MatrixSynapse/synapse/storage/state.py

850 lines
32 KiB
Python
Raw Normal View History

2014-10-15 11:04:55 +02:00
# -*- coding: utf-8 -*-
2016-01-07 05:26:29 +01:00
# Copyright 2014-2016 OpenMarket Ltd
2014-10-15 11:04:55 +02:00
#
# 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.
from ._base import SQLBaseStore
from synapse.util.caches.descriptors import cached, cachedList, cachedInlineCallbacks
2016-03-22 19:22:52 +01:00
from synapse.util.caches import intern_string
from synapse.storage.engines import PostgresEngine
2014-10-15 11:04:55 +02:00
2015-03-20 14:52:56 +01:00
from twisted.internet import defer
2015-01-06 12:18:12 +01:00
import logging
logger = logging.getLogger(__name__)
2014-10-15 11:04:55 +02:00
2016-09-02 11:41:38 +02:00
MAX_STATE_DELTA_HOPS = 100
2014-10-15 11:04:55 +02:00
class StateStore(SQLBaseStore):
""" 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 11:04:55 +02:00
STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
STATE_GROUP_INDEX_UPDATE_NAME = "state_group_state_type_index"
CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx"
def __init__(self, hs):
super(StateStore, self).__init__(hs)
self.register_background_update_handler(
self.STATE_GROUP_DEDUPLICATION_UPDATE_NAME,
self._background_deduplicate_state,
)
self.register_background_update_handler(
self.STATE_GROUP_INDEX_UPDATE_NAME,
self._background_index_state,
)
self.register_background_index_update(
self.CURRENT_STATE_INDEX_UPDATE_NAME,
index_name="current_state_events_member_index",
table="current_state_events",
columns=["state_key"],
where_clause="type='m.room.member'",
)
@cachedInlineCallbacks(max_entries=100000, iterable=True)
def get_current_state_ids(self, room_id):
rows = yield self._simple_select_list(
table="current_state_events",
keyvalues={"room_id": room_id},
retcols=["event_id", "type", "state_key"],
desc="_calculate_state_delta",
)
defer.returnValue({
(r["type"], r["state_key"]): r["event_id"] for r in rows
})
2015-05-13 12:13:31 +02:00
@defer.inlineCallbacks
def get_state_groups_ids(self, room_id, event_ids):
if not event_ids:
defer.returnValue({})
event_to_groups = yield self._get_state_group_for_events(
2015-10-12 16:06:14 +02:00
event_ids,
)
2017-03-24 11:57:02 +01:00
groups = set(event_to_groups.itervalues())
group_to_state = yield self._get_state_for_groups(groups)
2015-05-13 12:29:03 +02:00
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(
[
2017-03-24 11:57:02 +01:00
ev_id for group_ids in group_to_ids.itervalues()
for ev_id in group_ids.itervalues()
],
get_prev_content=False
)
defer.returnValue({
group: [
2017-03-24 11:57:02 +01:00
state_event_map[v] for v in event_id_map.itervalues()
if v in state_event_map
]
2017-03-24 11:57:02 +01:00
for group, event_id_map in group_to_ids.iteritems()
})
2015-06-03 15:45:55 +02:00
def _have_persisted_state_group_txn(self, txn, state_group):
txn.execute(
"SELECT count(*) FROM state_groups WHERE id = ?",
(state_group,)
)
row = txn.fetchone()
return row and row[0]
2015-06-25 18:18:19 +02:00
def _store_mult_state_groups_txn(self, txn, events_and_contexts):
state_groups = {}
for event, context in events_and_contexts:
if event.internal_metadata.is_outlier():
continue
if context.current_state_ids is None:
# AFAIK, this can never happen
logger.error(
"Non-outlier event %s had current_state_ids==None",
event.event_id)
continue
# if the event was rejected, just give it the same state as its
# predecessor.
if context.rejected:
state_groups[event.event_id] = context.prev_group
2015-06-25 18:18:19 +02:00
continue
state_groups[event.event_id] = context.state_group
if self._have_persisted_state_group_txn(txn, context.state_group):
2015-06-25 18:18:19 +02:00
continue
self._simple_insert_txn(
2014-10-15 11:04:55 +02:00
txn,
table="state_groups",
values={
"id": context.state_group,
2014-10-15 11:04:55 +02:00
"room_id": event.room_id,
"event_id": event.event_id,
},
2014-10-15 11:04:55 +02:00
)
2016-09-05 11:41:27 +02:00
# We persist as a delta if we can, while also ensuring the chain
# of deltas isn't tooo long, as otherwise read performance degrades.
if context.prev_group:
2016-09-02 11:41:38 +02:00
potential_hops = self._count_state_group_hops_txn(
txn, context.prev_group
)
2016-09-05 11:41:27 +02:00
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 11:41:38 +02:00
2016-09-05 11:41:27 +02:00
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,
}
2017-03-24 11:57:02 +01:00
for key, state_id in context.delta_ids.iteritems()
2016-09-05 11:41:27 +02:00
],
)
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,
}
2017-03-24 11:57:02 +01:00
for key, state_id in context.current_state_ids.iteritems()
],
)
2014-10-15 11:04:55 +02:00
2015-06-25 18:18:19 +02:00
self._simple_insert_many_txn(
2014-10-15 11:04:55 +02:00
txn,
table="event_to_state_groups",
2015-06-25 18:18:19 +02:00
values=[
{
"state_group": state_group_id,
"event_id": event_id,
2015-06-25 18:18:19 +02:00
}
2017-03-24 11:57:02 +01:00
for event_id, state_group_id in state_groups.iteritems()
2015-06-25 18:18:19 +02:00
],
2014-10-15 11:04:55 +02:00
)
2016-09-02 11:41:38 +02:00
def _count_state_group_hops_txn(self, txn, state_group):
2016-09-05 11:41:27 +02:00
"""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 11:41:38 +02:00
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:
# 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 11:41:38 +02:00
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
2017-01-16 16:08:17 +01:00
@cached(num_args=2, max_entries=100000, iterable=True)
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)
def _get_state_groups_from_groups(self, groups, types):
"""Returns dictionary state_group -> (dict of (type, state_key) -> event id)
2015-08-10 16:01:06 +02:00
"""
results = {}
chunks = [groups[i:i + 100] for i in xrange(0, len(groups), 100)]
for chunk in chunks:
res = yield self.runInteraction(
"_get_state_groups_from_groups",
self._get_state_groups_from_groups_txn, chunk, types,
)
results.update(res)
defer.returnValue(results)
def _get_state_groups_from_groups_txn(self, txn, groups, types=None):
results = {group: {} for group in groups}
2016-09-26 18:16:24 +02:00
if types is not None:
types = list(set(types)) # deduplicate types list
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")
2016-09-05 11:41:27 +02:00
# 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 15:53:19 +02:00
# 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.
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 type, state_key, last_value(event_id) OVER (
PARTITION BY type, state_key ORDER BY state_group ASC
2016-09-07 15:39:01 +02:00
ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
) AS event_id FROM state_groups_state
WHERE state_group IN (
SELECT state_group FROM state
)
%s
""")
# Turns out that postgres doesn't like doing a list of OR's and
# is about 1000x slower, so we just issue a query for each specific
# type seperately.
if types:
clause_to_args = [
(
"AND type = ? AND state_key = ?",
(etype, state_key)
)
for etype, state_key in types
]
else:
# If types is None we fetch all the state, and so just use an
# empty where clause with no extra args.
clause_to_args = [("", [])]
for where_clause, where_args in clause_to_args:
for group in groups:
args = [group]
args.extend(where_args)
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:
if types is not None:
where_clause = "AND (%s)" % (
" OR ".join(["(type = ? AND state_key = ?)"] * len(types)),
)
else:
where_clause = ""
# 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:
2016-09-26 18:16:24 +02:00
# 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
2017-01-05 14:39:43 +01:00
# without the right indices (which we can't add until
2016-09-26 18:16:24 +02:00
# after we finish deduping state, which requires this func)
2016-09-27 10:27:54 +02:00
args = [next_group]
if types:
args.extend(i for typ in types for i in typ)
txn.execute(
"SELECT type, state_key, event_id FROM state_groups_state"
" WHERE state_group = ? %s" % (where_clause,),
args
)
2017-03-24 11:57:02 +01:00
results[group].update(
((typ, state_key), event_id)
for typ, state_key, event_id in txn
2016-09-27 10:27:54 +02:00
if (typ, state_key) not in results[group]
2017-03-24 11:57:02 +01:00
)
2016-09-27 10:27:54 +02:00
# If the lengths match then we must have all the types,
# so no need to go walk further down the tree.
if types is not None and len(results[group]) == len(types):
break
2016-09-26 18:16:24 +02:00
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,
)
return results
@defer.inlineCallbacks
2015-10-12 16:06:14 +02:00
def get_state_for_events(self, event_ids, types):
2015-08-04 12:08:07 +02:00
"""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
"""
event_to_groups = yield self._get_state_group_for_events(
2015-10-12 16:06:14 +02:00
event_ids,
)
2017-03-24 11:57:02 +01:00
groups = set(event_to_groups.itervalues())
2015-08-13 18:11:30 +02:00
group_to_state = yield self._get_state_for_groups(groups, types)
state_event_map = yield self.get_events(
2017-03-24 11:57:02 +01:00
[ev_id for sd in group_to_state.itervalues() for ev_id in sd.itervalues()],
get_prev_content=False
)
event_to_state = {
event_id: {
k: state_event_map[v]
2017-03-24 11:57:02 +01:00
for k, v in group_to_state[group].iteritems()
if v in state_event_map
}
2017-03-24 11:57:02 +01:00
for event_id, group in event_to_groups.iteritems()
}
2015-08-13 18:11:30 +02:00
defer.returnValue({event: event_to_state[event] for event in event_ids})
2016-08-25 19:59:44 +02:00
@defer.inlineCallbacks
def get_state_ids_for_events(self, event_ids, types=None):
"""
Get the state dicts corresponding to a list of events
Args:
event_ids(list(str)): events 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 event_id -> (type, state_key) -> state_event
"""
2016-08-25 19:59:44 +02:00
event_to_groups = yield self._get_state_group_for_events(
event_ids,
)
2017-03-24 11:57:02 +01:00
groups = set(event_to_groups.itervalues())
2016-08-25 19:59:44 +02:00
group_to_state = yield self._get_state_for_groups(groups, types)
event_to_state = {
event_id: group_to_state[group]
2017-03-24 11:57:02 +01:00
for event_id, group in event_to_groups.iteritems()
2016-08-25 19:59:44 +02:00
}
defer.returnValue({event: event_to_state[event] for event in event_ids})
@defer.inlineCallbacks
def get_state_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_for_events([event_id], types)
defer.returnValue(state_map[event_id])
2016-08-25 19:59:44 +02:00
@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])
@cached(num_args=2, max_entries=100000)
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",
)
@cachedList(cached_method_name="_get_state_group_for_event",
list_name="event_ids", num_args=1, inlineCallbacks=True)
2015-10-12 16:06:14 +02:00
def _get_state_group_for_events(self, event_ids):
2015-08-10 16:01:06 +02:00
"""Returns mapping event_id -> state_group
"""
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",
)
defer.returnValue({row["event_id"]: row["state_group"] for row in rows})
def _get_some_state_from_cache(self, group, types):
2015-08-10 16:01:06 +02:00
"""Checks if group is in cache. See `_get_state_for_groups`
2015-08-11 10:12:41 +02:00
2015-08-11 12:40:40 +02:00
Returns 3-tuple (`state_dict`, `missing_types`, `got_all`).
`missing_types` is the list of types that aren't in the cache for that
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 16:01:06 +02:00
"""
is_all, state_dict_ids = self._state_group_cache.get(group)
type_to_key = {}
missing_types = set()
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)
if (typ, state_key) not in state_dict_ids:
missing_types.add((typ, state_key))
2015-08-07 11:17:38 +02:00
2015-08-11 10:12:41 +02:00
sentinel = object()
2015-08-11 10:12:41 +02:00
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 12:40:40 +02:00
got_all = not (missing_types or types is None)
2015-08-11 10:12:41 +02:00
return {
2017-03-24 11:57:02 +01:00
k: v for k, v in state_dict_ids.iteritems()
2015-08-11 10:12:41 +02:00
if include(k[0], k[1])
2015-08-11 12:40:40 +02:00
}, missing_types, got_all
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
"""
is_all, state_dict_ids = self._state_group_cache.get(group)
return state_dict_ids, is_all
@defer.inlineCallbacks
def _get_state_for_groups(self, groups, types=None):
2015-08-10 16:01:06 +02:00
"""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.
"""
if types:
types = frozenset(types)
results = {}
missing_groups = []
if types is not None:
for group in set(groups):
state_dict_ids, missing_types, got_all = self._get_some_state_from_cache(
group, types
)
results[group] = state_dict_ids
if not got_all:
missing_groups.append(group)
else:
for group in set(groups):
state_dict_ids, got_all = self._get_all_state_from_cache(
group
)
results[group] = state_dict_ids
2015-08-11 10:12:41 +02:00
if not got_all:
missing_groups.append(group)
if missing_groups:
# Okay, so we have some missing_types, lets fetch them.
cache_seq_num = self._state_group_cache.sequence
group_to_state_dict = yield self._get_state_groups_from_groups(
missing_groups, types
)
# Now we want to update the cache with all the things we fetched
# from the database.
2017-03-24 11:57:02 +01:00
for group, group_state_dict in group_to_state_dict.iteritems():
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 19:22:52 +01:00
state_dict = {
(intern_string(etype), intern_string(state_key)): None
for (etype, state_key) in types
}
state_dict.update(results[group])
results[group] = state_dict
else:
state_dict = results[group]
2017-03-24 11:57:02 +01:00
state_dict.update(
((intern_string(k[0]), intern_string(k[1])), v)
for k, v in group_state_dict.iteritems()
)
self._state_group_cache.update(
cache_seq_num,
key=group,
value=state_dict,
full=(types is None),
)
# Remove all the entries with None values. The None values were just
# used for bookkeeping in the cache.
2017-03-24 11:57:02 +01:00
for group, state_dict in results.iteritems():
results[group] = {
key: event_id
2017-03-24 11:57:02 +01:00
for key, event_id in state_dict.iteritems()
if event_id
}
defer.returnValue(results)
def get_next_state_group(self):
return self._state_groups_id_gen.get_next()
@defer.inlineCallbacks
def _background_deduplicate_state(self, progress, batch_size):
2016-09-05 11:41:27 +02:00
"""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._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
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], types=None
)
2016-09-05 15:57:14 +02:00
prev_state = prev_state[prev_group]
curr_state = self._get_state_groups_from_groups_txn(
txn, [state_group], types=None
)
2016-09-05 15:57:14 +02:00
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 = {
2017-03-24 11:57:02 +01:00
key: value for key, value in curr_state.iteritems()
if prev_state.get(key, None) != value
}
self._simple_delete_txn(
txn,
table="state_group_edges",
keyvalues={
"state_group": state_group,
}
)
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,
}
2017-03-24 11:57:02 +01:00
for key, state_id in delta_state.iteritems()
],
)
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)
defer.returnValue(result * BATCH_SIZE_SCALE_FACTOR)
@defer.inlineCallbacks
def _background_index_state(self, progress, batch_size):
def reindex_txn(conn):
conn.rollback()
2016-09-14 11:18:30 +02:00
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"
)
2016-09-14 11:18:30 +02:00
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.runWithConnection(reindex_txn)
yield self._end_background_update(self.STATE_GROUP_INDEX_UPDATE_NAME)
defer.returnValue(1)