diff --git a/synapse/storage/__init__.py b/synapse/storage/__init__.py
index d01d46338a..f8fbd02ceb 100644
--- a/synapse/storage/__init__.py
+++ b/synapse/storage/__init__.py
@@ -124,7 +124,6 @@ class DataStore(RoomMemberStore, RoomStore,
)
self._transaction_id_gen = IdGenerator(db_conn, "sent_transactions", "id")
- self._state_groups_id_gen = IdGenerator(db_conn, "state_groups", "id")
self._access_tokens_id_gen = IdGenerator(db_conn, "access_tokens", "id")
self._event_reports_id_gen = IdGenerator(db_conn, "event_reports", "id")
self._push_rule_id_gen = IdGenerator(db_conn, "push_rules", "id")
diff --git a/synapse/storage/engines/postgres.py b/synapse/storage/engines/postgres.py
index a6ae79dfad..8a0386c1a4 100644
--- a/synapse/storage/engines/postgres.py
+++ b/synapse/storage/engines/postgres.py
@@ -62,3 +62,9 @@ class PostgresEngine(object):
def lock_table(self, txn, table):
txn.execute("LOCK TABLE %s in EXCLUSIVE MODE" % (table,))
+
+ def get_next_state_group_id(self, txn):
+ """Returns an int that can be used as a new state_group ID
+ """
+ txn.execute("SELECT nextval('state_group_id_seq')")
+ return txn.fetchone()[0]
diff --git a/synapse/storage/engines/sqlite3.py b/synapse/storage/engines/sqlite3.py
index 755c9a1f07..60f0fa7fb3 100644
--- a/synapse/storage/engines/sqlite3.py
+++ b/synapse/storage/engines/sqlite3.py
@@ -16,6 +16,7 @@
from synapse.storage.prepare_database import prepare_database
import struct
+import threading
class Sqlite3Engine(object):
@@ -24,6 +25,11 @@ class Sqlite3Engine(object):
def __init__(self, database_module, database_config):
self.module = database_module
+ # The current max state_group, or None if we haven't looked
+ # in the DB yet.
+ self._current_state_group_id = None
+ self._current_state_group_id_lock = threading.Lock()
+
def check_database(self, txn):
pass
@@ -43,6 +49,19 @@ class Sqlite3Engine(object):
def lock_table(self, txn, table):
return
+ def get_next_state_group_id(self, txn):
+ """Returns an int that can be used as a new state_group ID
+ """
+ # We do application locking here since if we're using sqlite then
+ # we are a single process synapse.
+ with self._current_state_group_id_lock:
+ if self._current_state_group_id is None:
+ txn.execute("SELECT COALESCE(max(id), 0) FROM state_groups")
+ self._current_state_group_id = txn.fetchone()[0]
+
+ self._current_state_group_id += 1
+ return self._current_state_group_id
+
# Following functions taken from: https://github.com/coleifer/peewee
diff --git a/synapse/storage/events.py b/synapse/storage/events.py
index 3d5eb9bc02..d5bc423f18 100644
--- a/synapse/storage/events.py
+++ b/synapse/storage/events.py
@@ -739,9 +739,8 @@ class EventsStore(SQLBaseStore):
events_and_contexts=events_and_contexts,
)
- # Insert into the state_groups, state_groups_state, and
- # event_to_state_groups tables.
- self._store_mult_state_groups_txn(txn, events_and_contexts)
+ # Insert into event_to_state_groups.
+ self._store_event_state_mappings_txn(txn, events_and_contexts)
# _store_rejected_events_txn filters out any events which were
# rejected, and returns the filtered list.
@@ -976,10 +975,9 @@ class EventsStore(SQLBaseStore):
# an outlier in the database. We now have some state at that
# so we need to update the state_groups table with that state.
- # insert into the state_group, state_groups_state and
- # event_to_state_groups tables.
+ # insert into event_to_state_groups.
try:
- self._store_mult_state_groups_txn(txn, ((event, context),))
+ self._store_event_state_mappings_txn(txn, ((event, context),))
except Exception:
logger.exception("")
raise
@@ -2049,16 +2047,32 @@ class EventsStore(SQLBaseStore):
)
return self.runInteraction("get_all_new_events", get_all_new_events_txn)
- def delete_old_state(self, room_id, topological_ordering):
- return self.runInteraction(
- "delete_old_state",
- self._delete_old_state_txn, room_id, topological_ordering
- )
+ def purge_history(
+ self, room_id, topological_ordering, delete_local_events,
+ ):
+ """Deletes room history before a certain point
+
+ Args:
+ room_id (str):
+
+ topological_ordering (int):
+ minimum topo ordering to preserve
- def _delete_old_state_txn(self, txn, room_id, topological_ordering):
- """Deletes old room state
+ delete_local_events (bool):
+ if True, we will delete local events as well as remote ones
+ (instead of just marking them as outliers and deleting their
+ state groups).
"""
+ return self.runInteraction(
+ "purge_history",
+ self._purge_history_txn, room_id, topological_ordering,
+ delete_local_events,
+ )
+
+ def _purge_history_txn(
+ self, txn, room_id, topological_ordering, delete_local_events,
+ ):
# Tables that should be pruned:
# event_auth
# event_backward_extremities
@@ -2099,7 +2113,7 @@ class EventsStore(SQLBaseStore):
400, "topological_ordering is greater than forward extremeties"
)
- logger.debug("[purge] looking for events to delete")
+ logger.info("[purge] looking for events to delete")
txn.execute(
"SELECT event_id, state_key FROM events"
@@ -2111,16 +2125,16 @@ class EventsStore(SQLBaseStore):
to_delete = [
(event_id,) for event_id, state_key in event_rows
- if state_key is None and not self.hs.is_mine_id(event_id)
+ if state_key is None and (
+ delete_local_events or not self.hs.is_mine_id(event_id)
+ )
]
logger.info(
- "[purge] found %i events before cutoff, of which %i are remote"
- " non-state events to delete", len(event_rows), len(to_delete))
-
- for event_id, state_key in event_rows:
- txn.call_after(self._get_state_group_for_event.invalidate, (event_id,))
+ "[purge] found %i events before cutoff, of which %i can be deleted",
+ len(event_rows), len(to_delete),
+ )
- logger.debug("[purge] Finding new backward extremities")
+ logger.info("[purge] Finding new backward extremities")
# We calculate the new entries for the backward extremeties by finding
# all events that point to events that are to be purged
@@ -2134,7 +2148,7 @@ class EventsStore(SQLBaseStore):
)
new_backwards_extrems = txn.fetchall()
- logger.debug("[purge] replacing backward extremities: %r", new_backwards_extrems)
+ logger.info("[purge] replacing backward extremities: %r", new_backwards_extrems)
txn.execute(
"DELETE FROM event_backward_extremities WHERE room_id = ?",
@@ -2150,7 +2164,7 @@ class EventsStore(SQLBaseStore):
]
)
- logger.debug("[purge] finding redundant state groups")
+ logger.info("[purge] finding redundant state groups")
# Get all state groups that are only referenced by events that are
# to be deleted.
@@ -2167,15 +2181,15 @@ class EventsStore(SQLBaseStore):
)
state_rows = txn.fetchall()
- logger.debug("[purge] found %i redundant state groups", len(state_rows))
+ logger.info("[purge] found %i redundant state groups", len(state_rows))
# make a set of the redundant state groups, so that we can look them up
# efficiently
state_groups_to_delete = set([sg for sg, in state_rows])
# Now we get all the state groups that rely on these state groups
- logger.debug("[purge] finding state groups which depend on redundant"
- " state groups")
+ logger.info("[purge] finding state groups which depend on redundant"
+ " state groups")
remaining_state_groups = []
for i in xrange(0, len(state_rows), 100):
chunk = [sg for sg, in state_rows[i:i + 100]]
@@ -2200,7 +2214,7 @@ class EventsStore(SQLBaseStore):
# Now we turn the state groups that reference to-be-deleted state
# groups to non delta versions.
for sg in remaining_state_groups:
- logger.debug("[purge] de-delta-ing remaining state group %s", sg)
+ logger.info("[purge] de-delta-ing remaining state group %s", sg)
curr_state = self._get_state_groups_from_groups_txn(
txn, [sg], types=None
)
@@ -2237,7 +2251,7 @@ class EventsStore(SQLBaseStore):
],
)
- logger.debug("[purge] removing redundant state groups")
+ logger.info("[purge] removing redundant state groups")
txn.executemany(
"DELETE FROM state_groups_state WHERE state_group = ?",
state_rows
@@ -2247,18 +2261,15 @@ class EventsStore(SQLBaseStore):
state_rows
)
- # Delete all non-state
- logger.debug("[purge] removing events from event_to_state_groups")
+ logger.info("[purge] removing events from event_to_state_groups")
txn.executemany(
"DELETE FROM event_to_state_groups WHERE event_id = ?",
[(event_id,) for event_id, _ in event_rows]
)
-
- logger.debug("[purge] updating room_depth")
- txn.execute(
- "UPDATE room_depth SET min_depth = ? WHERE room_id = ?",
- (topological_ordering, room_id,)
- )
+ for event_id, _ in event_rows:
+ txn.call_after(self._get_state_group_for_event.invalidate, (
+ event_id,
+ ))
# Delete all remote non-state events
for table in (
@@ -2276,7 +2287,8 @@ class EventsStore(SQLBaseStore):
"event_signatures",
"rejections",
):
- logger.debug("[purge] removing remote non-state events from %s", table)
+ logger.info("[purge] removing remote non-state events from %s",
+ table)
txn.executemany(
"DELETE FROM %s WHERE event_id = ?" % (table,),
@@ -2284,16 +2296,30 @@ class EventsStore(SQLBaseStore):
)
# Mark all state and own events as outliers
- logger.debug("[purge] marking remaining events as outliers")
+ logger.info("[purge] marking remaining events as outliers")
txn.executemany(
"UPDATE events SET outlier = ?"
" WHERE event_id = ?",
[
(True, event_id,) for event_id, state_key in event_rows
- if state_key is not None or self.hs.is_mine_id(event_id)
+ if state_key is not None or (
+ not delete_local_events and self.hs.is_mine_id(event_id)
+ )
]
)
+ # synapse tries to take out an exclusive lock on room_depth whenever it
+ # persists events (because upsert), and once we run this update, we
+ # will block that for the rest of our transaction.
+ #
+ # So, let's stick it at the end so that we don't block event
+ # persistence.
+ logger.info("[purge] updating room_depth")
+ txn.execute(
+ "UPDATE room_depth SET min_depth = ? WHERE room_id = ?",
+ (topological_ordering, room_id,)
+ )
+
logger.info("[purge] done")
@defer.inlineCallbacks
diff --git a/synapse/storage/room.py b/synapse/storage/room.py
index cf2c4dae39..fff6652e05 100644
--- a/synapse/storage/room.py
+++ b/synapse/storage/room.py
@@ -16,11 +16,9 @@
from twisted.internet import defer
from synapse.api.errors import StoreError
+from synapse.storage.search import SearchStore
from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
-from ._base import SQLBaseStore
-from .engines import PostgresEngine, Sqlite3Engine
-
import collections
import logging
import ujson as json
@@ -40,7 +38,7 @@ RatelimitOverride = collections.namedtuple(
)
-class RoomStore(SQLBaseStore):
+class RoomStore(SearchStore):
@defer.inlineCallbacks
def store_room(self, room_id, room_creator_user_id, is_public):
@@ -263,8 +261,8 @@ class RoomStore(SQLBaseStore):
},
)
- self._store_event_search_txn(
- txn, event, "content.topic", event.content["topic"]
+ self.store_event_search_txn(
+ txn, event, "content.topic", event.content["topic"],
)
def _store_room_name_txn(self, txn, event):
@@ -279,14 +277,14 @@ class RoomStore(SQLBaseStore):
}
)
- self._store_event_search_txn(
- txn, event, "content.name", event.content["name"]
+ self.store_event_search_txn(
+ txn, event, "content.name", event.content["name"],
)
def _store_room_message_txn(self, txn, event):
if hasattr(event, "content") and "body" in event.content:
- self._store_event_search_txn(
- txn, event, "content.body", event.content["body"]
+ self.store_event_search_txn(
+ txn, event, "content.body", event.content["body"],
)
def _store_history_visibility_txn(self, txn, event):
@@ -308,31 +306,6 @@ class RoomStore(SQLBaseStore):
event.content[key]
))
- def _store_event_search_txn(self, txn, event, key, value):
- if isinstance(self.database_engine, PostgresEngine):
- sql = (
- "INSERT INTO event_search"
- " (event_id, room_id, key, vector, stream_ordering, origin_server_ts)"
- " VALUES (?,?,?,to_tsvector('english', ?),?,?)"
- )
- txn.execute(
- sql,
- (
- event.event_id, event.room_id, key, value,
- event.internal_metadata.stream_ordering,
- event.origin_server_ts,
- )
- )
- elif isinstance(self.database_engine, Sqlite3Engine):
- sql = (
- "INSERT INTO event_search (event_id, room_id, key, value)"
- " VALUES (?,?,?,?)"
- )
- txn.execute(sql, (event.event_id, event.room_id, key, value,))
- else:
- # This should be unreachable.
- raise Exception("Unrecognized database engine")
-
def add_event_report(self, room_id, event_id, user_id, reason, content,
received_ts):
next_id = self._event_reports_id_gen.get_next()
diff --git a/synapse/storage/schema/delta/47/state_group_seq.py b/synapse/storage/schema/delta/47/state_group_seq.py
new file mode 100644
index 0000000000..f6766501d2
--- /dev/null
+++ b/synapse/storage/schema/delta/47/state_group_seq.py
@@ -0,0 +1,37 @@
+# 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.
+
+from synapse.storage.engines import PostgresEngine
+
+
+def run_create(cur, database_engine, *args, **kwargs):
+ if isinstance(database_engine, PostgresEngine):
+ # if we already have some state groups, we want to start making new
+ # ones with a higher id.
+ cur.execute("SELECT max(id) FROM state_groups")
+ row = cur.fetchone()
+
+ if row[0] is None:
+ start_val = 1
+ else:
+ start_val = row[0] + 1
+
+ cur.execute(
+ "CREATE SEQUENCE state_group_id_seq START WITH %s",
+ (start_val, ),
+ )
+
+
+def run_upgrade(*args, **kwargs):
+ pass
diff --git a/synapse/storage/search.py b/synapse/storage/search.py
index 479b04c636..f1ac9ba0fd 100644
--- a/synapse/storage/search.py
+++ b/synapse/storage/search.py
@@ -13,19 +13,25 @@
# See the License for the specific language governing permissions and
# limitations under the License.
+from collections import namedtuple
+import logging
+import re
+import ujson as json
+
from twisted.internet import defer
from .background_updates import BackgroundUpdateStore
from synapse.api.errors import SynapseError
from synapse.storage.engines import PostgresEngine, Sqlite3Engine
-import logging
-import re
-import ujson as json
-
logger = logging.getLogger(__name__)
+SearchEntry = namedtuple('SearchEntry', [
+ 'key', 'value', 'event_id', 'room_id', 'stream_ordering',
+ 'origin_server_ts',
+])
+
class SearchStore(BackgroundUpdateStore):
@@ -49,16 +55,17 @@ class SearchStore(BackgroundUpdateStore):
@defer.inlineCallbacks
def _background_reindex_search(self, progress, batch_size):
+ # we work through the events table from highest stream id to lowest
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
TYPES = ["m.room.name", "m.room.message", "m.room.topic"]
def reindex_search_txn(txn):
sql = (
- "SELECT stream_ordering, event_id, room_id, type, content FROM events"
+ "SELECT stream_ordering, event_id, room_id, type, content, "
+ " origin_server_ts FROM events"
" WHERE ? <= stream_ordering AND stream_ordering < ?"
" AND (%s)"
" ORDER BY stream_ordering DESC"
@@ -67,6 +74,10 @@ class SearchStore(BackgroundUpdateStore):
txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
+ # we could stream straight from the results into
+ # store_search_entries_txn with a generator function, but that
+ # would mean having two cursors open on the database at once.
+ # Instead we just build a list of results.
rows = self.cursor_to_dict(txn)
if not rows:
return 0
@@ -79,6 +90,8 @@ class SearchStore(BackgroundUpdateStore):
event_id = row["event_id"]
room_id = row["room_id"]
etype = row["type"]
+ stream_ordering = row["stream_ordering"]
+ origin_server_ts = row["origin_server_ts"]
try:
content = json.loads(row["content"])
except Exception:
@@ -93,6 +106,8 @@ class SearchStore(BackgroundUpdateStore):
elif etype == "m.room.name":
key = "content.name"
value = content["name"]
+ else:
+ raise Exception("unexpected event type %s" % etype)
except (KeyError, AttributeError):
# If the event is missing a necessary field then
# skip over it.
@@ -103,25 +118,16 @@ class SearchStore(BackgroundUpdateStore):
# then skip over it
continue
- event_search_rows.append((event_id, room_id, key, value))
+ event_search_rows.append(SearchEntry(
+ key=key,
+ value=value,
+ event_id=event_id,
+ room_id=room_id,
+ stream_ordering=stream_ordering,
+ origin_server_ts=origin_server_ts,
+ ))
- if isinstance(self.database_engine, PostgresEngine):
- sql = (
- "INSERT INTO event_search (event_id, room_id, key, vector)"
- " VALUES (?,?,?,to_tsvector('english', ?))"
- )
- elif isinstance(self.database_engine, Sqlite3Engine):
- sql = (
- "INSERT INTO event_search (event_id, room_id, key, value)"
- " VALUES (?,?,?,?)"
- )
- else:
- # This should be unreachable.
- raise Exception("Unrecognized database engine")
-
- for index in range(0, len(event_search_rows), INSERT_CLUMP_SIZE):
- clump = event_search_rows[index:index + INSERT_CLUMP_SIZE]
- txn.executemany(sql, clump)
+ self.store_search_entries_txn(txn, event_search_rows)
progress = {
"target_min_stream_id_inclusive": target_min_stream_id,
@@ -242,6 +248,62 @@ class SearchStore(BackgroundUpdateStore):
defer.returnValue(num_rows)
+ def store_event_search_txn(self, txn, event, key, value):
+ """Add event to the search table
+
+ Args:
+ txn (cursor):
+ event (EventBase):
+ key (str):
+ value (str):
+ """
+ self.store_search_entries_txn(
+ txn,
+ (SearchEntry(
+ key=key,
+ value=value,
+ event_id=event.event_id,
+ room_id=event.room_id,
+ stream_ordering=event.internal_metadata.stream_ordering,
+ origin_server_ts=event.origin_server_ts,
+ ),),
+ )
+
+ def store_search_entries_txn(self, txn, entries):
+ """Add entries to the search table
+
+ Args:
+ txn (cursor):
+ entries (iterable[SearchEntry]):
+ entries to be added to the table
+ """
+ if isinstance(self.database_engine, PostgresEngine):
+ sql = (
+ "INSERT INTO event_search"
+ " (event_id, room_id, key, vector, stream_ordering, origin_server_ts)"
+ " VALUES (?,?,?,to_tsvector('english', ?),?,?)"
+ )
+
+ args = ((
+ entry.event_id, entry.room_id, entry.key, entry.value,
+ entry.stream_ordering, entry.origin_server_ts,
+ ) for entry in entries)
+
+ txn.executemany(sql, args)
+ elif isinstance(self.database_engine, Sqlite3Engine):
+ sql = (
+ "INSERT INTO event_search (event_id, room_id, key, value)"
+ " VALUES (?,?,?,?)"
+ )
+ args = ((
+ entry.event_id, entry.room_id, entry.key, entry.value,
+ ) for entry in entries)
+
+ txn.executemany(sql, args)
+ else:
+ # This should be unreachable.
+ raise Exception("Unrecognized database engine")
+
@defer.inlineCallbacks
def search_msgs(self, room_ids, search_term, keys):
"""Performs a full text search over events with given keys.
diff --git a/synapse/storage/state.py b/synapse/storage/state.py
index 360e3e4355..adb48df73e 100644
--- a/synapse/storage/state.py
+++ b/synapse/storage/state.py
@@ -42,11 +42,8 @@ class _GetStateGroupDelta(namedtuple("_GetStateGroupDelta", ("prev_group", "delt
return len(self.delta_ids) if self.delta_ids else 0
-class StateGroupReadStore(SQLBaseStore):
- """The read-only parts of StateGroupStore
-
- None of these functions write to the state tables, so are suitable for
- including in the SlavedStores.
+class StateGroupWorkerStore(SQLBaseStore):
+ """The parts of StateGroupStore that can be called from workers.
"""
STATE_GROUP_DEDUPLICATION_UPDATE_NAME = "state_group_state_deduplication"
@@ -54,7 +51,7 @@ class StateGroupReadStore(SQLBaseStore):
CURRENT_STATE_INDEX_UPDATE_NAME = "current_state_members_idx"
def __init__(self, db_conn, hs):
- super(StateGroupReadStore, self).__init__(db_conn, hs)
+ super(StateGroupWorkerStore, self).__init__(db_conn, hs)
self._state_group_cache = DictionaryCache(
"*stateGroupCache*", 100000 * CACHE_SIZE_FACTOR
@@ -549,116 +546,66 @@ class StateGroupReadStore(SQLBaseStore):
defer.returnValue(results)
+ def store_state_group(self, event_id, room_id, prev_group, delta_ids,
+ current_state_ids):
+ """Store a new set of state, returning a newly assigned state group.
-class StateStore(StateGroupReadStore, BackgroundUpdateStore):
- """ 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.
- """
-
- 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, db_conn, hs):
- super(StateStore, self).__init__(db_conn, 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'",
- )
-
- 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]
-
- 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
+ Args:
+ event_id (str): The event ID for which the state was calculated
+ room_id (str)
+ prev_group (int|None): A previous state group for the room, optional.
+ delta_ids (dict|None): The delta between state at `prev_group` and
+ `current_state_ids`, if `prev_group` was given. Same format as
+ `current_state_ids`.
+ current_state_ids (dict): The state to store. Map of (type, state_key)
+ to event_id.
- if context.current_state_ids is None:
+ Returns:
+ Deferred[int]: The state group ID
+ """
+ def _store_state_group_txn(txn):
+ if 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
+ raise Exception("current_state_ids cannot be None")
- # 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
- continue
-
- state_groups[event.event_id] = context.state_group
-
- if self._have_persisted_state_group_txn(txn, context.state_group):
- continue
+ state_group = self.database_engine.get_next_state_group_id(txn)
self._simple_insert_txn(
txn,
table="state_groups",
values={
- "id": context.state_group,
- "room_id": event.room_id,
- "event_id": event.event_id,
+ "id": state_group,
+ "room_id": room_id,
+ "event_id": event_id,
},
)
# 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:
+ if prev_group:
is_in_db = self._simple_select_one_onecol_txn(
txn,
table="state_groups",
- keyvalues={"id": context.prev_group},
+ keyvalues={"id": prev_group},
retcol="id",
allow_none=True,
)
if not is_in_db:
raise Exception(
"Trying to persist state with unpersisted prev_group: %r"
- % (context.prev_group,)
+ % (prev_group,)
)
potential_hops = self._count_state_group_hops_txn(
- txn, context.prev_group
+ txn, prev_group
)
- if context.prev_group and potential_hops < MAX_STATE_DELTA_HOPS:
+ if 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,
+ "state_group": state_group,
+ "prev_state_group": prev_group,
},
)
@@ -667,13 +614,13 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore):
table="state_groups_state",
values=[
{
- "state_group": context.state_group,
- "room_id": event.room_id,
+ "state_group": state_group,
+ "room_id": room_id,
"type": key[0],
"state_key": key[1],
"event_id": state_id,
}
- for key, state_id in context.delta_ids.iteritems()
+ for key, state_id in delta_ids.iteritems()
],
)
else:
@@ -682,13 +629,13 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore):
table="state_groups_state",
values=[
{
- "state_group": context.state_group,
- "room_id": event.room_id,
+ "state_group": state_group,
+ "room_id": room_id,
"type": key[0],
"state_key": key[1],
"event_id": state_id,
}
- for key, state_id in context.current_state_ids.iteritems()
+ for key, state_id in current_state_ids.iteritems()
],
)
@@ -699,11 +646,71 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore):
txn.call_after(
self._state_group_cache.update,
self._state_group_cache.sequence,
- key=context.state_group,
- value=dict(context.current_state_ids),
+ key=state_group,
+ value=dict(current_state_ids),
full=True,
)
+ return state_group
+
+ return self.runInteraction("store_state_group", _store_state_group_txn)
+
+
+class StateStore(StateGroupWorkerStore, BackgroundUpdateStore):
+ """ 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.
+ """
+
+ 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, db_conn, hs):
+ super(StateStore, self).__init__(db_conn, 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'",
+ )
+
+ def _store_event_state_mappings_txn(self, txn, events_and_contexts):
+ state_groups = {}
+ for event, context in events_and_contexts:
+ if event.internal_metadata.is_outlier():
+ 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
+ continue
+
+ state_groups[event.event_id] = context.state_group
+
self._simple_insert_many_txn(
txn,
table="event_to_state_groups",
@@ -763,9 +770,6 @@ class StateStore(StateGroupReadStore, BackgroundUpdateStore):
return count
- def get_next_state_group(self):
- return self._state_groups_id_gen.get_next()
-
@defer.inlineCallbacks
def _background_deduplicate_state(self, progress, batch_size):
"""This background update will slowly deduplicate state by reencoding
|
