2014-10-15 05:04:55 -04:00
|
|
|
# -*- coding: utf-8 -*-
|
2016-01-06 23:26:29 -05:00
|
|
|
# Copyright 2014-2016 OpenMarket Ltd
|
2014-10-15 05:04:55 -04:00
|
|
|
#
|
|
|
|
# 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.
|
|
|
|
|
2015-08-11 12:59:32 -04:00
|
|
|
from ._base import SQLBaseStore
|
2016-03-23 07:37:58 -04:00
|
|
|
from synapse.util.caches.descriptors import cached, cachedList
|
2016-03-22 14:22:52 -04:00
|
|
|
from synapse.util.caches import intern_string
|
2016-09-01 09:31:26 -04:00
|
|
|
from synapse.storage.engines import PostgresEngine
|
2014-10-15 05:04:55 -04:00
|
|
|
|
2015-03-20 09:52:56 -04:00
|
|
|
from twisted.internet import defer
|
|
|
|
|
2015-01-06 06:18:12 -05:00
|
|
|
import logging
|
|
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
2014-10-15 05:04:55 -04:00
|
|
|
|
2016-09-02 05:41:38 -04:00
|
|
|
MAX_STATE_DELTA_HOPS = 100
|
|
|
|
|
|
|
|
|
2014-10-15 05:04:55 -04:00
|
|
|
class StateStore(SQLBaseStore):
|
2014-11-12 09:33:34 -05:00
|
|
|
""" Keeps track of the state at a given event.
|
|
|
|
|
|
|
|
This is done by the concept of `state groups`. Every event is a assigned
|
|
|
|
a state group (identified by an arbitrary string), which references a
|
|
|
|
collection of state events. The current state of an event is then the
|
|
|
|
collection of state events referenced by the event's state group.
|
|
|
|
|
|
|
|
Hence, every change in the current state causes a new state group to be
|
|
|
|
generated. However, if no change happens (e.g., if we get a message event
|
|
|
|
with only one parent it inherits the state group from its parent.)
|
|
|
|
|
|
|
|
There are three tables:
|
|
|
|
* `state_groups`: Stores group name, first event with in the group and
|
|
|
|
room id.
|
|
|
|
* `event_to_state_groups`: Maps events to state groups.
|
|
|
|
* `state_groups_state`: Maps state group to state events.
|
|
|
|
"""
|
2014-10-15 05:04:55 -04:00
|
|
|
|
2016-09-05 04:34:24 -04:00
|
|
|
STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
|
|
|
|
|
|
|
|
def __init__(self, hs):
|
|
|
|
super(StateStore, self).__init__(hs)
|
|
|
|
self.register_background_update_handler(
|
|
|
|
self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME,
|
|
|
|
self._background_deduplicate_state,
|
|
|
|
)
|
|
|
|
|
2015-05-13 06:13:31 -04:00
|
|
|
@defer.inlineCallbacks
|
2016-08-25 08:28:31 -04:00
|
|
|
def get_state_groups_ids(self, room_id, event_ids):
|
2015-08-07 13:15:30 -04:00
|
|
|
if not event_ids:
|
|
|
|
defer.returnValue({})
|
2014-11-12 09:33:34 -05:00
|
|
|
|
2015-08-07 13:15:30 -04:00
|
|
|
event_to_groups = yield self._get_state_group_for_events(
|
2015-10-12 10:06:14 -04:00
|
|
|
event_ids,
|
2015-08-07 13:15:30 -04:00
|
|
|
)
|
2014-11-12 09:33:34 -05:00
|
|
|
|
2015-08-07 13:15:30 -04:00
|
|
|
groups = set(event_to_groups.values())
|
|
|
|
group_to_state = yield self._get_state_for_groups(groups)
|
2015-05-13 06:29:03 -04:00
|
|
|
|
2016-08-25 08:28:31 -04:00
|
|
|
defer.returnValue(group_to_state)
|
|
|
|
|
|
|
|
@defer.inlineCallbacks
|
|
|
|
def get_state_groups(self, room_id, event_ids):
|
|
|
|
""" Get the state groups for the given list of event_ids
|
|
|
|
|
|
|
|
The return value is a dict mapping group names to lists of events.
|
|
|
|
"""
|
|
|
|
if not event_ids:
|
|
|
|
defer.returnValue({})
|
|
|
|
|
|
|
|
group_to_ids = yield self.get_state_groups_ids(room_id, event_ids)
|
|
|
|
|
|
|
|
state_event_map = yield self.get_events(
|
|
|
|
[
|
|
|
|
ev_id for group_ids in group_to_ids.values()
|
|
|
|
for ev_id in group_ids.values()
|
|
|
|
],
|
|
|
|
get_prev_content=False
|
|
|
|
)
|
|
|
|
|
2015-08-05 10:06:51 -04:00
|
|
|
defer.returnValue({
|
2016-08-25 08:28:31 -04:00
|
|
|
group: [
|
|
|
|
state_event_map[v] for v in event_id_map.values() if v in state_event_map
|
|
|
|
]
|
|
|
|
for group, event_id_map in group_to_ids.items()
|
2015-08-05 10:06:51 -04:00
|
|
|
})
|
2015-06-03 09:45:55 -04:00
|
|
|
|
2016-08-31 05:09:46 -04:00
|
|
|
def _have_persisted_state_group_txn(self, txn, state_group):
|
|
|
|
txn.execute(
|
2016-08-31 11:25:57 -04:00
|
|
|
"SELECT count(*) FROM state_groups WHERE id = ?",
|
2016-08-31 05:09:46 -04:00
|
|
|
(state_group,)
|
|
|
|
)
|
|
|
|
row = txn.fetchone()
|
|
|
|
return row and row[0]
|
|
|
|
|
2015-06-25 12:18:19 -04:00
|
|
|
def _store_mult_state_groups_txn(self, txn, events_and_contexts):
|
|
|
|
state_groups = {}
|
|
|
|
for event, context in events_and_contexts:
|
2016-03-30 07:55:02 -04:00
|
|
|
if event.internal_metadata.is_outlier():
|
|
|
|
continue
|
|
|
|
|
2016-08-25 12:32:22 -04:00
|
|
|
if context.current_state_ids is None:
|
2015-06-25 12:18:19 -04:00
|
|
|
continue
|
2014-12-05 11:20:48 -05:00
|
|
|
|
2016-08-31 05:09:46 -04:00
|
|
|
state_groups[event.event_id] = context.state_group
|
|
|
|
|
|
|
|
if self._have_persisted_state_group_txn(txn, context.state_group):
|
2015-06-25 12:18:19 -04:00
|
|
|
continue
|
|
|
|
|
2015-04-07 07:05:36 -04:00
|
|
|
self._simple_insert_txn(
|
2014-10-15 05:04:55 -04:00
|
|
|
txn,
|
|
|
|
table="state_groups",
|
|
|
|
values={
|
2016-08-31 05:09:46 -04:00
|
|
|
"id": context.state_group,
|
2014-10-15 05:04:55 -04:00
|
|
|
"room_id": event.room_id,
|
|
|
|
"event_id": event.event_id,
|
2014-11-12 06:22:51 -05:00
|
|
|
},
|
2014-10-15 05:04:55 -04:00
|
|
|
)
|
|
|
|
|
2016-09-05 05:41:27 -04:00
|
|
|
# We persist as a delta if we can, while also ensuring the chain
|
|
|
|
# of deltas isn't tooo long, as otherwise read performance degrades.
|
2016-09-01 09:31:26 -04:00
|
|
|
if context.prev_group:
|
2016-09-02 05:41:38 -04:00
|
|
|
potential_hops = self._count_state_group_hops_txn(
|
|
|
|
txn, context.prev_group
|
2016-09-01 09:31:26 -04:00
|
|
|
)
|
2016-09-05 05:41:27 -04:00
|
|
|
if context.prev_group and potential_hops < MAX_STATE_DELTA_HOPS:
|
|
|
|
self._simple_insert_txn(
|
|
|
|
txn,
|
|
|
|
table="state_group_edges",
|
|
|
|
values={
|
|
|
|
"state_group": context.state_group,
|
|
|
|
"prev_state_group": context.prev_group,
|
|
|
|
},
|
|
|
|
)
|
2016-09-02 05:41:38 -04:00
|
|
|
|
2016-09-05 05:41:27 -04:00
|
|
|
self._simple_insert_many_txn(
|
|
|
|
txn,
|
|
|
|
table="state_groups_state",
|
|
|
|
values=[
|
|
|
|
{
|
|
|
|
"state_group": context.state_group,
|
|
|
|
"room_id": event.room_id,
|
|
|
|
"type": key[0],
|
|
|
|
"state_key": key[1],
|
|
|
|
"event_id": state_id,
|
|
|
|
}
|
|
|
|
for key, state_id in context.delta_ids.items()
|
|
|
|
],
|
|
|
|
)
|
2016-09-01 09:31:26 -04:00
|
|
|
else:
|
|
|
|
self._simple_insert_many_txn(
|
|
|
|
txn,
|
|
|
|
table="state_groups_state",
|
|
|
|
values=[
|
|
|
|
{
|
|
|
|
"state_group": context.state_group,
|
|
|
|
"room_id": event.room_id,
|
|
|
|
"type": key[0],
|
|
|
|
"state_key": key[1],
|
|
|
|
"event_id": state_id,
|
|
|
|
}
|
2016-09-05 05:41:27 -04:00
|
|
|
for key, state_id in context.current_state_ids.items()
|
2016-09-01 09:31:26 -04:00
|
|
|
],
|
|
|
|
)
|
2014-10-15 05:04:55 -04:00
|
|
|
|
2015-06-25 12:18:19 -04:00
|
|
|
self._simple_insert_many_txn(
|
2014-10-15 05:04:55 -04:00
|
|
|
txn,
|
|
|
|
table="event_to_state_groups",
|
2015-06-25 12:18:19 -04:00
|
|
|
values=[
|
|
|
|
{
|
2016-03-30 07:55:02 -04:00
|
|
|
"state_group": state_group_id,
|
|
|
|
"event_id": event_id,
|
2015-06-25 12:18:19 -04:00
|
|
|
}
|
2016-03-30 07:55:02 -04:00
|
|
|
for event_id, state_group_id in state_groups.items()
|
2015-06-25 12:18:19 -04:00
|
|
|
],
|
2014-10-15 05:04:55 -04:00
|
|
|
)
|
2015-03-19 11:59:48 -04:00
|
|
|
|
2016-09-02 05:41:38 -04:00
|
|
|
def _count_state_group_hops_txn(self, txn, state_group):
|
2016-09-05 05:41:27 -04:00
|
|
|
"""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.
|
|
|
|
"""
|
2016-09-02 05:41:38 -04:00
|
|
|
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:
|
2016-09-05 09:50:36 -04:00
|
|
|
# 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)
|
2016-09-02 05:41:38 -04:00
|
|
|
next_group = state_group
|
|
|
|
count = 0
|
|
|
|
|
|
|
|
while next_group:
|
|
|
|
next_group = self._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
|
|
|
|
|
2015-03-20 09:52:56 -04:00
|
|
|
@defer.inlineCallbacks
|
|
|
|
def get_current_state(self, room_id, event_type=None, state_key=""):
|
2015-05-21 06:13:19 -04:00
|
|
|
if event_type and state_key is not None:
|
|
|
|
result = yield self.get_current_state_for_key(
|
|
|
|
room_id, event_type, state_key
|
|
|
|
)
|
|
|
|
defer.returnValue(result)
|
|
|
|
|
2015-04-30 12:12:52 -04:00
|
|
|
def f(txn):
|
|
|
|
sql = (
|
2015-04-30 13:49:26 -04:00
|
|
|
"SELECT event_id FROM current_state_events"
|
|
|
|
" WHERE room_id = ? "
|
2015-04-30 12:12:52 -04:00
|
|
|
)
|
|
|
|
|
|
|
|
if event_type and state_key is not None:
|
2015-04-30 13:49:26 -04:00
|
|
|
sql += " AND type = ? AND state_key = ? "
|
2015-04-30 12:12:52 -04:00
|
|
|
args = (room_id, event_type, state_key)
|
|
|
|
elif event_type:
|
2015-04-30 13:49:26 -04:00
|
|
|
sql += " AND type = ?"
|
2015-04-30 12:12:52 -04:00
|
|
|
args = (room_id, event_type)
|
|
|
|
else:
|
|
|
|
args = (room_id, )
|
2015-03-20 09:52:56 -04:00
|
|
|
|
2015-04-30 12:12:52 -04:00
|
|
|
txn.execute(sql, args)
|
2015-05-15 10:20:05 -04:00
|
|
|
results = txn.fetchall()
|
2015-03-20 09:52:56 -04:00
|
|
|
|
2015-05-15 10:20:05 -04:00
|
|
|
return [r[0] for r in results]
|
2015-03-20 09:52:56 -04:00
|
|
|
|
2015-05-15 10:20:05 -04:00
|
|
|
event_ids = yield self.runInteraction("get_current_state", f)
|
2015-05-15 10:33:01 -04:00
|
|
|
events = yield self._get_events(event_ids, get_prev_content=False)
|
2015-03-20 09:52:56 -04:00
|
|
|
defer.returnValue(events)
|
|
|
|
|
2016-03-23 07:37:58 -04:00
|
|
|
@defer.inlineCallbacks
|
2015-05-21 06:13:19 -04:00
|
|
|
def get_current_state_for_key(self, room_id, event_type, state_key):
|
2016-03-23 12:13:05 -04:00
|
|
|
event_ids = yield self._get_current_state_for_key(room_id, event_type, state_key)
|
2016-03-23 07:37:58 -04:00
|
|
|
events = yield self._get_events(event_ids, get_prev_content=False)
|
|
|
|
defer.returnValue(events)
|
|
|
|
|
|
|
|
@cached(num_args=3)
|
|
|
|
def _get_current_state_for_key(self, room_id, event_type, state_key):
|
2015-05-21 06:13:19 -04:00
|
|
|
def f(txn):
|
|
|
|
sql = (
|
|
|
|
"SELECT event_id FROM current_state_events"
|
|
|
|
" WHERE room_id = ? AND type = ? AND state_key = ?"
|
|
|
|
)
|
|
|
|
|
|
|
|
args = (room_id, event_type, state_key)
|
|
|
|
txn.execute(sql, args)
|
|
|
|
results = txn.fetchall()
|
|
|
|
return [r[0] for r in results]
|
2016-03-23 07:37:58 -04:00
|
|
|
return self.runInteraction("get_current_state_for_key", f)
|
2015-05-21 06:13:19 -04:00
|
|
|
|
2016-08-19 06:59:29 -04:00
|
|
|
@cached(num_args=2, max_entries=1000)
|
2016-04-19 12:22:03 -04:00
|
|
|
def _get_state_group_from_group(self, group, types):
|
|
|
|
raise NotImplementedError()
|
|
|
|
|
|
|
|
@cachedList(cached_method_name="_get_state_group_from_group",
|
|
|
|
list_name="groups", num_args=2, inlineCallbacks=True)
|
2016-02-10 08:24:42 -05:00
|
|
|
def _get_state_groups_from_groups(self, groups, types):
|
2016-03-22 07:59:31 -04:00
|
|
|
"""Returns dictionary state_group -> (dict of (type, state_key) -> event id)
|
2015-08-10 10:01:06 -04:00
|
|
|
"""
|
2016-04-19 12:22:03 -04:00
|
|
|
results = {}
|
|
|
|
|
2016-02-10 08:24:42 -05:00
|
|
|
chunks = [groups[i:i + 100] for i in xrange(0, len(groups), 100)]
|
|
|
|
for chunk in chunks:
|
2016-04-19 12:22:03 -04:00
|
|
|
res = yield self.runInteraction(
|
2016-02-10 08:24:42 -05:00
|
|
|
"_get_state_groups_from_groups",
|
2016-09-05 04:34:24 -04:00
|
|
|
self._get_state_groups_from_groups_txn, chunk, types,
|
2016-02-10 08:24:42 -05:00
|
|
|
)
|
2016-04-19 12:22:03 -04:00
|
|
|
results.update(res)
|
|
|
|
|
|
|
|
defer.returnValue(results)
|
2015-08-07 13:15:30 -04:00
|
|
|
|
2016-09-05 04:34:24 -04:00
|
|
|
def _get_state_groups_from_groups_txn(self, txn, groups, types=None):
|
|
|
|
if types is not None:
|
|
|
|
where_clause = "AND (%s)" % (
|
|
|
|
" OR ".join(["(type = ? AND state_key = ?)"] * len(types)),
|
|
|
|
)
|
|
|
|
else:
|
|
|
|
where_clause = ""
|
|
|
|
|
|
|
|
results = {group: {} for group in groups}
|
|
|
|
if isinstance(self.database_engine, PostgresEngine):
|
2016-09-08 04:38:54 -04:00
|
|
|
# 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")
|
|
|
|
|
2016-09-05 05:41:27 -04:00
|
|
|
# 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.
|
2016-09-07 09:53:19 -04:00
|
|
|
# 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.
|
2016-09-05 04:34:24 -04:00
|
|
|
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
|
|
|
|
)
|
2016-09-07 09:22:22 -04:00
|
|
|
SELECT type, state_key, last_value(event_id) OVER (
|
|
|
|
PARTITION BY type, state_key ORDER BY state_group ASC
|
2016-09-07 09:39:01 -04:00
|
|
|
ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
|
2016-09-07 09:22:22 -04:00
|
|
|
) AS event_id FROM state_groups_state
|
|
|
|
WHERE state_group IN (
|
|
|
|
SELECT state_group FROM state
|
2016-09-05 04:34:24 -04:00
|
|
|
)
|
|
|
|
%s;
|
|
|
|
""") % (where_clause,)
|
|
|
|
|
|
|
|
for group in groups:
|
|
|
|
args = [group]
|
|
|
|
if types is not None:
|
|
|
|
args.extend([i for typ in types for i in typ])
|
|
|
|
|
|
|
|
txn.execute(sql, args)
|
|
|
|
rows = self.cursor_to_dict(txn)
|
|
|
|
for row in rows:
|
|
|
|
key = (row["type"], row["state_key"])
|
|
|
|
results[group][key] = row["event_id"]
|
|
|
|
else:
|
2016-09-05 09:50:36 -04:00
|
|
|
# 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)
|
2016-09-05 04:34:24 -04:00
|
|
|
for group in groups:
|
|
|
|
group_tree = [group]
|
|
|
|
next_group = group
|
|
|
|
|
|
|
|
while next_group:
|
|
|
|
next_group = self._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:
|
|
|
|
group_tree.append(next_group)
|
|
|
|
|
|
|
|
sql = ("""
|
|
|
|
SELECT type, state_key, event_id FROM state_groups_state
|
|
|
|
INNER JOIN (
|
|
|
|
SELECT type, state_key, max(state_group) as state_group
|
|
|
|
FROM state_groups_state
|
|
|
|
WHERE state_group IN (%s) %s
|
|
|
|
GROUP BY type, state_key
|
|
|
|
) USING (type, state_key, state_group);
|
|
|
|
""") % (",".join("?" for _ in group_tree), where_clause,)
|
|
|
|
|
|
|
|
args = list(group_tree)
|
|
|
|
if types is not None:
|
|
|
|
args.extend([i for typ in types for i in typ])
|
|
|
|
|
|
|
|
txn.execute(sql, args)
|
|
|
|
rows = self.cursor_to_dict(txn)
|
|
|
|
for row in rows:
|
|
|
|
key = (row["type"], row["state_key"])
|
|
|
|
results[group][key] = row["event_id"]
|
|
|
|
|
|
|
|
return results
|
|
|
|
|
2015-08-04 04:32:23 -04:00
|
|
|
@defer.inlineCallbacks
|
2015-10-12 10:06:14 -04:00
|
|
|
def get_state_for_events(self, event_ids, types):
|
2015-08-04 06:08:07 -04:00
|
|
|
"""Given a list of event_ids and type tuples, return a list of state
|
|
|
|
dicts for each event. The state dicts will only have the type/state_keys
|
|
|
|
that are in the `types` list.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
event_ids (list)
|
|
|
|
types (list): List of (type, state_key) tuples which are used to
|
|
|
|
filter the state fetched. `state_key` may be None, which matches
|
|
|
|
any `state_key`
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
deferred: A list of dicts corresponding to the event_ids given.
|
|
|
|
The dicts are mappings from (type, state_key) -> state_events
|
|
|
|
"""
|
2015-08-07 13:15:30 -04:00
|
|
|
event_to_groups = yield self._get_state_group_for_events(
|
2015-10-12 10:06:14 -04:00
|
|
|
event_ids,
|
2015-08-07 13:15:30 -04:00
|
|
|
)
|
2015-08-04 04:32:23 -04:00
|
|
|
|
2015-08-07 13:15:30 -04:00
|
|
|
groups = set(event_to_groups.values())
|
2015-08-13 12:11:30 -04:00
|
|
|
group_to_state = yield self._get_state_for_groups(groups, types)
|
2015-07-02 11:20:10 -04:00
|
|
|
|
2016-08-25 08:28:31 -04:00
|
|
|
state_event_map = yield self.get_events(
|
|
|
|
[ev_id for sd in group_to_state.values() for ev_id in sd.values()],
|
|
|
|
get_prev_content=False
|
|
|
|
)
|
|
|
|
|
2015-08-04 04:32:23 -04:00
|
|
|
event_to_state = {
|
2016-08-25 08:28:31 -04:00
|
|
|
event_id: {
|
|
|
|
k: state_event_map[v]
|
|
|
|
for k, v in group_to_state[group].items()
|
|
|
|
if v in state_event_map
|
|
|
|
}
|
2015-08-07 13:15:30 -04:00
|
|
|
for event_id, group in event_to_groups.items()
|
2015-07-02 11:20:10 -04:00
|
|
|
}
|
|
|
|
|
2015-08-13 12:11:30 -04:00
|
|
|
defer.returnValue({event: event_to_state[event] for event in event_ids})
|
2015-07-02 11:20:10 -04:00
|
|
|
|
2016-08-25 13:59:44 -04:00
|
|
|
@defer.inlineCallbacks
|
|
|
|
def get_state_ids_for_events(self, event_ids, types):
|
|
|
|
event_to_groups = yield self._get_state_group_for_events(
|
|
|
|
event_ids,
|
|
|
|
)
|
|
|
|
|
|
|
|
groups = set(event_to_groups.values())
|
|
|
|
group_to_state = yield self._get_state_for_groups(groups, types)
|
|
|
|
|
|
|
|
event_to_state = {
|
|
|
|
event_id: group_to_state[group]
|
|
|
|
for event_id, group in event_to_groups.items()
|
|
|
|
}
|
|
|
|
|
|
|
|
defer.returnValue({event: event_to_state[event] for event in event_ids})
|
|
|
|
|
2015-11-10 13:27:23 -05:00
|
|
|
@defer.inlineCallbacks
|
|
|
|
def get_state_for_event(self, event_id, types=None):
|
|
|
|
"""
|
|
|
|
Get the state dict corresponding to a particular event
|
|
|
|
|
2016-04-01 11:08:59 -04:00
|
|
|
Args:
|
|
|
|
event_id(str): event whose state should be returned
|
|
|
|
types(list[(str, str)]|None): List of (type, state_key) tuples
|
|
|
|
which are used to filter the state fetched. May be None, which
|
|
|
|
matches any key
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
A deferred dict from (type, state_key) -> state_event
|
2015-11-10 13:27:23 -05:00
|
|
|
"""
|
|
|
|
state_map = yield self.get_state_for_events([event_id], types)
|
|
|
|
defer.returnValue(state_map[event_id])
|
|
|
|
|
2016-08-25 13:59:44 -04:00
|
|
|
@defer.inlineCallbacks
|
|
|
|
def get_state_ids_for_event(self, event_id, types=None):
|
|
|
|
"""
|
|
|
|
Get the state dict corresponding to a particular event
|
|
|
|
|
|
|
|
Args:
|
|
|
|
event_id(str): event whose state should be returned
|
|
|
|
types(list[(str, str)]|None): List of (type, state_key) tuples
|
|
|
|
which are used to filter the state fetched. May be None, which
|
|
|
|
matches any key
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
A deferred dict from (type, state_key) -> state_event
|
|
|
|
"""
|
|
|
|
state_map = yield self.get_state_ids_for_events([event_id], types)
|
|
|
|
defer.returnValue(state_map[event_id])
|
|
|
|
|
2016-08-19 06:59:29 -04:00
|
|
|
@cached(num_args=2, max_entries=10000)
|
2015-08-05 10:06:51 -04:00
|
|
|
def _get_state_group_for_event(self, room_id, event_id):
|
|
|
|
return self._simple_select_one_onecol(
|
|
|
|
table="event_to_state_groups",
|
|
|
|
keyvalues={
|
|
|
|
"event_id": event_id,
|
|
|
|
},
|
|
|
|
retcol="state_group",
|
|
|
|
allow_none=True,
|
|
|
|
desc="_get_state_group_for_event",
|
|
|
|
)
|
|
|
|
|
2016-04-06 08:08:05 -04:00
|
|
|
@cachedList(cached_method_name="_get_state_group_for_event",
|
|
|
|
list_name="event_ids", num_args=1, inlineCallbacks=True)
|
2015-10-12 10:06:14 -04:00
|
|
|
def _get_state_group_for_events(self, event_ids):
|
2015-08-10 10:01:06 -04:00
|
|
|
"""Returns mapping event_id -> state_group
|
|
|
|
"""
|
2016-02-10 07:57:50 -05:00
|
|
|
rows = yield self._simple_select_many_batch(
|
|
|
|
table="event_to_state_groups",
|
|
|
|
column="event_id",
|
|
|
|
iterable=event_ids,
|
|
|
|
keyvalues={},
|
|
|
|
retcols=("event_id", "state_group",),
|
|
|
|
desc="_get_state_group_for_events",
|
|
|
|
)
|
2015-08-07 13:15:30 -04:00
|
|
|
|
2016-03-23 12:29:46 -04:00
|
|
|
defer.returnValue({row["event_id"]: row["state_group"] for row in rows})
|
2015-08-07 13:15:30 -04:00
|
|
|
|
2015-08-12 12:06:21 -04:00
|
|
|
def _get_some_state_from_cache(self, group, types):
|
2015-08-10 10:01:06 -04:00
|
|
|
"""Checks if group is in cache. See `_get_state_for_groups`
|
2015-08-11 04:12:41 -04:00
|
|
|
|
2015-08-11 06:40:40 -04:00
|
|
|
Returns 3-tuple (`state_dict`, `missing_types`, `got_all`).
|
|
|
|
`missing_types` is the list of types that aren't in the cache for that
|
2015-08-12 12:06:21 -04:00
|
|
|
group. `got_all` is a bool indicating if we successfully retrieved all
|
|
|
|
requests state from the cache, if False we need to query the DB for the
|
|
|
|
missing state.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
group: The state group to lookup
|
|
|
|
types (list): List of 2-tuples of the form (`type`, `state_key`),
|
|
|
|
where a `state_key` of `None` matches all state_keys for the
|
|
|
|
`type`.
|
2015-08-10 10:01:06 -04:00
|
|
|
"""
|
2016-03-22 07:59:31 -04:00
|
|
|
is_all, state_dict_ids = self._state_group_cache.get(group)
|
2015-08-05 10:06:51 -04:00
|
|
|
|
|
|
|
type_to_key = {}
|
|
|
|
missing_types = set()
|
2015-08-12 12:06:21 -04:00
|
|
|
for typ, state_key in types:
|
|
|
|
if state_key is None:
|
|
|
|
type_to_key[typ] = None
|
|
|
|
missing_types.add((typ, state_key))
|
|
|
|
else:
|
|
|
|
if type_to_key.get(typ, object()) is not None:
|
|
|
|
type_to_key.setdefault(typ, set()).add(state_key)
|
2015-08-05 10:06:51 -04:00
|
|
|
|
2016-03-22 07:59:31 -04:00
|
|
|
if (typ, state_key) not in state_dict_ids:
|
2015-08-12 12:06:21 -04:00
|
|
|
missing_types.add((typ, state_key))
|
2015-08-07 05:17:38 -04:00
|
|
|
|
2015-08-11 04:12:41 -04:00
|
|
|
sentinel = object()
|
2015-08-05 10:06:51 -04:00
|
|
|
|
2015-08-11 04:12:41 -04:00
|
|
|
def include(typ, state_key):
|
|
|
|
valid_state_keys = type_to_key.get(typ, sentinel)
|
|
|
|
if valid_state_keys is sentinel:
|
|
|
|
return False
|
|
|
|
if valid_state_keys is None:
|
|
|
|
return True
|
|
|
|
if state_key in valid_state_keys:
|
|
|
|
return True
|
|
|
|
return False
|
|
|
|
|
2015-08-11 06:40:40 -04:00
|
|
|
got_all = not (missing_types or types is None)
|
|
|
|
|
2015-08-11 04:12:41 -04:00
|
|
|
return {
|
2016-03-22 07:59:31 -04:00
|
|
|
k: v for k, v in state_dict_ids.items()
|
2015-08-11 04:12:41 -04:00
|
|
|
if include(k[0], k[1])
|
2015-08-11 06:40:40 -04:00
|
|
|
}, missing_types, got_all
|
2015-08-07 13:15:30 -04:00
|
|
|
|
2015-08-12 12:06:21 -04:00
|
|
|
def _get_all_state_from_cache(self, group):
|
|
|
|
"""Checks if group is in cache. See `_get_state_for_groups`
|
|
|
|
|
|
|
|
Returns 2-tuple (`state_dict`, `got_all`). `got_all` is a bool
|
|
|
|
indicating if we successfully retrieved all requests state from the
|
|
|
|
cache, if False we need to query the DB for the missing state.
|
|
|
|
|
|
|
|
Args:
|
|
|
|
group: The state group to lookup
|
|
|
|
"""
|
2016-03-22 07:59:31 -04:00
|
|
|
is_all, state_dict_ids = self._state_group_cache.get(group)
|
|
|
|
|
|
|
|
return state_dict_ids, is_all
|
2015-08-12 12:06:21 -04:00
|
|
|
|
2015-08-07 13:15:30 -04:00
|
|
|
@defer.inlineCallbacks
|
|
|
|
def _get_state_for_groups(self, groups, types=None):
|
2015-08-10 10:01:06 -04:00
|
|
|
"""Given list of groups returns dict of group -> list of state events
|
|
|
|
with matching types. `types` is a list of `(type, state_key)`, where
|
|
|
|
a `state_key` of None matches all state_keys. If `types` is None then
|
|
|
|
all events are returned.
|
|
|
|
"""
|
2016-04-19 12:22:03 -04:00
|
|
|
if types:
|
|
|
|
types = frozenset(types)
|
2015-08-07 13:15:30 -04:00
|
|
|
results = {}
|
2016-02-10 08:24:42 -05:00
|
|
|
missing_groups = []
|
2015-08-12 12:06:21 -04:00
|
|
|
if types is not None:
|
|
|
|
for group in set(groups):
|
2016-03-22 07:59:31 -04:00
|
|
|
state_dict_ids, missing_types, got_all = self._get_some_state_from_cache(
|
2015-08-12 12:06:21 -04:00
|
|
|
group, types
|
|
|
|
)
|
2016-03-22 07:59:31 -04:00
|
|
|
results[group] = state_dict_ids
|
2015-08-12 12:06:21 -04:00
|
|
|
|
|
|
|
if not got_all:
|
2016-02-10 08:24:42 -05:00
|
|
|
missing_groups.append(group)
|
2015-08-12 12:06:21 -04:00
|
|
|
else:
|
|
|
|
for group in set(groups):
|
2016-03-22 07:59:31 -04:00
|
|
|
state_dict_ids, got_all = self._get_all_state_from_cache(
|
2015-08-12 12:06:21 -04:00
|
|
|
group
|
|
|
|
)
|
2016-03-22 07:59:31 -04:00
|
|
|
|
|
|
|
results[group] = state_dict_ids
|
2015-08-11 04:12:41 -04:00
|
|
|
|
2015-08-12 12:06:21 -04:00
|
|
|
if not got_all:
|
2016-02-10 08:24:42 -05:00
|
|
|
missing_groups.append(group)
|
2015-08-07 13:15:30 -04:00
|
|
|
|
2016-03-22 07:59:31 -04:00
|
|
|
if missing_groups:
|
|
|
|
# Okay, so we have some missing_types, lets fetch them.
|
|
|
|
cache_seq_num = self._state_group_cache.sequence
|
2015-08-05 10:06:51 -04:00
|
|
|
|
2016-03-22 07:59:31 -04:00
|
|
|
group_to_state_dict = yield self._get_state_groups_from_groups(
|
|
|
|
missing_groups, types
|
|
|
|
)
|
2015-08-07 13:15:30 -04:00
|
|
|
|
2016-03-22 07:59:31 -04:00
|
|
|
# Now we want to update the cache with all the things we fetched
|
|
|
|
# from the database.
|
|
|
|
for group, group_state_dict in group_to_state_dict.items():
|
|
|
|
if types:
|
|
|
|
# We delibrately put key -> None mappings into the cache to
|
|
|
|
# cache absence of the key, on the assumption that if we've
|
|
|
|
# explicitly asked for some types then we will probably ask
|
|
|
|
# for them again.
|
2016-03-22 14:22:52 -04:00
|
|
|
state_dict = {
|
|
|
|
(intern_string(etype), intern_string(state_key)): None
|
|
|
|
for (etype, state_key) in types
|
|
|
|
}
|
2016-03-22 07:59:31 -04:00
|
|
|
state_dict.update(results[group])
|
|
|
|
results[group] = state_dict
|
|
|
|
else:
|
|
|
|
state_dict = results[group]
|
|
|
|
|
|
|
|
state_dict.update(group_state_dict)
|
|
|
|
|
|
|
|
self._state_group_cache.update(
|
|
|
|
cache_seq_num,
|
|
|
|
key=group,
|
|
|
|
value=state_dict,
|
|
|
|
full=(types is None),
|
|
|
|
)
|
2015-08-05 10:06:51 -04:00
|
|
|
|
2015-08-18 06:44:10 -04:00
|
|
|
# Remove all the entries with None values. The None values were just
|
|
|
|
# used for bookkeeping in the cache.
|
|
|
|
for group, state_dict in results.items():
|
2015-08-17 04:39:12 -04:00
|
|
|
results[group] = {
|
2016-08-25 08:28:31 -04:00
|
|
|
key: event_id
|
2016-03-22 07:59:31 -04:00
|
|
|
for key, event_id in state_dict.items()
|
2016-08-25 08:28:31 -04:00
|
|
|
if event_id
|
2015-08-17 04:39:12 -04:00
|
|
|
}
|
2015-08-07 13:15:30 -04:00
|
|
|
|
|
|
|
defer.returnValue(results)
|
2016-03-30 10:58:20 -04:00
|
|
|
|
2016-08-31 05:09:46 -04:00
|
|
|
def get_next_state_group(self):
|
|
|
|
return self._state_groups_id_gen.get_next()
|
2016-09-05 04:34:24 -04:00
|
|
|
|
|
|
|
@defer.inlineCallbacks
|
|
|
|
def _background_deduplicate_state(self, progress, batch_size):
|
2016-09-05 05:41:27 -04:00
|
|
|
"""This background update will slowly deduplicate state by reencoding
|
|
|
|
them as deltas.
|
|
|
|
"""
|
2016-09-05 04:34:24 -04:00
|
|
|
last_state_group = progress.get("last_state_group", 0)
|
|
|
|
rows_inserted = progress.get("rows_inserted", 0)
|
|
|
|
max_group = progress.get("max_group", None)
|
|
|
|
|
2016-09-05 10:49:57 -04:00
|
|
|
BATCH_SIZE_SCALE_FACTOR = 100
|
|
|
|
|
|
|
|
batch_size = max(1, int(batch_size / BATCH_SIZE_SCALE_FACTOR))
|
|
|
|
|
2016-09-05 04:34:24 -04:00
|
|
|
if max_group is None:
|
|
|
|
rows = yield self._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 xrange(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
|
|
|
|
|
2016-09-05 10:07:23 -04:00
|
|
|
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
|
|
|
|
|
2016-09-05 04:34:24 -04:00
|
|
|
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], types=None
|
|
|
|
)
|
2016-09-05 09:57:14 -04:00
|
|
|
prev_state = prev_state[prev_group]
|
2016-09-05 04:34:24 -04:00
|
|
|
|
|
|
|
curr_state = self._get_state_groups_from_groups_txn(
|
|
|
|
txn, [state_group], types=None
|
|
|
|
)
|
2016-09-05 09:57:14 -04:00
|
|
|
curr_state = curr_state[state_group]
|
2016-09-05 04:34:24 -04:00
|
|
|
|
|
|
|
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
|
|
|
|
}
|
|
|
|
|
2016-09-05 10:07:23 -04:00
|
|
|
self._simple_delete_txn(
|
|
|
|
txn,
|
|
|
|
table="state_group_edges",
|
|
|
|
keyvalues={
|
|
|
|
"state_group": state_group,
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
2016-09-05 04:34:24 -04:00
|
|
|
self._simple_insert_txn(
|
|
|
|
txn,
|
|
|
|
table="state_group_edges",
|
|
|
|
values={
|
|
|
|
"state_group": state_group,
|
|
|
|
"prev_state_group": prev_group,
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
|
|
|
self._simple_delete_txn(
|
|
|
|
txn,
|
|
|
|
table="state_groups_state",
|
|
|
|
keyvalues={
|
|
|
|
"state_group": state_group,
|
|
|
|
}
|
|
|
|
)
|
|
|
|
|
|
|
|
self._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._background_update_progress_txn(
|
|
|
|
txn, self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, progress
|
|
|
|
)
|
|
|
|
|
|
|
|
return False, batch_size
|
|
|
|
|
|
|
|
finished, result = yield self.runInteraction(
|
|
|
|
self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME, reindex_txn
|
|
|
|
)
|
|
|
|
|
|
|
|
if finished:
|
|
|
|
yield self._end_background_update(self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME)
|
|
|
|
|
2016-09-05 10:49:57 -04:00
|
|
|
defer.returnValue(result * BATCH_SIZE_SCALE_FACTOR)
|