values()))
+ txn.execute(sql, list(keyvalues.values()))
+ return txn.rowcount
def _simple_delete_many(self, table, column, iterable, keyvalues, desc):
return self.runInteraction(
@@ -1298,9 +1299,12 @@ class SQLBaseStore(object):
column : column name to test for inclusion against `iterable`
iterable : list
keyvalues : dict of column names and values to select the rows with
+
+ Returns:
+ int: Number rows deleted
"""
if not iterable:
- return
+ return 0
sql = "DELETE FROM %s" % table
@@ -1315,7 +1319,9 @@ class SQLBaseStore(object):
if clauses:
sql = "%s WHERE %s" % (sql, " AND ".join(clauses))
- return txn.execute(sql, values)
+ txn.execute(sql, values)
+
+ return txn.rowcount
def _get_cache_dict(
self, db_conn, table, entity_column, stream_column, max_value, limit=100000
diff --git a/synapse/storage/events.py b/synapse/storage/events.py
index 7a7f841c6c..f9162be9b9 100644
--- a/synapse/storage/events.py
+++ b/synapse/storage/events.py
@@ -1,6 +1,7 @@
# -*- coding: utf-8 -*-
# Copyright 2014-2016 OpenMarket Ltd
-# Copyright 2018 New Vector Ltd
+# Copyright 2018-2019 New Vector Ltd
+# Copyright 2019 The Matrix.org Foundation C.I.C.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
@@ -219,41 +220,11 @@ class EventsStore(
EventsWorkerStore,
BackgroundUpdateStore,
):
- EVENT_ORIGIN_SERVER_TS_NAME = "event_origin_server_ts"
- EVENT_FIELDS_SENDER_URL_UPDATE_NAME = "event_fields_sender_url"
def __init__(self, db_conn, hs):
super(EventsStore, self).__init__(db_conn, hs)
- self.register_background_update_handler(
- self.EVENT_ORIGIN_SERVER_TS_NAME, self._background_reindex_origin_server_ts
- )
- self.register_background_update_handler(
- self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME,
- self._background_reindex_fields_sender,
- )
-
- self.register_background_index_update(
- "event_contains_url_index",
- index_name="event_contains_url_index",
- table="events",
- columns=["room_id", "topological_ordering", "stream_ordering"],
- where_clause="contains_url = true AND outlier = false",
- )
-
- # an event_id index on event_search is useful for the purge_history
- # api. Plus it means we get to enforce some integrity with a UNIQUE
- # clause
- self.register_background_index_update(
- "event_search_event_id_idx",
- index_name="event_search_event_id_idx",
- table="event_search",
- columns=["event_id"],
- unique=True,
- psql_only=True,
- )
self._event_persist_queue = _EventPeristenceQueue()
-
self._state_resolution_handler = hs.get_state_resolution_handler()
@defer.inlineCallbacks
@@ -554,10 +525,18 @@ class EventsStore(
e_id for event in new_events for e_id in event.prev_event_ids()
)
- # Finally, remove any events which are prev_events of any existing events.
+ # Remove any events which are prev_events of any existing events.
existing_prevs = yield self._get_events_which_are_prevs(result)
result.difference_update(existing_prevs)
+ # Finally handle the case where the new events have soft-failed prev
+ # events. If they do we need to remove them and their prev events,
+ # otherwise we end up with dangling extremities.
+ existing_prevs = yield self._get_prevs_before_rejected(
+ e_id for event in new_events for e_id in event.prev_event_ids()
+ )
+ result.difference_update(existing_prevs)
+
defer.returnValue(result)
@defer.inlineCallbacks
@@ -573,12 +552,13 @@ class EventsStore(
"""
results = []
- def _get_events(txn, batch):
+ def _get_events_which_are_prevs_txn(txn, batch):
sql = """
- SELECT prev_event_id
+ SELECT prev_event_id, internal_metadata
FROM event_edges
INNER JOIN events USING (event_id)
LEFT JOIN rejections USING (event_id)
+ LEFT JOIN event_json USING (event_id)
WHERE
prev_event_id IN (%s)
AND NOT events.outlier
@@ -588,14 +568,86 @@ class EventsStore(
)
txn.execute(sql, batch)
- results.extend(r[0] for r in txn)
+ results.extend(
+ r[0]
+ for r in txn
+ if not json.loads(r[1]).get("soft_failed")
+ )
for chunk in batch_iter(event_ids, 100):
- yield self.runInteraction("_get_events_which_are_prevs", _get_events, chunk)
+ yield self.runInteraction(
+ "_get_events_which_are_prevs",
+ _get_events_which_are_prevs_txn,
+ chunk,
+ )
defer.returnValue(results)
@defer.inlineCallbacks
+ def _get_prevs_before_rejected(self, event_ids):
+ """Get soft-failed ancestors to remove from the extremities.
+
+ Given a set of events, find all those that have been soft-failed or
+ rejected. Returns those soft failed/rejected events and their prev
+ events (whether soft-failed/rejected or not), and recurses up the
+ prev-event graph until it finds no more soft-failed/rejected events.
+
+ This is used to find extremities that are ancestors of new events, but
+ are separated by soft failed events.
+
+ Args:
+ event_ids (Iterable[str]): Events to find prev events for. Note
+ that these must have already been persisted.
+
+ Returns:
+ Deferred[set[str]]
+ """
+
+ # The set of event_ids to return. This includes all soft-failed events
+ # and their prev events.
+ existing_prevs = set()
+
+ def _get_prevs_before_rejected_txn(txn, batch):
+ to_recursively_check = batch
+
+ while to_recursively_check:
+ sql = """
+ SELECT
+ event_id, prev_event_id, internal_metadata,
+ rejections.event_id IS NOT NULL
+ FROM event_edges
+ INNER JOIN events USING (event_id)
+ LEFT JOIN rejections USING (event_id)
+ LEFT JOIN event_json USING (event_id)
+ WHERE
+ event_id IN (%s)
+ AND NOT events.outlier
+ """ % (
+ ",".join("?" for _ in to_recursively_check),
+ )
+
+ txn.execute(sql, to_recursively_check)
+ to_recursively_check = []
+
+ for event_id, prev_event_id, metadata, rejected in txn:
+ if prev_event_id in existing_prevs:
+ continue
+
+ soft_failed = json.loads(metadata).get("soft_failed")
+ if soft_failed or rejected:
+ to_recursively_check.append(prev_event_id)
+ existing_prevs.add(prev_event_id)
+
+ for chunk in batch_iter(event_ids, 100):
+ yield self.runInteraction(
+ "_get_prevs_before_rejected",
+ _get_prevs_before_rejected_txn,
+ chunk,
+ )
+
+ defer.returnValue(existing_prevs)
+
+ @defer.inlineCallbacks
def _get_new_state_after_events(
self, room_id, events_context, old_latest_event_ids, new_latest_event_ids
):
@@ -1325,6 +1377,9 @@ class EventsStore(
txn, event.room_id, event.redacts
)
+ # Remove from relations table.
+ self._handle_redaction(txn, event.redacts)
+
# Update the event_forward_extremities, event_backward_extremities and
# event_edges tables.
self._handle_mult_prev_events(
@@ -1351,6 +1406,8 @@ class EventsStore(
# Insert into the event_search table.
self._store_guest_access_txn(txn, event)
+ self._handle_event_relations(txn, event)
+
# Insert into the room_memberships table.
self._store_room_members_txn(
txn,
@@ -1493,153 +1550,6 @@ class EventsStore(
ret = yield self.runInteraction("count_daily_active_rooms", _count)
defer.returnValue(ret)
- @defer.inlineCallbacks
- def _background_reindex_fields_sender(self, progress, batch_size):
- target_min_stream_id = progress["target_min_stream_id_inclusive"]
- max_stream_id = progress["max_stream_id_exclusive"]
- rows_inserted = progress.get("rows_inserted", 0)
-
- INSERT_CLUMP_SIZE = 1000
-
- def reindex_txn(txn):
- sql = (
- "SELECT stream_ordering, event_id, json FROM events"
- " INNER JOIN event_json USING (event_id)"
- " WHERE ? <= stream_ordering AND stream_ordering < ?"
- " ORDER BY stream_ordering DESC"
- " LIMIT ?"
- )
-
- txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
-
- rows = txn.fetchall()
- if not rows:
- return 0
-
- min_stream_id = rows[-1][0]
-
- update_rows = []
- for row in rows:
- try:
- event_id = row[1]
- event_json = json.loads(row[2])
- sender = event_json["sender"]
- content = event_json["content"]
-
- contains_url = "url" in content
- if contains_url:
- contains_url &= isinstance(content["url"], text_type)
- except (KeyError, AttributeError):
- # If the event is missing a necessary field then
- # skip over it.
- continue
-
- update_rows.append((sender, contains_url, event_id))
-
- sql = "UPDATE events SET sender = ?, contains_url = ? WHERE event_id = ?"
-
- for index in range(0, len(update_rows), INSERT_CLUMP_SIZE):
- clump = update_rows[index : index + INSERT_CLUMP_SIZE]
- txn.executemany(sql, clump)
-
- progress = {
- "target_min_stream_id_inclusive": target_min_stream_id,
- "max_stream_id_exclusive": min_stream_id,
- "rows_inserted": rows_inserted + len(rows),
- }
-
- self._background_update_progress_txn(
- txn, self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME, progress
- )
-
- return len(rows)
-
- result = yield self.runInteraction(
- self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME, reindex_txn
- )
-
- if not result:
- yield self._end_background_update(self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME)
-
- defer.returnValue(result)
-
- @defer.inlineCallbacks
- def _background_reindex_origin_server_ts(self, progress, batch_size):
- target_min_stream_id = progress["target_min_stream_id_inclusive"]
- max_stream_id = progress["max_stream_id_exclusive"]
- rows_inserted = progress.get("rows_inserted", 0)
-
- INSERT_CLUMP_SIZE = 1000
-
- def reindex_search_txn(txn):
- sql = (
- "SELECT stream_ordering, event_id FROM events"
- " WHERE ? <= stream_ordering AND stream_ordering < ?"
- " ORDER BY stream_ordering DESC"
- " LIMIT ?"
- )
-
- txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
-
- rows = txn.fetchall()
- if not rows:
- return 0
-
- min_stream_id = rows[-1][0]
- event_ids = [row[1] for row in rows]
-
- rows_to_update = []
-
- chunks = [event_ids[i : i + 100] for i in range(0, len(event_ids), 100)]
- for chunk in chunks:
- ev_rows = self._simple_select_many_txn(
- txn,
- table="event_json",
- column="event_id",
- iterable=chunk,
- retcols=["event_id", "json"],
- keyvalues={},
- )
-
- for row in ev_rows:
- event_id = row["event_id"]
- event_json = json.loads(row["json"])
- try:
- origin_server_ts = event_json["origin_server_ts"]
- except (KeyError, AttributeError):
- # If the event is missing a necessary field then
- # skip over it.
- continue
-
- rows_to_update.append((origin_server_ts, event_id))
-
- sql = "UPDATE events SET origin_server_ts = ? WHERE event_id = ?"
-
- for index in range(0, len(rows_to_update), INSERT_CLUMP_SIZE):
- clump = rows_to_update[index : index + INSERT_CLUMP_SIZE]
- txn.executemany(sql, clump)
-
- progress = {
- "target_min_stream_id_inclusive": target_min_stream_id,
- "max_stream_id_exclusive": min_stream_id,
- "rows_inserted": rows_inserted + len(rows_to_update),
- }
-
- self._background_update_progress_txn(
- txn, self.EVENT_ORIGIN_SERVER_TS_NAME, progress
- )
-
- return len(rows_to_update)
-
- result = yield self.runInteraction(
- self.EVENT_ORIGIN_SERVER_TS_NAME, reindex_search_txn
- )
-
- if not result:
- yield self._end_background_update(self.EVENT_ORIGIN_SERVER_TS_NAME)
-
- defer.returnValue(result)
-
def get_current_backfill_token(self):
"""The current minimum token that backfilled events have reached"""
return -self._backfill_id_gen.get_current_token()
@@ -1655,10 +1565,11 @@ class EventsStore(
def get_all_new_forward_event_rows(txn):
sql = (
"SELECT e.stream_ordering, e.event_id, e.room_id, e.type,"
- " state_key, redacts"
+ " state_key, redacts, relates_to_id"
" FROM events AS e"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
+ " LEFT JOIN event_relations USING (event_id)"
" WHERE ? < stream_ordering AND stream_ordering <= ?"
" ORDER BY stream_ordering ASC"
" LIMIT ?"
@@ -1673,11 +1584,12 @@ class EventsStore(
sql = (
"SELECT event_stream_ordering, e.event_id, e.room_id, e.type,"
- " state_key, redacts"
+ " state_key, redacts, relates_to_id"
" FROM events AS e"
" INNER JOIN ex_outlier_stream USING (event_id)"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
+ " LEFT JOIN event_relations USING (event_id)"
" WHERE ? < event_stream_ordering"
" AND event_stream_ordering <= ?"
" ORDER BY event_stream_ordering DESC"
@@ -1698,10 +1610,11 @@ class EventsStore(
def get_all_new_backfill_event_rows(txn):
sql = (
"SELECT -e.stream_ordering, e.event_id, e.room_id, e.type,"
- " state_key, redacts"
+ " state_key, redacts, relates_to_id"
" FROM events AS e"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
+ " LEFT JOIN event_relations USING (event_id)"
" WHERE ? > stream_ordering AND stream_ordering >= ?"
" ORDER BY stream_ordering ASC"
" LIMIT ?"
@@ -1716,11 +1629,12 @@ class EventsStore(
sql = (
"SELECT -event_stream_ordering, e.event_id, e.room_id, e.type,"
- " state_key, redacts"
+ " state_key, redacts, relates_to_id"
" FROM events AS e"
" INNER JOIN ex_outlier_stream USING (event_id)"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
+ " LEFT JOIN event_relations USING (event_id)"
" WHERE ? > event_stream_ordering"
" AND event_stream_ordering >= ?"
" ORDER BY event_stream_ordering DESC"
diff --git a/synapse/storage/events_bg_updates.py b/synapse/storage/events_bg_updates.py
new file mode 100644
index 0000000000..75c1935bf3
--- /dev/null
+++ b/synapse/storage/events_bg_updates.py
@@ -0,0 +1,401 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 The Matrix.org Foundation C.I.C.
+#
+# 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.
+
+import logging
+
+from six import text_type
+
+from canonicaljson import json
+
+from twisted.internet import defer
+
+from synapse.storage.background_updates import BackgroundUpdateStore
+
+logger = logging.getLogger(__name__)
+
+
+class EventsBackgroundUpdatesStore(BackgroundUpdateStore):
+
+ EVENT_ORIGIN_SERVER_TS_NAME = "event_origin_server_ts"
+ EVENT_FIELDS_SENDER_URL_UPDATE_NAME = "event_fields_sender_url"
+ DELETE_SOFT_FAILED_EXTREMITIES = "delete_soft_failed_extremities"
+
+ def __init__(self, db_conn, hs):
+ super(EventsBackgroundUpdatesStore, self).__init__(db_conn, hs)
+
+ self.register_background_update_handler(
+ self.EVENT_ORIGIN_SERVER_TS_NAME, self._background_reindex_origin_server_ts
+ )
+ self.register_background_update_handler(
+ self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME,
+ self._background_reindex_fields_sender,
+ )
+
+ self.register_background_index_update(
+ "event_contains_url_index",
+ index_name="event_contains_url_index",
+ table="events",
+ columns=["room_id", "topological_ordering", "stream_ordering"],
+ where_clause="contains_url = true AND outlier = false",
+ )
+
+ # an event_id index on event_search is useful for the purge_history
+ # api. Plus it means we get to enforce some integrity with a UNIQUE
+ # clause
+ self.register_background_index_update(
+ "event_search_event_id_idx",
+ index_name="event_search_event_id_idx",
+ table="event_search",
+ columns=["event_id"],
+ unique=True,
+ psql_only=True,
+ )
+
+ self.register_background_update_handler(
+ self.DELETE_SOFT_FAILED_EXTREMITIES,
+ self._cleanup_extremities_bg_update,
+ )
+
+ @defer.inlineCallbacks
+ def _background_reindex_fields_sender(self, progress, batch_size):
+ target_min_stream_id = progress["target_min_stream_id_inclusive"]
+ max_stream_id = progress["max_stream_id_exclusive"]
+ rows_inserted = progress.get("rows_inserted", 0)
+
+ INSERT_CLUMP_SIZE = 1000
+
+ def reindex_txn(txn):
+ sql = (
+ "SELECT stream_ordering, event_id, json FROM events"
+ " INNER JOIN event_json USING (event_id)"
+ " WHERE ? <= stream_ordering AND stream_ordering < ?"
+ " ORDER BY stream_ordering DESC"
+ " LIMIT ?"
+ )
+
+ txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
+
+ rows = txn.fetchall()
+ if not rows:
+ return 0
+
+ min_stream_id = rows[-1][0]
+
+ update_rows = []
+ for row in rows:
+ try:
+ event_id = row[1]
+ event_json = json.loads(row[2])
+ sender = event_json["sender"]
+ content = event_json["content"]
+
+ contains_url = "url" in content
+ if contains_url:
+ contains_url &= isinstance(content["url"], text_type)
+ except (KeyError, AttributeError):
+ # If the event is missing a necessary field then
+ # skip over it.
+ continue
+
+ update_rows.append((sender, contains_url, event_id))
+
+ sql = "UPDATE events SET sender = ?, contains_url = ? WHERE event_id = ?"
+
+ for index in range(0, len(update_rows), INSERT_CLUMP_SIZE):
+ clump = update_rows[index : index + INSERT_CLUMP_SIZE]
+ txn.executemany(sql, clump)
+
+ progress = {
+ "target_min_stream_id_inclusive": target_min_stream_id,
+ "max_stream_id_exclusive": min_stream_id,
+ "rows_inserted": rows_inserted + len(rows),
+ }
+
+ self._background_update_progress_txn(
+ txn, self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME, progress
+ )
+
+ return len(rows)
+
+ result = yield self.runInteraction(
+ self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME, reindex_txn
+ )
+
+ if not result:
+ yield self._end_background_update(self.EVENT_FIELDS_SENDER_URL_UPDATE_NAME)
+
+ defer.returnValue(result)
+
+ @defer.inlineCallbacks
+ def _background_reindex_origin_server_ts(self, progress, batch_size):
+ target_min_stream_id = progress["target_min_stream_id_inclusive"]
+ max_stream_id = progress["max_stream_id_exclusive"]
+ rows_inserted = progress.get("rows_inserted", 0)
+
+ INSERT_CLUMP_SIZE = 1000
+
+ def reindex_search_txn(txn):
+ sql = (
+ "SELECT stream_ordering, event_id FROM events"
+ " WHERE ? <= stream_ordering AND stream_ordering < ?"
+ " ORDER BY stream_ordering DESC"
+ " LIMIT ?"
+ )
+
+ txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
+
+ rows = txn.fetchall()
+ if not rows:
+ return 0
+
+ min_stream_id = rows[-1][0]
+ event_ids = [row[1] for row in rows]
+
+ rows_to_update = []
+
+ chunks = [event_ids[i : i + 100] for i in range(0, len(event_ids), 100)]
+ for chunk in chunks:
+ ev_rows = self._simple_select_many_txn(
+ txn,
+ table="event_json",
+ column="event_id",
+ iterable=chunk,
+ retcols=["event_id", "json"],
+ keyvalues={},
+ )
+
+ for row in ev_rows:
+ event_id = row["event_id"]
+ event_json = json.loads(row["json"])
+ try:
+ origin_server_ts = event_json["origin_server_ts"]
+ except (KeyError, AttributeError):
+ # If the event is missing a necessary field then
+ # skip over it.
+ continue
+
+ rows_to_update.append((origin_server_ts, event_id))
+
+ sql = "UPDATE events SET origin_server_ts = ? WHERE event_id = ?"
+
+ for index in range(0, len(rows_to_update), INSERT_CLUMP_SIZE):
+ clump = rows_to_update[index : index + INSERT_CLUMP_SIZE]
+ txn.executemany(sql, clump)
+
+ progress = {
+ "target_min_stream_id_inclusive": target_min_stream_id,
+ "max_stream_id_exclusive": min_stream_id,
+ "rows_inserted": rows_inserted + len(rows_to_update),
+ }
+
+ self._background_update_progress_txn(
+ txn, self.EVENT_ORIGIN_SERVER_TS_NAME, progress
+ )
+
+ return len(rows_to_update)
+
+ result = yield self.runInteraction(
+ self.EVENT_ORIGIN_SERVER_TS_NAME, reindex_search_txn
+ )
+
+ if not result:
+ yield self._end_background_update(self.EVENT_ORIGIN_SERVER_TS_NAME)
+
+ defer.returnValue(result)
+
+ @defer.inlineCallbacks
+ def _cleanup_extremities_bg_update(self, progress, batch_size):
+ """Background update to clean out extremities that should have been
+ deleted previously.
+
+ Mainly used to deal with the aftermath of #5269.
+ """
+
+ # This works by first copying all existing forward extremities into the
+ # `_extremities_to_check` table at start up, and then checking each
+ # event in that table whether we have any descendants that are not
+ # soft-failed/rejected. If that is the case then we delete that event
+ # from the forward extremities table.
+ #
+ # For efficiency, we do this in batches by recursively pulling out all
+ # descendants of a batch until we find the non soft-failed/rejected
+ # events, i.e. the set of descendants whose chain of prev events back
+ # to the batch of extremities are all soft-failed or rejected.
+ # Typically, we won't find any such events as extremities will rarely
+ # have any descendants, but if they do then we should delete those
+ # extremities.
+
+ def _cleanup_extremities_bg_update_txn(txn):
+ # The set of extremity event IDs that we're checking this round
+ original_set = set()
+
+ # A dict[str, set[str]] of event ID to their prev events.
+ graph = {}
+
+ # The set of descendants of the original set that are not rejected
+ # nor soft-failed. Ancestors of these events should be removed
+ # from the forward extremities table.
+ non_rejected_leaves = set()
+
+ # Set of event IDs that have been soft failed, and for which we
+ # should check if they have descendants which haven't been soft
+ # failed.
+ soft_failed_events_to_lookup = set()
+
+ # First, we get `batch_size` events from the table, pulling out
+ # their successor events, if any, and the successor events'
+ # rejection status.
+ txn.execute(
+ """SELECT prev_event_id, event_id, internal_metadata,
+ rejections.event_id IS NOT NULL, events.outlier
+ FROM (
+ SELECT event_id AS prev_event_id
+ FROM _extremities_to_check
+ LIMIT ?
+ ) AS f
+ LEFT JOIN event_edges USING (prev_event_id)
+ LEFT JOIN events USING (event_id)
+ LEFT JOIN event_json USING (event_id)
+ LEFT JOIN rejections USING (event_id)
+ """, (batch_size,)
+ )
+
+ for prev_event_id, event_id, metadata, rejected, outlier in txn:
+ original_set.add(prev_event_id)
+
+ if not event_id or outlier:
+ # Common case where the forward extremity doesn't have any
+ # descendants.
+ continue
+
+ graph.setdefault(event_id, set()).add(prev_event_id)
+
+ soft_failed = False
+ if metadata:
+ soft_failed = json.loads(metadata).get("soft_failed")
+
+ if soft_failed or rejected:
+ soft_failed_events_to_lookup.add(event_id)
+ else:
+ non_rejected_leaves.add(event_id)
+
+ # Now we recursively check all the soft-failed descendants we
+ # found above in the same way, until we have nothing left to
+ # check.
+ while soft_failed_events_to_lookup:
+ # We only want to do 100 at a time, so we split given list
+ # into two.
+ batch = list(soft_failed_events_to_lookup)
+ to_check, to_defer = batch[:100], batch[100:]
+ soft_failed_events_to_lookup = set(to_defer)
+
+ sql = """SELECT prev_event_id, event_id, internal_metadata,
+ rejections.event_id IS NOT NULL
+ FROM event_edges
+ INNER JOIN events USING (event_id)
+ INNER JOIN event_json USING (event_id)
+ LEFT JOIN rejections USING (event_id)
+ WHERE
+ prev_event_id IN (%s)
+ AND NOT events.outlier
+ """ % (
+ ",".join("?" for _ in to_check),
+ )
+ txn.execute(sql, to_check)
+
+ for prev_event_id, event_id, metadata, rejected in txn:
+ if event_id in graph:
+ # Already handled this event previously, but we still
+ # want to record the edge.
+ graph[event_id].add(prev_event_id)
+ continue
+
+ graph[event_id] = {prev_event_id}
+
+ soft_failed = json.loads(metadata).get("soft_failed")
+ if soft_failed or rejected:
+ soft_failed_events_to_lookup.add(event_id)
+ else:
+ non_rejected_leaves.add(event_id)
+
+ # We have a set of non-soft-failed descendants, so we recurse up
+ # the graph to find all ancestors and add them to the set of event
+ # IDs that we can delete from forward extremities table.
+ to_delete = set()
+ while non_rejected_leaves:
+ event_id = non_rejected_leaves.pop()
+ prev_event_ids = graph.get(event_id, set())
+ non_rejected_leaves.update(prev_event_ids)
+ to_delete.update(prev_event_ids)
+
+ to_delete.intersection_update(original_set)
+
+ deleted = self._simple_delete_many_txn(
+ txn=txn,
+ table="event_forward_extremities",
+ column="event_id",
+ iterable=to_delete,
+ keyvalues={},
+ )
+
+ logger.info(
+ "Deleted %d forward extremities of %d checked, to clean up #5269",
+ deleted,
+ len(original_set),
+ )
+
+ if deleted:
+ # We now need to invalidate the caches of these rooms
+ rows = self._simple_select_many_txn(
+ txn,
+ table="events",
+ column="event_id",
+ iterable=to_delete,
+ keyvalues={},
+ retcols=("room_id",)
+ )
+ room_ids = set(row["room_id"] for row in rows)
+ for room_id in room_ids:
+ txn.call_after(
+ self.get_latest_event_ids_in_room.invalidate,
+ (room_id,)
+ )
+
+ self._simple_delete_many_txn(
+ txn=txn,
+ table="_extremities_to_check",
+ column="event_id",
+ iterable=original_set,
+ keyvalues={},
+ )
+
+ return len(original_set)
+
+ num_handled = yield self.runInteraction(
+ "_cleanup_extremities_bg_update", _cleanup_extremities_bg_update_txn,
+ )
+
+ if not num_handled:
+ yield self._end_background_update(self.DELETE_SOFT_FAILED_EXTREMITIES)
+
+ def _drop_table_txn(txn):
+ txn.execute("DROP TABLE _extremities_to_check")
+
+ yield self.runInteraction(
+ "_cleanup_extremities_bg_update_drop_table",
+ _drop_table_txn,
+ )
+
+ defer.returnValue(num_handled)
diff --git a/synapse/storage/relations.py b/synapse/storage/relations.py
new file mode 100644
index 0000000000..4c83800cca
--- /dev/null
+++ b/synapse/storage/relations.py
@@ -0,0 +1,476 @@
+# -*- coding: utf-8 -*-
+# Copyright 2019 New Vector 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.
+
+import logging
+
+import attr
+
+from twisted.internet import defer
+
+from synapse.api.constants import RelationTypes
+from synapse.api.errors import SynapseError
+from synapse.storage._base import SQLBaseStore
+from synapse.storage.stream import generate_pagination_where_clause
+from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
+
+logger = logging.getLogger(__name__)
+
+
+@attr.s
+class PaginationChunk(object):
+ """Returned by relation pagination APIs.
+
+ Attributes:
+ chunk (list): The rows returned by pagination
+ next_batch (Any|None): Token to fetch next set of results with, if
+ None then there are no more results.
+ prev_batch (Any|None): Token to fetch previous set of results with, if
+ None then there are no previous results.
+ """
+
+ chunk = attr.ib()
+ next_batch = attr.ib(default=None)
+ prev_batch = attr.ib(default=None)
+
+ def to_dict(self):
+ d = {"chunk": self.chunk}
+
+ if self.next_batch:
+ d["next_batch"] = self.next_batch.to_string()
+
+ if self.prev_batch:
+ d["prev_batch"] = self.prev_batch.to_string()
+
+ return d
+
+
+@attr.s(frozen=True, slots=True)
+class RelationPaginationToken(object):
+ """Pagination token for relation pagination API.
+
+ As the results are order by topological ordering, we can use the
+ `topological_ordering` and `stream_ordering` fields of the events at the
+ boundaries of the chunk as pagination tokens.
+
+ Attributes:
+ topological (int): The topological ordering of the boundary event
+ stream (int): The stream ordering of the boundary event.
+ """
+
+ topological = attr.ib()
+ stream = attr.ib()
+
+ @staticmethod
+ def from_string(string):
+ try:
+ t, s = string.split("-")
+ return RelationPaginationToken(int(t), int(s))
+ except ValueError:
+ raise SynapseError(400, "Invalid token")
+
+ def to_string(self):
+ return "%d-%d" % (self.topological, self.stream)
+
+ def as_tuple(self):
+ return attr.astuple(self)
+
+
+@attr.s(frozen=True, slots=True)
+class AggregationPaginationToken(object):
+ """Pagination token for relation aggregation pagination API.
+
+ As the results are order by count and then MAX(stream_ordering) of the
+ aggregation groups, we can just use them as our pagination token.
+
+ Attributes:
+ count (int): The count of relations in the boundar group.
+ stream (int): The MAX stream ordering in the boundary group.
+ """
+
+ count = attr.ib()
+ stream = attr.ib()
+
+ @staticmethod
+ def from_string(string):
+ try:
+ c, s = string.split("-")
+ return AggregationPaginationToken(int(c), int(s))
+ except ValueError:
+ raise SynapseError(400, "Invalid token")
+
+ def to_string(self):
+ return "%d-%d" % (self.count, self.stream)
+
+ def as_tuple(self):
+ return attr.astuple(self)
+
+
+class RelationsWorkerStore(SQLBaseStore):
+ @cached(tree=True)
+ def get_relations_for_event(
+ self,
+ event_id,
+ relation_type=None,
+ event_type=None,
+ aggregation_key=None,
+ limit=5,
+ direction="b",
+ from_token=None,
+ to_token=None,
+ ):
+ """Get a list of relations for an event, ordered by topological ordering.
+
+ Args:
+ event_id (str): Fetch events that relate to this event ID.
+ relation_type (str|None): Only fetch events with this relation
+ type, if given.
+ event_type (str|None): Only fetch events with this event type, if
+ given.
+ aggregation_key (str|None): Only fetch events with this aggregation
+ key, if given.
+ limit (int): Only fetch the most recent `limit` events.
+ direction (str): Whether to fetch the most recent first (`"b"`) or
+ the oldest first (`"f"`).
+ from_token (RelationPaginationToken|None): Fetch rows from the given
+ token, or from the start if None.
+ to_token (RelationPaginationToken|None): Fetch rows up to the given
+ token, or up to the end if None.
+
+ Returns:
+ Deferred[PaginationChunk]: List of event IDs that match relations
+ requested. The rows are of the form `{"event_id": "..."}`.
+ """
+
+ where_clause = ["relates_to_id = ?"]
+ where_args = [event_id]
+
+ if relation_type is not None:
+ where_clause.append("relation_type = ?")
+ where_args.append(relation_type)
+
+ if event_type is not None:
+ where_clause.append("type = ?")
+ where_args.append(event_type)
+
+ if aggregation_key:
+ where_clause.append("aggregation_key = ?")
+ where_args.append(aggregation_key)
+
+ pagination_clause = generate_pagination_where_clause(
+ direction=direction,
+ column_names=("topological_ordering", "stream_ordering"),
+ from_token=attr.astuple(from_token) if from_token else None,
+ to_token=attr.astuple(to_token) if to_token else None,
+ engine=self.database_engine,
+ )
+
+ if pagination_clause:
+ where_clause.append(pagination_clause)
+
+ if direction == "b":
+ order = "DESC"
+ else:
+ order = "ASC"
+
+ sql = """
+ SELECT event_id, topological_ordering, stream_ordering
+ FROM event_relations
+ INNER JOIN events USING (event_id)
+ WHERE %s
+ ORDER BY topological_ordering %s, stream_ordering %s
+ LIMIT ?
+ """ % (
+ " AND ".join(where_clause),
+ order,
+ order,
+ )
+
+ def _get_recent_references_for_event_txn(txn):
+ txn.execute(sql, where_args + [limit + 1])
+
+ last_topo_id = None
+ last_stream_id = None
+ events = []
+ for row in txn:
+ events.append({"event_id": row[0]})
+ last_topo_id = row[1]
+ last_stream_id = row[2]
+
+ next_batch = None
+ if len(events) > limit and last_topo_id and last_stream_id:
+ next_batch = RelationPaginationToken(last_topo_id, last_stream_id)
+
+ return PaginationChunk(
+ chunk=list(events[:limit]), next_batch=next_batch, prev_batch=from_token
+ )
+
+ return self.runInteraction(
+ "get_recent_references_for_event", _get_recent_references_for_event_txn
+ )
+
+ @cached(tree=True)
+ def get_aggregation_groups_for_event(
+ self,
+ event_id,
+ event_type=None,
+ limit=5,
+ direction="b",
+ from_token=None,
+ to_token=None,
+ ):
+ """Get a list of annotations on the event, grouped by event type and
+ aggregation key, sorted by count.
+
+ This is used e.g. to get the what and how many reactions have happend
+ on an event.
+
+ Args:
+ event_id (str): Fetch events that relate to this event ID.
+ event_type (str|None): Only fetch events with this event type, if
+ given.
+ limit (int): Only fetch the `limit` groups.
+ direction (str): Whether to fetch the highest count first (`"b"`) or
+ the lowest count first (`"f"`).
+ from_token (AggregationPaginationToken|None): Fetch rows from the
+ given token, or from the start if None.
+ to_token (AggregationPaginationToken|None): Fetch rows up to the
+ given token, or up to the end if None.
+
+
+ Returns:
+ Deferred[PaginationChunk]: List of groups of annotations that
+ match. Each row is a dict with `type`, `key` and `count` fields.
+ """
+
+ where_clause = ["relates_to_id = ?", "relation_type = ?"]
+ where_args = [event_id, RelationTypes.ANNOTATION]
+
+ if event_type:
+ where_clause.append("type = ?")
+ where_args.append(event_type)
+
+ having_clause = generate_pagination_where_clause(
+ direction=direction,
+ column_names=("COUNT(*)", "MAX(stream_ordering)"),
+ from_token=attr.astuple(from_token) if from_token else None,
+ to_token=attr.astuple(to_token) if to_token else None,
+ engine=self.database_engine,
+ )
+
+ if direction == "b":
+ order = "DESC"
+ else:
+ order = "ASC"
+
+ if having_clause:
+ having_clause = "HAVING " + having_clause
+ else:
+ having_clause = ""
+
+ sql = """
+ SELECT type, aggregation_key, COUNT(DISTINCT sender), MAX(stream_ordering)
+ FROM event_relations
+ INNER JOIN events USING (event_id)
+ WHERE {where_clause}
+ GROUP BY relation_type, type, aggregation_key
+ {having_clause}
+ ORDER BY COUNT(*) {order}, MAX(stream_ordering) {order}
+ LIMIT ?
+ """.format(
+ where_clause=" AND ".join(where_clause),
+ order=order,
+ having_clause=having_clause,
+ )
+
+ def _get_aggregation_groups_for_event_txn(txn):
+ txn.execute(sql, where_args + [limit + 1])
+
+ next_batch = None
+ events = []
+ for row in txn:
+ events.append({"type": row[0], "key": row[1], "count": row[2]})
+ next_batch = AggregationPaginationToken(row[2], row[3])
+
+ if len(events) <= limit:
+ next_batch = None
+
+ return PaginationChunk(
+ chunk=list(events[:limit]), next_batch=next_batch, prev_batch=from_token
+ )
+
+ return self.runInteraction(
+ "get_aggregation_groups_for_event", _get_aggregation_groups_for_event_txn
+ )
+
+ @cachedInlineCallbacks()
+ def get_applicable_edit(self, event_id):
+ """Get the most recent edit (if any) that has happened for the given
+ event.
+
+ Correctly handles checking whether edits were allowed to happen.
+
+ Args:
+ event_id (str): The original event ID
+
+ Returns:
+ Deferred[EventBase|None]: Returns the most recent edit, if any.
+ """
+
+ # We only allow edits for `m.room.message` events that have the same sender
+ # and event type. We can't assert these things during regular event auth so
+ # we have to do the checks post hoc.
+
+ # Fetches latest edit that has the same type and sender as the
+ # original, and is an `m.room.message`.
+ sql = """
+ SELECT edit.event_id FROM events AS edit
+ INNER JOIN event_relations USING (event_id)
+ INNER JOIN events AS original ON
+ original.event_id = relates_to_id
+ AND edit.type = original.type
+ AND edit.sender = original.sender
+ WHERE
+ relates_to_id = ?
+ AND relation_type = ?
+ AND edit.type = 'm.room.message'
+ ORDER by edit.origin_server_ts DESC, edit.event_id DESC
+ LIMIT 1
+ """
+
+ def _get_applicable_edit_txn(txn):
+ txn.execute(sql, (event_id, RelationTypes.REPLACE))
+ row = txn.fetchone()
+ if row:
+ return row[0]
+
+ edit_id = yield self.runInteraction(
+ "get_applicable_edit", _get_applicable_edit_txn
+ )
+
+ if not edit_id:
+ return
+
+ edit_event = yield self.get_event(edit_id, allow_none=True)
+ defer.returnValue(edit_event)
+
+ def has_user_annotated_event(self, parent_id, event_type, aggregation_key, sender):
+ """Check if a user has already annotated an event with the same key
+ (e.g. already liked an event).
+
+ Args:
+ parent_id (str): The event being annotated
+ event_type (str): The event type of the annotation
+ aggregation_key (str): The aggregation key of the annotation
+ sender (str): The sender of the annotation
+
+ Returns:
+ Deferred[bool]
+ """
+
+ sql = """
+ SELECT 1 FROM event_relations
+ INNER JOIN events USING (event_id)
+ WHERE
+ relates_to_id = ?
+ AND relation_type = ?
+ AND type = ?
+ AND sender = ?
+ AND aggregation_key = ?
+ LIMIT 1;
+ """
+
+ def _get_if_user_has_annotated_event(txn):
+ txn.execute(
+ sql,
+ (
+ parent_id,
+ RelationTypes.ANNOTATION,
+ event_type,
+ sender,
+ aggregation_key,
+ ),
+ )
+
+ return bool(txn.fetchone())
+
+ return self.runInteraction(
+ "get_if_user_has_annotated_event", _get_if_user_has_annotated_event
+ )
+
+
+class RelationsStore(RelationsWorkerStore):
+ def _handle_event_relations(self, txn, event):
+ """Handles inserting relation data during peristence of events
+
+ Args:
+ txn
+ event (EventBase)
+ """
+ relation = event.content.get("m.relates_to")
+ if not relation:
+ # No relations
+ return
+
+ rel_type = relation.get("rel_type")
+ if rel_type not in (
+ RelationTypes.ANNOTATION,
+ RelationTypes.REFERENCE,
+ RelationTypes.REPLACE,
+ ):
+ # Unknown relation type
+ return
+
+ parent_id = relation.get("event_id")
+ if not parent_id:
+ # Invalid relation
+ return
+
+ aggregation_key = relation.get("key")
+
+ self._simple_insert_txn(
+ txn,
+ table="event_relations",
+ values={
+ "event_id": event.event_id,
+ "relates_to_id": parent_id,
+ "relation_type": rel_type,
+ "aggregation_key": aggregation_key,
+ },
+ )
+
+ txn.call_after(self.get_relations_for_event.invalidate_many, (parent_id,))
+ txn.call_after(
+ self.get_aggregation_groups_for_event.invalidate_many, (parent_id,)
+ )
+
+ if rel_type == RelationTypes.REPLACE:
+ txn.call_after(self.get_applicable_edit.invalidate, (parent_id,))
+
+ def _handle_redaction(self, txn, redacted_event_id):
+ """Handles receiving a redaction and checking whether we need to remove
+ any redacted relations from the database.
+
+ Args:
+ txn
+ redacted_event_id (str): The event that was redacted.
+ """
+
+ self._simple_delete_txn(
+ txn,
+ table="event_relations",
+ keyvalues={
+ "event_id": redacted_event_id,
+ }
+ )
diff --git a/synapse/storage/schema/delta/54/delete_forward_extremities.sql b/synapse/storage/schema/delta/54/delete_forward_extremities.sql
new file mode 100644
index 0000000000..b062ec840c
--- /dev/null
+++ b/synapse/storage/schema/delta/54/delete_forward_extremities.sql
@@ -0,0 +1,23 @@
+/* Copyright 2019 The Matrix.org Foundation C.I.C.
+ *
+ * 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.
+ */
+
+-- Start a background job to cleanup extremities that were incorrectly added
+-- by bug #5269.
+INSERT INTO background_updates (update_name, progress_json) VALUES
+ ('delete_soft_failed_extremities', '{}');
+
+DROP TABLE IF EXISTS _extremities_to_check; -- To make this delta schema file idempotent.
+CREATE TABLE _extremities_to_check AS SELECT event_id FROM event_forward_extremities;
+CREATE INDEX _extremities_to_check_id ON _extremities_to_check(event_id);
diff --git a/synapse/storage/schema/delta/54/relations.sql b/synapse/storage/schema/delta/54/relations.sql
new file mode 100644
index 0000000000..134862b870
--- /dev/null
+++ b/synapse/storage/schema/delta/54/relations.sql
@@ -0,0 +1,27 @@
+/* Copyright 2019 New Vector 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.
+ */
+
+-- Tracks related events, like reactions, replies, edits, etc. Note that things
+-- in this table are not necessarily "valid", e.g. it may contain edits from
+-- people who don't have power to edit other peoples events.
+CREATE TABLE IF NOT EXISTS event_relations (
+ event_id TEXT NOT NULL,
+ relates_to_id TEXT NOT NULL,
+ relation_type TEXT NOT NULL,
+ aggregation_key TEXT
+);
+
+CREATE UNIQUE INDEX event_relations_id ON event_relations(event_id);
+CREATE INDEX event_relations_relates ON event_relations(relates_to_id, relation_type, aggregation_key);
diff --git a/synapse/storage/stream.py b/synapse/storage/stream.py
index d105b6b17d..529ad4ea79 100644
--- a/synapse/storage/stream.py
+++ b/synapse/storage/stream.py
@@ -64,59 +64,135 @@ _EventDictReturn = namedtuple(
)
-def lower_bound(token, engine, inclusive=False):
- inclusive = "=" if inclusive else ""
- if token.topological is None:
- return "(%d <%s %s)" % (token.stream, inclusive, "stream_ordering")
- else:
- 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",
+def generate_pagination_where_clause(
+ direction, column_names, from_token, to_token, engine,
+):
+ """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 (str): Whether we're paginating backwards("b") or
+ forwards ("f").
+ column_names (tuple[str, str]): The column names to bound. Must *not*
+ be user defined as these get inserted directly into the SQL
+ statement without escapes.
+ from_token (tuple[int, int]|None): The start point for the pagination.
+ This is an exclusive minimum bound if direction is "f", and an
+ inclusive maximum bound if direction is "b".
+ to_token (tuple[int, int]|None): The endpoint point for the pagination.
+ This is an inclusive maximum bound if direction is "f", and an
+ exclusive minimum bound if direction is "b".
+ engine: The database engine to generate the clauses for
+
+ Returns:
+ str: The sql expression
+ """
+ assert direction in ("b", "f")
+
+ where_clause = []
+ if from_token:
+ where_clause.append(
+ _make_generic_sql_bound(
+ bound=">=" if direction == "b" else "<",
+ column_names=column_names,
+ values=from_token,
+ engine=engine,
)
- return "(%d < %s OR (%d = %s AND %d <%s %s))" % (
- token.topological,
- "topological_ordering",
- token.topological,
- "topological_ordering",
- token.stream,
- inclusive,
- "stream_ordering",
- )
-
-
-def upper_bound(token, engine, inclusive=True):
- inclusive = "=" if inclusive else ""
- if token.topological is None:
- return "(%d >%s %s)" % (token.stream, inclusive, "stream_ordering")
- else:
- 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",
+ )
+
+ if to_token:
+ where_clause.append(
+ _make_generic_sql_bound(
+ bound="<" if direction == "b" else ">=",
+ column_names=column_names,
+ values=to_token,
+ engine=engine,
)
- return "(%d > %s OR (%d = %s AND %d >%s %s))" % (
- token.topological,
- "topological_ordering",
- token.topological,
- "topological_ordering",
- token.stream,
- inclusive,
- "stream_ordering",
)
+ return " AND ".join(where_clause)
+
+
+def _make_generic_sql_bound(bound, column_names, values, engine):
+ """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 (str): The comparison operator to use. One of ">", "<", ">=",
+ "<=", where the values are on the left and columns on the right.
+ names (tuple[str, str]): The column names. Must *not* be user defined
+ as these get inserted directly into the SQL statement without
+ escapes.
+ values (tuple[int|None, int]): 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:
+ str
+ """
+
+ 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<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))" % (
+ val1, val2,
+ bound,
+ name1, name2,
+ )
+
+ # We want to generate queries of e.g. the form:
+ #
+ # (val1 < name1 OR (val1 = name1 AND val2 <= name2))
+ #
+ # which is equivalent to (val1, val2) < (name1, name2)
+
+ return """(
+ {val1:d} {strict_bound} {name1}
+ OR ({val1:d} = {name1} AND {val2:d} {bound} {name2})
+ )""".format(
+ name1=name1,
+ val1=val1,
+ name2=name2,
+ val2=val2,
+ strict_bound=bound[0], # The first bound must always be strict equality here
+ bound=bound,
+ )
+
def filter_to_clause(event_filter):
# NB: This may create SQL clauses that don't optimise well (and we don't
@@ -762,20 +838,16 @@ class StreamWorkerStore(EventsWorkerStore, SQLBaseStore):
args = [False, room_id]
if direction == 'b':
order = "DESC"
- bounds = upper_bound(from_token, self.database_engine)
- if to_token:
- bounds = "%s AND %s" % (
- bounds,
- lower_bound(to_token, self.database_engine),
- )
else:
order = "ASC"
- bounds = lower_bound(from_token, self.database_engine)
- if to_token:
- bounds = "%s AND %s" % (
- bounds,
- upper_bound(to_token, self.database_engine),
- )
+
+ bounds = generate_pagination_where_clause(
+ direction=direction,
+ column_names=("topological_ordering", "stream_ordering"),
+ from_token=from_token,
+ to_token=to_token,
+ engine=self.database_engine,
+ )
filter_clause, filter_args = filter_to_clause(event_filter)
|
