forked-synapse/synapse/storage/databases/state/bg_updates.py

375 lines
14 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
# Copyright 2014-2016 OpenMarket 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.storage._base import SQLBaseStore
from synapse.storage.database import DatabasePool
from synapse.storage.engines import PostgresEngine
from synapse.storage.state import StateFilter
logger = logging.getLogger(__name__)
MAX_STATE_DELTA_HOPS = 100
class StateGroupBackgroundUpdateStore(SQLBaseStore):
"""Defines functions related to state groups needed to run the state backgroud
updates.
"""
def _count_state_group_hops_txn(self, txn, state_group):
"""Given a state group, count how many hops there are in the tree.
This is used to ensure the delta chains don't get too long.
"""
if isinstance(self.database_engine, PostgresEngine):
sql = """
WITH RECURSIVE state(state_group) AS (
VALUES(?::bigint)
UNION ALL
SELECT prev_state_group FROM state_group_edges e, state s
WHERE s.state_group = e.state_group
)
SELECT count(*) FROM state;
"""
txn.execute(sql, (state_group,))
row = txn.fetchone()
if row and row[0]:
return row[0]
else:
return 0
else:
# We don't use WITH RECURSIVE on sqlite3 as there are distributions
# that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
next_group = state_group
count = 0
while next_group:
next_group = self.db_pool.simple_select_one_onecol_txn(
txn,
table="state_group_edges",
keyvalues={"state_group": next_group},
retcol="prev_state_group",
allow_none=True,
)
if next_group:
count += 1
return count
def _get_state_groups_from_groups_txn(
self, txn, groups, state_filter=StateFilter.all()
):
results = {group: {} for group in groups}
where_clause, where_args = state_filter.make_sql_filter_clause()
# Unless the filter clause is empty, we're going to append it after an
# existing where clause
if where_clause:
where_clause = " AND (%s)" % (where_clause,)
if isinstance(self.database_engine, PostgresEngine):
# Temporarily disable sequential scans in this transaction. This is
# a temporary hack until we can add the right indices in
txn.execute("SET LOCAL enable_seqscan=off")
# The below query walks the state_group tree so that the "state"
# table includes all state_groups in the tree. It then joins
# against `state_groups_state` to fetch the latest state.
# It assumes that previous state groups are always numerically
# lesser.
# The PARTITION is used to get the event_id in the greatest state
# group for the given type, state_key.
# This may return multiple rows per (type, state_key), but last_value
# should be the same.
sql = """
WITH RECURSIVE state(state_group) AS (
VALUES(?::bigint)
UNION ALL
SELECT prev_state_group FROM state_group_edges e, state s
WHERE s.state_group = e.state_group
)
Improve performance of _get_state_groups_from_groups_txn (#7567) The query keeps showing up in my slow query log. This changes the plan under the top-level Sort node from ``` WindowAgg (cost=280335.88..292963.15 rows=561212 width=80) (actual time=138.651..160.562 rows=27112 loops=1) -> Sort (cost=280335.88..281738.91 rows=561212 width=84) (actual time=138.597..140.622 rows=27112 loops=1) Sort Key: state_groups_state.type, state_groups_state.state_key, state_groups_state.state_group Sort Method: quicksort Memory: 4581kB -> Nested Loop (cost=2.83..226745.22 rows=561212 width=84) (actual time=21.548..47.657 rows=27112 loops=1) -> HashAggregate (cost=2.27..3.28 rows=101 width=8) (actual time=21.526..21.535 rows=20 loops=1) Group Key: state.state_group -> CTE Scan on state (cost=0.00..2.02 rows=101 width=8) (actual time=21.280..21.493 rows=20 loops=1) -> Index Scan using state_groups_state_type_idx on state_groups_state (cost=0.56..2189.40 rows=5557 width=84) (actual time=0.005..0.991 rows=1356 loops=20) Index Cond: (state_group = state.state_group) ``` to ``` Nested Loop (cost=2.83..226745.22 rows=561212 width=84) (actual time=24.194..52.834 rows=27112 loops=1) -> HashAggregate (cost=2.27..3.28 rows=101 width=8) (actual time=24.130..24.138 rows=20 loops=1) Group Key: state.state_group -> CTE Scan on state (cost=0.00..2.02 rows=101 width=8) (actual time=23.887..24.113 rows=20 loops=1) -> Index Scan using state_groups_state_type_idx on state_groups_state (cost=0.56..2189.40 rows=5557 width=84) (actual time=0.016..1.159 rows=1356 loops=20) Index Cond: (state_group = state.state_group) ``` This cuts the execution time from ~190ms to ~130ms, i.e. a reduction of ~30%. The full plans are visualised at https://explain.depesz.com/s/WpbT and https://explain.depesz.com/s/KlEk Signed-off-by: Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
2020-06-01 10:23:43 -04:00
SELECT DISTINCT ON (type, state_key)
type, state_key, event_id
FROM state_groups_state
WHERE state_group IN (
SELECT state_group FROM state
Improve performance of _get_state_groups_from_groups_txn (#7567) The query keeps showing up in my slow query log. This changes the plan under the top-level Sort node from ``` WindowAgg (cost=280335.88..292963.15 rows=561212 width=80) (actual time=138.651..160.562 rows=27112 loops=1) -> Sort (cost=280335.88..281738.91 rows=561212 width=84) (actual time=138.597..140.622 rows=27112 loops=1) Sort Key: state_groups_state.type, state_groups_state.state_key, state_groups_state.state_group Sort Method: quicksort Memory: 4581kB -> Nested Loop (cost=2.83..226745.22 rows=561212 width=84) (actual time=21.548..47.657 rows=27112 loops=1) -> HashAggregate (cost=2.27..3.28 rows=101 width=8) (actual time=21.526..21.535 rows=20 loops=1) Group Key: state.state_group -> CTE Scan on state (cost=0.00..2.02 rows=101 width=8) (actual time=21.280..21.493 rows=20 loops=1) -> Index Scan using state_groups_state_type_idx on state_groups_state (cost=0.56..2189.40 rows=5557 width=84) (actual time=0.005..0.991 rows=1356 loops=20) Index Cond: (state_group = state.state_group) ``` to ``` Nested Loop (cost=2.83..226745.22 rows=561212 width=84) (actual time=24.194..52.834 rows=27112 loops=1) -> HashAggregate (cost=2.27..3.28 rows=101 width=8) (actual time=24.130..24.138 rows=20 loops=1) Group Key: state.state_group -> CTE Scan on state (cost=0.00..2.02 rows=101 width=8) (actual time=23.887..24.113 rows=20 loops=1) -> Index Scan using state_groups_state_type_idx on state_groups_state (cost=0.56..2189.40 rows=5557 width=84) (actual time=0.016..1.159 rows=1356 loops=20) Index Cond: (state_group = state.state_group) ``` This cuts the execution time from ~190ms to ~130ms, i.e. a reduction of ~30%. The full plans are visualised at https://explain.depesz.com/s/WpbT and https://explain.depesz.com/s/KlEk Signed-off-by: Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
2020-06-01 10:23:43 -04:00
) %s
ORDER BY type, state_key, state_group DESC
"""
for group in groups:
args = [group]
args.extend(where_args)
Improve performance of _get_state_groups_from_groups_txn (#7567) The query keeps showing up in my slow query log. This changes the plan under the top-level Sort node from ``` WindowAgg (cost=280335.88..292963.15 rows=561212 width=80) (actual time=138.651..160.562 rows=27112 loops=1) -> Sort (cost=280335.88..281738.91 rows=561212 width=84) (actual time=138.597..140.622 rows=27112 loops=1) Sort Key: state_groups_state.type, state_groups_state.state_key, state_groups_state.state_group Sort Method: quicksort Memory: 4581kB -> Nested Loop (cost=2.83..226745.22 rows=561212 width=84) (actual time=21.548..47.657 rows=27112 loops=1) -> HashAggregate (cost=2.27..3.28 rows=101 width=8) (actual time=21.526..21.535 rows=20 loops=1) Group Key: state.state_group -> CTE Scan on state (cost=0.00..2.02 rows=101 width=8) (actual time=21.280..21.493 rows=20 loops=1) -> Index Scan using state_groups_state_type_idx on state_groups_state (cost=0.56..2189.40 rows=5557 width=84) (actual time=0.005..0.991 rows=1356 loops=20) Index Cond: (state_group = state.state_group) ``` to ``` Nested Loop (cost=2.83..226745.22 rows=561212 width=84) (actual time=24.194..52.834 rows=27112 loops=1) -> HashAggregate (cost=2.27..3.28 rows=101 width=8) (actual time=24.130..24.138 rows=20 loops=1) Group Key: state.state_group -> CTE Scan on state (cost=0.00..2.02 rows=101 width=8) (actual time=23.887..24.113 rows=20 loops=1) -> Index Scan using state_groups_state_type_idx on state_groups_state (cost=0.56..2189.40 rows=5557 width=84) (actual time=0.016..1.159 rows=1356 loops=20) Index Cond: (state_group = state.state_group) ``` This cuts the execution time from ~190ms to ~130ms, i.e. a reduction of ~30%. The full plans are visualised at https://explain.depesz.com/s/WpbT and https://explain.depesz.com/s/KlEk Signed-off-by: Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>
2020-06-01 10:23:43 -04:00
txn.execute(sql % (where_clause,), args)
for row in txn:
typ, state_key, event_id = row
key = (typ, state_key)
results[group][key] = event_id
else:
max_entries_returned = state_filter.max_entries_returned()
# We don't use WITH RECURSIVE on sqlite3 as there are distributions
# that ship with an sqlite3 version that doesn't support it (e.g. wheezy)
for group in groups:
next_group = group
while next_group:
# We did this before by getting the list of group ids, and
# then passing that list to sqlite to get latest event for
# each (type, state_key). However, that was terribly slow
# without the right indices (which we can't add until
# after we finish deduping state, which requires this func)
args = [next_group]
args.extend(where_args)
txn.execute(
"SELECT type, state_key, event_id FROM state_groups_state"
" WHERE state_group = ? " + where_clause,
args,
)
results[group].update(
((typ, state_key), event_id)
for typ, state_key, event_id in txn
if (typ, state_key) not in results[group]
)
# If the number of entries in the (type,state_key)->event_id dict
# matches the number of (type,state_keys) types we were searching
# for, then we must have found them all, so no need to go walk
# further down the tree... UNLESS our types filter contained
# wildcards (i.e. Nones) in which case we have to do an exhaustive
# search
if (
max_entries_returned is not None
and len(results[group]) == max_entries_returned
):
break
next_group = self.db_pool.simple_select_one_onecol_txn(
txn,
table="state_group_edges",
keyvalues={"state_group": next_group},
retcol="prev_state_group",
allow_none=True,
)
return results
class StateBackgroundUpdateStore(StateGroupBackgroundUpdateStore):
STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
STATE_GROUPS_ROOM_INDEX_UPDATE_NAME = "state_groups_room_id_idx"
def __init__(self, database: DatabasePool, db_conn, hs):
super(StateBackgroundUpdateStore, self).__init__(database, db_conn, hs)
self.db_pool.updates.register_background_update_handler(
self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME,
self._background_deduplicate_state,
)
self.db_pool.updates.register_background_update_handler(
self.STATE_GROUP_INDEX_UPDATE_NAME, self._background_index_state
)
self.db_pool.updates.register_background_index_update(
self.STATE_GROUPS_ROOM_INDEX_UPDATE_NAME,
index_name="state_groups_room_id_idx",
table="state_groups",
columns=["room_id"],
)
@defer.inlineCallbacks
def _background_deduplicate_state(self, progress, batch_size):
"""This background update will slowly deduplicate state by reencoding
them as deltas.
"""
last_state_group = progress.get("last_state_group", 0)
rows_inserted = progress.get("rows_inserted", 0)
max_group = progress.get("max_group", None)
BATCH_SIZE_SCALE_FACTOR = 100
batch_size = max(1, int(batch_size / BATCH_SIZE_SCALE_FACTOR))
if max_group is None:
rows = yield self.db_pool.execute(
"_background_deduplicate_state",
None,
"SELECT coalesce(max(id), 0) FROM state_groups",
)
max_group = rows[0][0]
def reindex_txn(txn):
new_last_state_group = last_state_group
for count in range(batch_size):
txn.execute(
"SELECT id, room_id FROM state_groups"
" WHERE ? < id AND id <= ?"
" ORDER BY id ASC"
" LIMIT 1",
(new_last_state_group, max_group),
)
row = txn.fetchone()
if row:
state_group, room_id = row
if not row or not state_group:
return True, count
txn.execute(
"SELECT state_group FROM state_group_edges"
" WHERE state_group = ?",
(state_group,),
)
# If we reach a point where we've already started inserting
# edges we should stop.
if txn.fetchall():
return True, count
txn.execute(
"SELECT coalesce(max(id), 0) FROM state_groups"
" WHERE id < ? AND room_id = ?",
(state_group, room_id),
)
(prev_group,) = txn.fetchone()
new_last_state_group = state_group
if prev_group:
potential_hops = self._count_state_group_hops_txn(txn, prev_group)
if potential_hops >= MAX_STATE_DELTA_HOPS:
# We want to ensure chains are at most this long,#
# otherwise read performance degrades.
continue
prev_state = self._get_state_groups_from_groups_txn(
txn, [prev_group]
)
prev_state = prev_state[prev_group]
curr_state = self._get_state_groups_from_groups_txn(
txn, [state_group]
)
curr_state = curr_state[state_group]
if not set(prev_state.keys()) - set(curr_state.keys()):
# We can only do a delta if the current has a strict super set
# of keys
delta_state = {
key: value
for key, value in curr_state.items()
if prev_state.get(key, None) != value
}
self.db_pool.simple_delete_txn(
txn,
table="state_group_edges",
keyvalues={"state_group": state_group},
)
self.db_pool.simple_insert_txn(
txn,
table="state_group_edges",
values={
"state_group": state_group,
"prev_state_group": prev_group,
},
)
self.db_pool.simple_delete_txn(
txn,
table="state_groups_state",
keyvalues={"state_group": state_group},
)
self.db_pool.simple_insert_many_txn(
txn,
table="state_groups_state",
values=[
{
"state_group": state_group,
"room_id": room_id,
"type": key[0],
"state_key": key[1],
"event_id": state_id,
}
for key, state_id in delta_state.items()
],
)
progress = {
"last_state_group": state_group,
"rows_inserted": rows_inserted + batch_size,
"max_group": max_group,
}
self.db_pool.updates._background_update_progress_txn(
txn, self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, progress
)
return False, batch_size
finished, result = yield self.db_pool.runInteraction(
self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, reindex_txn
)
if finished:
yield self.db_pool.updates._end_background_update(
self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME
)
return result * BATCH_SIZE_SCALE_FACTOR
@defer.inlineCallbacks
def _background_index_state(self, progress, batch_size):
def reindex_txn(conn):
conn.rollback()
if isinstance(self.database_engine, PostgresEngine):
# postgres insists on autocommit for the index
conn.set_session(autocommit=True)
try:
txn = conn.cursor()
txn.execute(
"CREATE INDEX CONCURRENTLY state_groups_state_type_idx"
" ON state_groups_state(state_group, type, state_key)"
)
txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
finally:
conn.set_session(autocommit=False)
else:
txn = conn.cursor()
txn.execute(
"CREATE INDEX state_groups_state_type_idx"
" ON state_groups_state(state_group, type, state_key)"
)
txn.execute("DROP INDEX IF EXISTS state_groups_state_id")
yield self.db_pool.runWithConnection(reindex_txn)
yield self.db_pool.updates._end_background_update(
self.STATE_GROUP_INDEX_UPDATE_NAME
)
return 1