diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py
index 7522d3fd57..71316f7d09 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
@@ -55,6 +56,7 @@ from .roommember import RoomMemberStore
from .search import SearchStore
from .signatures import SignatureStore
from .state import StateStore
+from .stats import StatsStore
from .stream import StreamStore
from .tags import TagsStore
from .transactions import TransactionStore
@@ -65,6 +67,7 @@ logger = logging.getLogger(__name__)
class DataStore(
+ EventsBackgroundUpdatesStore,
RoomMemberStore,
RoomStore,
RegistrationStore,
@@ -100,6 +103,7 @@ class DataStore(
GroupServerStore,
UserErasureStore,
MonthlyActiveUsersStore,
+ StatsStore,
RelationsStore,
):
def __init__(self, db_conn, hs):
diff --git a/synapse/storage/_base.py b/synapse/storage/_base.py
index fa6839ceca..52891bb9eb 100644
--- a/synapse/storage/_base.py
+++ b/synapse/storage/_base.py
@@ -16,6 +16,7 @@
# limitations under the License.
import itertools
import logging
+import random
import sys
import threading
import time
@@ -247,6 +248,8 @@ class SQLBaseStore(object):
self._check_safe_to_upsert,
)
+ self.rand = random.SystemRandom()
+
if self._account_validity.enabled:
self._clock.call_later(
0.0,
@@ -308,21 +311,36 @@ class SQLBaseStore(object):
res = self.cursor_to_dict(txn)
if res:
for user in res:
- self.set_expiration_date_for_user_txn(txn, user["name"])
+ self.set_expiration_date_for_user_txn(
+ txn,
+ user["name"],
+ use_delta=True,
+ )
yield self.runInteraction(
"get_users_with_no_expiration_date",
select_users_with_no_expiration_date_txn,
)
- def set_expiration_date_for_user_txn(self, txn, user_id):
+ def set_expiration_date_for_user_txn(self, txn, user_id, use_delta=False):
"""Sets an expiration date to the account with the given user ID.
Args:
user_id (str): User ID to set an expiration date for.
+ use_delta (bool): If set to False, the expiration date for the user will be
+ now + validity period. If set to True, this expiration date will be a
+ random value in the [now + period - d ; now + period] range, d being a
+ delta equal to 10% of the validity period.
"""
now_ms = self._clock.time_msec()
expiration_ts = now_ms + self._account_validity.period
+
+ if use_delta:
+ expiration_ts = self.rand.randrange(
+ expiration_ts - self._account_validity.startup_job_max_delta,
+ expiration_ts,
+ )
+
self._simple_insert_txn(
txn,
"account_validity",
@@ -1261,7 +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(
@@ -1280,9 +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
@@ -1297,7 +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 2ffc27ff41..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
@@ -573,7 +552,7 @@ class EventsStore(
"""
results = []
- def _get_events(txn, batch):
+ def _get_events_which_are_prevs_txn(txn, batch):
sql = """
SELECT prev_event_id, internal_metadata
FROM event_edges
@@ -596,11 +575,79 @@ class EventsStore(
)
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
):
@@ -1503,153 +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()
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/events_worker.py b/synapse/storage/events_worker.py
index adc6cf26b5..1782428048 100644
--- a/synapse/storage/events_worker.py
+++ b/synapse/storage/events_worker.py
@@ -13,6 +13,8 @@
# See the License for the specific language governing permissions and
# limitations under the License.
+from __future__ import division
+
import itertools
import logging
from collections import namedtuple
@@ -610,4 +612,79 @@ class EventsWorkerStore(SQLBaseStore):
return res
- return self.runInteraction("get_rejection_reasons", f)
+ return self.runInteraction("get_seen_events_with_rejections", f)
+
+ def _get_total_state_event_counts_txn(self, txn, room_id):
+ """
+ See get_total_state_event_counts.
+ """
+ # We join against the events table as that has an index on room_id
+ sql = """
+ SELECT COUNT(*) FROM state_events
+ INNER JOIN events USING (room_id, event_id)
+ WHERE room_id=?
+ """
+ txn.execute(sql, (room_id,))
+ row = txn.fetchone()
+ return row[0] if row else 0
+
+ def get_total_state_event_counts(self, room_id):
+ """
+ Gets the total number of state events in a room.
+
+ Args:
+ room_id (str)
+
+ Returns:
+ Deferred[int]
+ """
+ return self.runInteraction(
+ "get_total_state_event_counts",
+ self._get_total_state_event_counts_txn, room_id
+ )
+
+ def _get_current_state_event_counts_txn(self, txn, room_id):
+ """
+ See get_current_state_event_counts.
+ """
+ sql = "SELECT COUNT(*) FROM current_state_events WHERE room_id=?"
+ txn.execute(sql, (room_id,))
+ row = txn.fetchone()
+ return row[0] if row else 0
+
+ def get_current_state_event_counts(self, room_id):
+ """
+ Gets the current number of state events in a room.
+
+ Args:
+ room_id (str)
+
+ Returns:
+ Deferred[int]
+ """
+ return self.runInteraction(
+ "get_current_state_event_counts",
+ self._get_current_state_event_counts_txn, room_id
+ )
+
+ @defer.inlineCallbacks
+ def get_room_complexity(self, room_id):
+ """
+ Get a rough approximation of the complexity of the room. This is used by
+ remote servers to decide whether they wish to join the room or not.
+ Higher complexity value indicates that being in the room will consume
+ more resources.
+
+ Args:
+ room_id (str)
+
+ Returns:
+ Deferred[dict[str:int]] of complexity version to complexity.
+ """
+ state_events = yield self.get_current_state_event_counts(room_id)
+
+ # Call this one "v1", so we can introduce new ones as we want to develop
+ # it.
+ complexity_v1 = round(state_events / 500, 2)
+
+ defer.returnValue({"v1": complexity_v1})
diff --git a/synapse/storage/keys.py b/synapse/storage/keys.py
index 7036541792..5300720dbb 100644
--- a/synapse/storage/keys.py
+++ b/synapse/storage/keys.py
@@ -19,6 +19,7 @@ import logging
import six
+import attr
from signedjson.key import decode_verify_key_bytes
from synapse.util import batch_iter
@@ -36,6 +37,12 @@ else:
db_binary_type = memoryview
+@attr.s(slots=True, frozen=True)
+class FetchKeyResult(object):
+ verify_key = attr.ib() # VerifyKey: the key itself
+ valid_until_ts = attr.ib() # int: how long we can use this key for
+
+
class KeyStore(SQLBaseStore):
"""Persistence for signature verification keys
"""
@@ -54,8 +61,8 @@ class KeyStore(SQLBaseStore):
iterable of (server_name, key-id) tuples to fetch keys for
Returns:
- Deferred: resolves to dict[Tuple[str, str], VerifyKey|None]:
- map from (server_name, key_id) -> VerifyKey, or None if the key is
+ Deferred: resolves to dict[Tuple[str, str], FetchKeyResult|None]:
+ map from (server_name, key_id) -> FetchKeyResult, or None if the key is
unknown
"""
keys = {}
@@ -65,17 +72,19 @@ class KeyStore(SQLBaseStore):
# batch_iter always returns tuples so it's safe to do len(batch)
sql = (
- "SELECT server_name, key_id, verify_key FROM server_signature_keys "
- "WHERE 1=0"
+ "SELECT server_name, key_id, verify_key, ts_valid_until_ms "
+ "FROM server_signature_keys WHERE 1=0"
) + " OR (server_name=? AND key_id=?)" * len(batch)
txn.execute(sql, tuple(itertools.chain.from_iterable(batch)))
for row in txn:
- server_name, key_id, key_bytes = row
- keys[(server_name, key_id)] = decode_verify_key_bytes(
- key_id, bytes(key_bytes)
+ server_name, key_id, key_bytes, ts_valid_until_ms = row
+ res = FetchKeyResult(
+ verify_key=decode_verify_key_bytes(key_id, bytes(key_bytes)),
+ valid_until_ts=ts_valid_until_ms,
)
+ keys[(server_name, key_id)] = res
def _txn(txn):
for batch in batch_iter(server_name_and_key_ids, 50):
@@ -84,38 +93,53 @@ class KeyStore(SQLBaseStore):
return self.runInteraction("get_server_verify_keys", _txn)
- def store_server_verify_key(
- self, server_name, from_server, time_now_ms, verify_key
- ):
- """Stores a NACL verification key for the given server.
+ def store_server_verify_keys(self, from_server, ts_added_ms, verify_keys):
+ """Stores NACL verification keys for remote servers.
Args:
- server_name (str): The name of the server.
- from_server (str): Where the verification key was looked up
- time_now_ms (int): The time now in milliseconds
- verify_key (nacl.signing.VerifyKey): The NACL verify key.
+ from_server (str): Where the verification keys were looked up
+ ts_added_ms (int): The time to record that the key was added
+ verify_keys (iterable[tuple[str, str, FetchKeyResult]]):
+ keys to be stored. Each entry is a triplet of
+ (server_name, key_id, key).
"""
- key_id = "%s:%s" % (verify_key.alg, verify_key.version)
-
- # XXX fix this to not need a lock (#3819)
- def _txn(txn):
- self._simple_upsert_txn(
- txn,
- table="server_signature_keys",
- keyvalues={"server_name": server_name, "key_id": key_id},
- values={
- "from_server": from_server,
- "ts_added_ms": time_now_ms,
- "verify_key": db_binary_type(verify_key.encode()),
- },
+ key_values = []
+ value_values = []
+ invalidations = []
+ for server_name, key_id, fetch_result in verify_keys:
+ key_values.append((server_name, key_id))
+ value_values.append(
+ (
+ from_server,
+ ts_added_ms,
+ fetch_result.valid_until_ts,
+ db_binary_type(fetch_result.verify_key.encode()),
+ )
)
# invalidate takes a tuple corresponding to the params of
# _get_server_verify_key. _get_server_verify_key only takes one
# param, which is itself the 2-tuple (server_name, key_id).
- txn.call_after(
- self._get_server_verify_key.invalidate, ((server_name, key_id),)
- )
-
- return self.runInteraction("store_server_verify_key", _txn)
+ invalidations.append((server_name, key_id))
+
+ def _invalidate(res):
+ f = self._get_server_verify_key.invalidate
+ for i in invalidations:
+ f((i, ))
+ return res
+
+ return self.runInteraction(
+ "store_server_verify_keys",
+ self._simple_upsert_many_txn,
+ table="server_signature_keys",
+ key_names=("server_name", "key_id"),
+ key_values=key_values,
+ value_names=(
+ "from_server",
+ "ts_added_ms",
+ "ts_valid_until_ms",
+ "verify_key",
+ ),
+ value_values=value_values,
+ ).addCallback(_invalidate)
def store_server_keys_json(
self, server_name, key_id, from_server, ts_now_ms, ts_expires_ms, key_json_bytes
diff --git a/synapse/storage/roommember.py b/synapse/storage/roommember.py
index 57df17bcc2..7617913326 100644
--- a/synapse/storage/roommember.py
+++ b/synapse/storage/roommember.py
@@ -142,6 +142,27 @@ class RoomMemberWorkerStore(EventsWorkerStore):
return self.runInteraction("get_room_summary", _get_room_summary_txn)
+ def _get_user_counts_in_room_txn(self, txn, room_id):
+ """
+ Get the user count in a room by membership.
+
+ Args:
+ room_id (str)
+ membership (Membership)
+
+ Returns:
+ Deferred[int]
+ """
+ sql = """
+ SELECT m.membership, count(*) FROM room_memberships as m
+ INNER JOIN current_state_events as c USING(event_id)
+ WHERE c.type = 'm.room.member' AND c.room_id = ?
+ GROUP BY m.membership
+ """
+
+ txn.execute(sql, (room_id,))
+ return {row[0]: row[1] for row in txn}
+
@cached()
def get_invited_rooms_for_user(self, user_id):
""" Get all the rooms the user is invited to
diff --git a/synapse/storage/schema/delta/54/account_validity.sql b/synapse/storage/schema/delta/54/account_validity_with_renewal.sql
index 2357626000..0adb2ad55e 100644
--- a/synapse/storage/schema/delta/54/account_validity.sql
+++ b/synapse/storage/schema/delta/54/account_validity_with_renewal.sql
@@ -13,6 +13,9 @@
* limitations under the License.
*/
+-- We previously changed the schema for this table without renaming the file, which means
+-- that some databases might still be using the old schema. This ensures Synapse uses the
+-- right schema for the table.
DROP TABLE IF EXISTS account_validity;
-- Track what users are in public rooms.
diff --git a/synapse/storage/schema/delta/54/add_validity_to_server_keys.sql b/synapse/storage/schema/delta/54/add_validity_to_server_keys.sql
new file mode 100644
index 0000000000..c01aa9d2d9
--- /dev/null
+++ b/synapse/storage/schema/delta/54/add_validity_to_server_keys.sql
@@ -0,0 +1,23 @@
+/* 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.
+ */
+
+/* When we can use this key until, before we have to refresh it. */
+ALTER TABLE server_signature_keys ADD COLUMN ts_valid_until_ms BIGINT;
+
+UPDATE server_signature_keys SET ts_valid_until_ms = (
+ SELECT MAX(ts_valid_until_ms) FROM server_keys_json skj WHERE
+ skj.server_name = server_signature_keys.server_name AND
+ skj.key_id = server_signature_keys.key_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/stats.sql b/synapse/storage/schema/delta/54/stats.sql
new file mode 100644
index 0000000000..652e58308e
--- /dev/null
+++ b/synapse/storage/schema/delta/54/stats.sql
@@ -0,0 +1,80 @@
+/* Copyright 2018 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.
+ */
+
+CREATE TABLE stats_stream_pos (
+ Lock CHAR(1) NOT NULL DEFAULT 'X' UNIQUE, -- Makes sure this table only has one row.
+ stream_id BIGINT,
+ CHECK (Lock='X')
+);
+
+INSERT INTO stats_stream_pos (stream_id) VALUES (null);
+
+CREATE TABLE user_stats (
+ user_id TEXT NOT NULL,
+ ts BIGINT NOT NULL,
+ bucket_size INT NOT NULL,
+ public_rooms INT NOT NULL,
+ private_rooms INT NOT NULL
+);
+
+CREATE UNIQUE INDEX user_stats_user_ts ON user_stats(user_id, ts);
+
+CREATE TABLE room_stats (
+ room_id TEXT NOT NULL,
+ ts BIGINT NOT NULL,
+ bucket_size INT NOT NULL,
+ current_state_events INT NOT NULL,
+ joined_members INT NOT NULL,
+ invited_members INT NOT NULL,
+ left_members INT NOT NULL,
+ banned_members INT NOT NULL,
+ state_events INT NOT NULL
+);
+
+CREATE UNIQUE INDEX room_stats_room_ts ON room_stats(room_id, ts);
+
+-- cache of current room state; useful for the publicRooms list
+CREATE TABLE room_state (
+ room_id TEXT NOT NULL,
+ join_rules TEXT,
+ history_visibility TEXT,
+ encryption TEXT,
+ name TEXT,
+ topic TEXT,
+ avatar TEXT,
+ canonical_alias TEXT
+ -- get aliases straight from the right table
+);
+
+CREATE UNIQUE INDEX room_state_room ON room_state(room_id);
+
+CREATE TABLE room_stats_earliest_token (
+ room_id TEXT NOT NULL,
+ token BIGINT NOT NULL
+);
+
+CREATE UNIQUE INDEX room_stats_earliest_token_idx ON room_stats_earliest_token(room_id);
+
+-- Set up staging tables
+INSERT INTO background_updates (update_name, progress_json) VALUES
+ ('populate_stats_createtables', '{}');
+
+-- Run through each room and update stats
+INSERT INTO background_updates (update_name, progress_json, depends_on) VALUES
+ ('populate_stats_process_rooms', '{}', 'populate_stats_createtables');
+
+-- Clean up staging tables
+INSERT INTO background_updates (update_name, progress_json, depends_on) VALUES
+ ('populate_stats_cleanup', '{}', 'populate_stats_process_rooms');
diff --git a/synapse/storage/schema/delta/54/stats2.sql b/synapse/storage/schema/delta/54/stats2.sql
new file mode 100644
index 0000000000..3b2d48447f
--- /dev/null
+++ b/synapse/storage/schema/delta/54/stats2.sql
@@ -0,0 +1,28 @@
+/* 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.
+ */
+
+-- This delta file gets run after `54/stats.sql` delta.
+
+-- We want to add some indices to the temporary stats table, so we re-insert
+-- 'populate_stats_createtables' if we are still processing the rooms update.
+INSERT INTO background_updates (update_name, progress_json)
+ SELECT 'populate_stats_createtables', '{}'
+ WHERE
+ 'populate_stats_process_rooms' IN (
+ SELECT update_name FROM background_updates
+ )
+ AND 'populate_stats_createtables' NOT IN ( -- don't insert if already exists
+ SELECT update_name FROM background_updates
+ );
diff --git a/synapse/storage/state_deltas.py b/synapse/storage/state_deltas.py
index 31a0279b18..5fdb442104 100644
--- a/synapse/storage/state_deltas.py
+++ b/synapse/storage/state_deltas.py
@@ -84,10 +84,16 @@ class StateDeltasStore(SQLBaseStore):
"get_current_state_deltas", get_current_state_deltas_txn
)
- def get_max_stream_id_in_current_state_deltas(self):
- return self._simple_select_one_onecol(
+ def _get_max_stream_id_in_current_state_deltas_txn(self, txn):
+ return self._simple_select_one_onecol_txn(
+ txn,
table="current_state_delta_stream",
keyvalues={},
retcol="COALESCE(MAX(stream_id), -1)",
- desc="get_max_stream_id_in_current_state_deltas",
+ )
+
+ def get_max_stream_id_in_current_state_deltas(self):
+ return self.runInteraction(
+ "get_max_stream_id_in_current_state_deltas",
+ self._get_max_stream_id_in_current_state_deltas_txn,
)
diff --git a/synapse/storage/stats.py b/synapse/storage/stats.py
new file mode 100644
index 0000000000..1c0b183a56
--- /dev/null
+++ b/synapse/storage/stats.py
@@ -0,0 +1,452 @@
+# -*- coding: utf-8 -*-
+# Copyright 2018, 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
+
+from twisted.internet import defer
+
+from synapse.api.constants import EventTypes, Membership
+from synapse.storage.prepare_database import get_statements
+from synapse.storage.state_deltas import StateDeltasStore
+from synapse.util.caches.descriptors import cached
+
+logger = logging.getLogger(__name__)
+
+# these fields track absolutes (e.g. total number of rooms on the server)
+ABSOLUTE_STATS_FIELDS = {
+ "room": (
+ "current_state_events",
+ "joined_members",
+ "invited_members",
+ "left_members",
+ "banned_members",
+ "state_events",
+ ),
+ "user": ("public_rooms", "private_rooms"),
+}
+
+TYPE_TO_ROOM = {"room": ("room_stats", "room_id"), "user": ("user_stats", "user_id")}
+
+TEMP_TABLE = "_temp_populate_stats"
+
+
+class StatsStore(StateDeltasStore):
+ def __init__(self, db_conn, hs):
+ super(StatsStore, self).__init__(db_conn, hs)
+
+ self.server_name = hs.hostname
+ self.clock = self.hs.get_clock()
+ self.stats_enabled = hs.config.stats_enabled
+ self.stats_bucket_size = hs.config.stats_bucket_size
+
+ self.register_background_update_handler(
+ "populate_stats_createtables", self._populate_stats_createtables
+ )
+ self.register_background_update_handler(
+ "populate_stats_process_rooms", self._populate_stats_process_rooms
+ )
+ self.register_background_update_handler(
+ "populate_stats_cleanup", self._populate_stats_cleanup
+ )
+
+ @defer.inlineCallbacks
+ def _populate_stats_createtables(self, progress, batch_size):
+
+ if not self.stats_enabled:
+ yield self._end_background_update("populate_stats_createtables")
+ defer.returnValue(1)
+
+ # Get all the rooms that we want to process.
+ def _make_staging_area(txn):
+ # Create the temporary tables
+ stmts = get_statements("""
+ -- We just recreate the table, we'll be reinserting the
+ -- correct entries again later anyway.
+ DROP TABLE IF EXISTS {temp}_rooms;
+
+ CREATE TABLE IF NOT EXISTS {temp}_rooms(
+ room_id TEXT NOT NULL,
+ events BIGINT NOT NULL
+ );
+
+ CREATE INDEX {temp}_rooms_events
+ ON {temp}_rooms(events);
+ CREATE INDEX {temp}_rooms_id
+ ON {temp}_rooms(room_id);
+ """.format(temp=TEMP_TABLE).splitlines())
+
+ for statement in stmts:
+ txn.execute(statement)
+
+ sql = (
+ "CREATE TABLE IF NOT EXISTS "
+ + TEMP_TABLE
+ + "_position(position TEXT NOT NULL)"
+ )
+ txn.execute(sql)
+
+ # Get rooms we want to process from the database, only adding
+ # those that we haven't (i.e. those not in room_stats_earliest_token)
+ sql = """
+ INSERT INTO %s_rooms (room_id, events)
+ SELECT c.room_id, count(*) FROM current_state_events AS c
+ LEFT JOIN room_stats_earliest_token AS t USING (room_id)
+ WHERE t.room_id IS NULL
+ GROUP BY c.room_id
+ """ % (TEMP_TABLE,)
+ txn.execute(sql)
+
+ new_pos = yield self.get_max_stream_id_in_current_state_deltas()
+ yield self.runInteraction("populate_stats_temp_build", _make_staging_area)
+ yield self._simple_insert(TEMP_TABLE + "_position", {"position": new_pos})
+ self.get_earliest_token_for_room_stats.invalidate_all()
+
+ yield self._end_background_update("populate_stats_createtables")
+ defer.returnValue(1)
+
+ @defer.inlineCallbacks
+ def _populate_stats_cleanup(self, progress, batch_size):
+ """
+ Update the user directory stream position, then clean up the old tables.
+ """
+ if not self.stats_enabled:
+ yield self._end_background_update("populate_stats_cleanup")
+ defer.returnValue(1)
+
+ position = yield self._simple_select_one_onecol(
+ TEMP_TABLE + "_position", None, "position"
+ )
+ yield self.update_stats_stream_pos(position)
+
+ def _delete_staging_area(txn):
+ txn.execute("DROP TABLE IF EXISTS " + TEMP_TABLE + "_rooms")
+ txn.execute("DROP TABLE IF EXISTS " + TEMP_TABLE + "_position")
+
+ yield self.runInteraction("populate_stats_cleanup", _delete_staging_area)
+
+ yield self._end_background_update("populate_stats_cleanup")
+ defer.returnValue(1)
+
+ @defer.inlineCallbacks
+ def _populate_stats_process_rooms(self, progress, batch_size):
+
+ if not self.stats_enabled:
+ yield self._end_background_update("populate_stats_process_rooms")
+ defer.returnValue(1)
+
+ # If we don't have progress filed, delete everything.
+ if not progress:
+ yield self.delete_all_stats()
+
+ def _get_next_batch(txn):
+ # Only fetch 250 rooms, so we don't fetch too many at once, even
+ # if those 250 rooms have less than batch_size state events.
+ sql = """
+ SELECT room_id, events FROM %s_rooms
+ ORDER BY events DESC
+ LIMIT 250
+ """ % (
+ TEMP_TABLE,
+ )
+ txn.execute(sql)
+ rooms_to_work_on = txn.fetchall()
+
+ if not rooms_to_work_on:
+ return None
+
+ # Get how many are left to process, so we can give status on how
+ # far we are in processing
+ txn.execute("SELECT COUNT(*) FROM " + TEMP_TABLE + "_rooms")
+ progress["remaining"] = txn.fetchone()[0]
+
+ return rooms_to_work_on
+
+ rooms_to_work_on = yield self.runInteraction(
+ "populate_stats_temp_read", _get_next_batch
+ )
+
+ # No more rooms -- complete the transaction.
+ if not rooms_to_work_on:
+ yield self._end_background_update("populate_stats_process_rooms")
+ defer.returnValue(1)
+
+ logger.info(
+ "Processing the next %d rooms of %d remaining",
+ len(rooms_to_work_on), progress["remaining"],
+ )
+
+ # Number of state events we've processed by going through each room
+ processed_event_count = 0
+
+ for room_id, event_count in rooms_to_work_on:
+
+ current_state_ids = yield self.get_current_state_ids(room_id)
+
+ join_rules_id = current_state_ids.get((EventTypes.JoinRules, ""))
+ history_visibility_id = current_state_ids.get(
+ (EventTypes.RoomHistoryVisibility, "")
+ )
+ encryption_id = current_state_ids.get((EventTypes.RoomEncryption, ""))
+ name_id = current_state_ids.get((EventTypes.Name, ""))
+ topic_id = current_state_ids.get((EventTypes.Topic, ""))
+ avatar_id = current_state_ids.get((EventTypes.RoomAvatar, ""))
+ canonical_alias_id = current_state_ids.get((EventTypes.CanonicalAlias, ""))
+
+ state_events = yield self.get_events([
+ join_rules_id, history_visibility_id, encryption_id, name_id,
+ topic_id, avatar_id, canonical_alias_id,
+ ])
+
+ def _get_or_none(event_id, arg):
+ event = state_events.get(event_id)
+ if event:
+ return event.content.get(arg)
+ return None
+
+ yield self.update_room_state(
+ room_id,
+ {
+ "join_rules": _get_or_none(join_rules_id, "join_rule"),
+ "history_visibility": _get_or_none(
+ history_visibility_id, "history_visibility"
+ ),
+ "encryption": _get_or_none(encryption_id, "algorithm"),
+ "name": _get_or_none(name_id, "name"),
+ "topic": _get_or_none(topic_id, "topic"),
+ "avatar": _get_or_none(avatar_id, "url"),
+ "canonical_alias": _get_or_none(canonical_alias_id, "alias"),
+ },
+ )
+
+ now = self.hs.get_reactor().seconds()
+
+ # quantise time to the nearest bucket
+ now = (now // self.stats_bucket_size) * self.stats_bucket_size
+
+ def _fetch_data(txn):
+
+ # Get the current token of the room
+ current_token = self._get_max_stream_id_in_current_state_deltas_txn(txn)
+
+ current_state_events = len(current_state_ids)
+
+ membership_counts = self._get_user_counts_in_room_txn(txn, room_id)
+
+ total_state_events = self._get_total_state_event_counts_txn(
+ txn, room_id
+ )
+
+ self._update_stats_txn(
+ txn,
+ "room",
+ room_id,
+ now,
+ {
+ "bucket_size": self.stats_bucket_size,
+ "current_state_events": current_state_events,
+ "joined_members": membership_counts.get(Membership.JOIN, 0),
+ "invited_members": membership_counts.get(Membership.INVITE, 0),
+ "left_members": membership_counts.get(Membership.LEAVE, 0),
+ "banned_members": membership_counts.get(Membership.BAN, 0),
+ "state_events": total_state_events,
+ },
+ )
+ self._simple_insert_txn(
+ txn,
+ "room_stats_earliest_token",
+ {"room_id": room_id, "token": current_token},
+ )
+
+ # We've finished a room. Delete it from the table.
+ self._simple_delete_one_txn(
+ txn, TEMP_TABLE + "_rooms", {"room_id": room_id},
+ )
+
+ yield self.runInteraction("update_room_stats", _fetch_data)
+
+ # Update the remaining counter.
+ progress["remaining"] -= 1
+ yield self.runInteraction(
+ "populate_stats",
+ self._background_update_progress_txn,
+ "populate_stats_process_rooms",
+ progress,
+ )
+
+ processed_event_count += event_count
+
+ if processed_event_count > batch_size:
+ # Don't process any more rooms, we've hit our batch size.
+ defer.returnValue(processed_event_count)
+
+ defer.returnValue(processed_event_count)
+
+ def delete_all_stats(self):
+ """
+ Delete all statistics records.
+ """
+
+ def _delete_all_stats_txn(txn):
+ txn.execute("DELETE FROM room_state")
+ txn.execute("DELETE FROM room_stats")
+ txn.execute("DELETE FROM room_stats_earliest_token")
+ txn.execute("DELETE FROM user_stats")
+
+ return self.runInteraction("delete_all_stats", _delete_all_stats_txn)
+
+ def get_stats_stream_pos(self):
+ return self._simple_select_one_onecol(
+ table="stats_stream_pos",
+ keyvalues={},
+ retcol="stream_id",
+ desc="stats_stream_pos",
+ )
+
+ def update_stats_stream_pos(self, stream_id):
+ return self._simple_update_one(
+ table="stats_stream_pos",
+ keyvalues={},
+ updatevalues={"stream_id": stream_id},
+ desc="update_stats_stream_pos",
+ )
+
+ def update_room_state(self, room_id, fields):
+ """
+ Args:
+ room_id (str)
+ fields (dict[str:Any])
+ """
+ return self._simple_upsert(
+ table="room_state",
+ keyvalues={"room_id": room_id},
+ values=fields,
+ desc="update_room_state",
+ )
+
+ def get_deltas_for_room(self, room_id, start, size=100):
+ """
+ Get statistics deltas for a given room.
+
+ Args:
+ room_id (str)
+ start (int): Pagination start. Number of entries, not timestamp.
+ size (int): How many entries to return.
+
+ Returns:
+ Deferred[list[dict]], where the dict has the keys of
+ ABSOLUTE_STATS_FIELDS["room"] and "ts".
+ """
+ return self._simple_select_list_paginate(
+ "room_stats",
+ {"room_id": room_id},
+ "ts",
+ start,
+ size,
+ retcols=(list(ABSOLUTE_STATS_FIELDS["room"]) + ["ts"]),
+ order_direction="DESC",
+ )
+
+ def get_all_room_state(self):
+ return self._simple_select_list(
+ "room_state", None, retcols=("name", "topic", "canonical_alias")
+ )
+
+ @cached()
+ def get_earliest_token_for_room_stats(self, room_id):
+ """
+ Fetch the "earliest token". This is used by the room stats delta
+ processor to ignore deltas that have been processed between the
+ start of the background task and any particular room's stats
+ being calculated.
+
+ Returns:
+ Deferred[int]
+ """
+ return self._simple_select_one_onecol(
+ "room_stats_earliest_token",
+ {"room_id": room_id},
+ retcol="token",
+ allow_none=True,
+ )
+
+ def update_stats(self, stats_type, stats_id, ts, fields):
+ table, id_col = TYPE_TO_ROOM[stats_type]
+ return self._simple_upsert(
+ table=table,
+ keyvalues={id_col: stats_id, "ts": ts},
+ values=fields,
+ desc="update_stats",
+ )
+
+ def _update_stats_txn(self, txn, stats_type, stats_id, ts, fields):
+ table, id_col = TYPE_TO_ROOM[stats_type]
+ return self._simple_upsert_txn(
+ txn, table=table, keyvalues={id_col: stats_id, "ts": ts}, values=fields
+ )
+
+ def update_stats_delta(self, ts, stats_type, stats_id, field, value):
+ def _update_stats_delta(txn):
+ table, id_col = TYPE_TO_ROOM[stats_type]
+
+ sql = (
+ "SELECT * FROM %s"
+ " WHERE %s=? and ts=("
+ " SELECT MAX(ts) FROM %s"
+ " WHERE %s=?"
+ ")"
+ ) % (table, id_col, table, id_col)
+ txn.execute(sql, (stats_id, stats_id))
+ rows = self.cursor_to_dict(txn)
+ if len(rows) == 0:
+ # silently skip as we don't have anything to apply a delta to yet.
+ # this tries to minimise any race between the initial sync and
+ # subsequent deltas arriving.
+ return
+
+ current_ts = ts
+ latest_ts = rows[0]["ts"]
+ if current_ts < latest_ts:
+ # This one is in the past, but we're just encountering it now.
+ # Mark it as part of the current bucket.
+ current_ts = latest_ts
+ elif ts != latest_ts:
+ # we have to copy our absolute counters over to the new entry.
+ values = {
+ key: rows[0][key] for key in ABSOLUTE_STATS_FIELDS[stats_type]
+ }
+ values[id_col] = stats_id
+ values["ts"] = ts
+ values["bucket_size"] = self.stats_bucket_size
+
+ self._simple_insert_txn(txn, table=table, values=values)
+
+ # actually update the new value
+ if stats_type in ABSOLUTE_STATS_FIELDS[stats_type]:
+ self._simple_update_txn(
+ txn,
+ table=table,
+ keyvalues={id_col: stats_id, "ts": current_ts},
+ updatevalues={field: value},
+ )
+ else:
+ sql = ("UPDATE %s SET %s=%s+? WHERE %s=? AND ts=?") % (
+ table,
+ field,
+ field,
+ id_col,
+ )
+ txn.execute(sql, (value, stats_id, current_ts))
+
+ return self.runInteraction("update_stats_delta", _update_stats_delta)
|
