diff options
Diffstat (limited to 'synapse/state/__init__.py')
| -rw-r--r-- | synapse/state/__init__.py | 269 |
1 files changed, 142 insertions, 127 deletions
diff --git a/synapse/state/__init__.py b/synapse/state/__init__.py index 2b0f4c79ee..2fa529fcd0 100644 --- a/synapse/state/__init__.py +++ b/synapse/state/__init__.py @@ -16,6 +16,7 @@ import logging from collections import namedtuple +from typing import Dict, Iterable, List, Optional, Set from six import iteritems, itervalues @@ -27,13 +28,15 @@ from twisted.internet import defer from synapse.api.constants import EventTypes from synapse.api.room_versions import KNOWN_ROOM_VERSIONS, StateResolutionVersions +from synapse.events import EventBase from synapse.events.snapshot import EventContext from synapse.logging.utils import log_function from synapse.state import v1, v2 +from synapse.storage.data_stores.main.events_worker import EventRedactBehaviour +from synapse.types import StateMap from synapse.util.async_helpers import Linearizer -from synapse.util.caches import get_cache_factor_for from synapse.util.caches.expiringcache import ExpiringCache -from synapse.util.metrics import Measure +from synapse.util.metrics import Measure, measure_func logger = logging.getLogger(__name__) @@ -49,7 +52,6 @@ state_groups_histogram = Histogram( KeyStateTuple = namedtuple("KeyStateTuple", ("context", "type", "state_key")) -SIZE_OF_CACHE = 100000 * get_cache_factor_for("state_cache") EVICTION_TIMEOUT_SECONDS = 60 * 60 @@ -103,6 +105,7 @@ class StateHandler(object): def __init__(self, hs): self.clock = hs.get_clock() self.store = hs.get_datastore() + self.state_store = hs.get_storage().state self.hs = hs self._state_resolution_handler = hs.get_state_resolution_handler() @@ -191,24 +194,37 @@ class StateHandler(object): return joined_users @defer.inlineCallbacks - def get_current_hosts_in_room(self, room_id, latest_event_ids=None): - if not latest_event_ids: - latest_event_ids = yield self.store.get_latest_event_ids_in_room(room_id) - logger.debug("calling resolve_state_groups from get_current_hosts_in_room") - entry = yield self.resolve_state_groups_for_events(room_id, latest_event_ids) + def get_current_hosts_in_room(self, room_id): + event_ids = yield self.store.get_latest_event_ids_in_room(room_id) + return (yield self.get_hosts_in_room_at_events(room_id, event_ids)) + + @defer.inlineCallbacks + def get_hosts_in_room_at_events(self, room_id, event_ids): + """Get the hosts that were in a room at the given event ids + + Args: + room_id (str): + event_ids (list[str]): + + Returns: + Deferred[list[str]]: the hosts in the room at the given events + """ + entry = yield self.resolve_state_groups_for_events(room_id, event_ids) joined_hosts = yield self.store.get_joined_hosts(room_id, entry) return joined_hosts @defer.inlineCallbacks - def compute_event_context(self, event, old_state=None): + def compute_event_context( + self, event: EventBase, old_state: Optional[Iterable[EventBase]] = None + ): """Build an EventContext structure for the event. This works out what the current state should be for the event, and generates a new state group if necessary. Args: - event (synapse.events.EventBase): - old_state (dict|None): The state at the event if it can't be + event: + old_state: The state at the event if it can't be calculated from existing events. This is normally only specified when receiving an event from federation where we don't have the prev events for, e.g. when backfilling. @@ -220,6 +236,9 @@ class StateHandler(object): # If this is an outlier, then we know it shouldn't have any current # state. Certainly store.get_current_state won't return any, and # persisting the event won't store the state group. + + # FIXME: why do we populate current_state_ids? I thought the point was + # that we weren't supposed to have any state for outliers? if old_state: prev_state_ids = {(s.type, s.state_key): s.event_id for s in old_state} if event.is_state(): @@ -236,114 +255,105 @@ class StateHandler(object): # group for it. context = EventContext.with_state( state_group=None, + state_group_before_event=None, current_state_ids=current_state_ids, prev_state_ids=prev_state_ids, ) return context + # + # first of all, figure out the state before the event + # + if old_state: - # We already have the state, so we don't need to calculate it. - # Let's just correctly fill out the context and create a - # new state group for it. - - prev_state_ids = {(s.type, s.state_key): s.event_id for s in old_state} - - if event.is_state(): - key = (event.type, event.state_key) - if key in prev_state_ids: - replaces = prev_state_ids[key] - if replaces != event.event_id: # Paranoia check - event.unsigned["replaces_state"] = replaces - current_state_ids = dict(prev_state_ids) - current_state_ids[key] = event.event_id - else: - current_state_ids = prev_state_ids + # if we're given the state before the event, then we use that + state_ids_before_event = { + (s.type, s.state_key): s.event_id for s in old_state + } + state_group_before_event = None + state_group_before_event_prev_group = None + deltas_to_state_group_before_event = None - state_group = yield self.store.store_state_group( - event.event_id, - event.room_id, - prev_group=None, - delta_ids=None, - current_state_ids=current_state_ids, - ) + else: + # otherwise, we'll need to resolve the state across the prev_events. + logger.debug("calling resolve_state_groups from compute_event_context") - context = EventContext.with_state( - state_group=state_group, - current_state_ids=current_state_ids, - prev_state_ids=prev_state_ids, + entry = yield self.resolve_state_groups_for_events( + event.room_id, event.prev_event_ids() ) - return context + state_ids_before_event = entry.state + state_group_before_event = entry.state_group + state_group_before_event_prev_group = entry.prev_group + deltas_to_state_group_before_event = entry.delta_ids - logger.debug("calling resolve_state_groups from compute_event_context") + # + # make sure that we have a state group at that point. If it's not a state event, + # that will be the state group for the new event. If it *is* a state event, + # it might get rejected (in which case we'll need to persist it with the + # previous state group) + # - entry = yield self.resolve_state_groups_for_events( - event.room_id, event.prev_event_ids() - ) + if not state_group_before_event: + state_group_before_event = yield self.state_store.store_state_group( + event.event_id, + event.room_id, + prev_group=state_group_before_event_prev_group, + delta_ids=deltas_to_state_group_before_event, + current_state_ids=state_ids_before_event, + ) - prev_state_ids = entry.state - prev_group = None - delta_ids = None + # XXX: can we update the state cache entry for the new state group? or + # could we set a flag on resolve_state_groups_for_events to tell it to + # always make a state group? + + # + # now if it's not a state event, we're done + # + + if not event.is_state(): + return EventContext.with_state( + state_group_before_event=state_group_before_event, + state_group=state_group_before_event, + current_state_ids=state_ids_before_event, + prev_state_ids=state_ids_before_event, + prev_group=state_group_before_event_prev_group, + delta_ids=deltas_to_state_group_before_event, + ) - if event.is_state(): - # If this is a state event then we need to create a new state - # group for the state after this event. + # + # otherwise, we'll need to create a new state group for after the event + # - key = (event.type, event.state_key) - if key in prev_state_ids: - replaces = prev_state_ids[key] + key = (event.type, event.state_key) + if key in state_ids_before_event: + replaces = state_ids_before_event[key] + if replaces != event.event_id: event.unsigned["replaces_state"] = replaces - current_state_ids = dict(prev_state_ids) - current_state_ids[key] = event.event_id - - if entry.state_group: - # If the state at the event has a state group assigned then - # we can use that as the prev group - prev_group = entry.state_group - delta_ids = {key: event.event_id} - elif entry.prev_group: - # If the state at the event only has a prev group, then we can - # use that as a prev group too. - prev_group = entry.prev_group - delta_ids = dict(entry.delta_ids) - delta_ids[key] = event.event_id - - state_group = yield self.store.store_state_group( - event.event_id, - event.room_id, - prev_group=prev_group, - delta_ids=delta_ids, - current_state_ids=current_state_ids, - ) - else: - current_state_ids = prev_state_ids - prev_group = entry.prev_group - delta_ids = entry.delta_ids - - if entry.state_group is None: - entry.state_group = yield self.store.store_state_group( - event.event_id, - event.room_id, - prev_group=entry.prev_group, - delta_ids=entry.delta_ids, - current_state_ids=current_state_ids, - ) - entry.state_id = entry.state_group - - state_group = entry.state_group - - context = EventContext.with_state( - state_group=state_group, - current_state_ids=current_state_ids, - prev_state_ids=prev_state_ids, - prev_group=prev_group, + state_ids_after_event = dict(state_ids_before_event) + state_ids_after_event[key] = event.event_id + delta_ids = {key: event.event_id} + + state_group_after_event = yield self.state_store.store_state_group( + event.event_id, + event.room_id, + prev_group=state_group_before_event, delta_ids=delta_ids, + current_state_ids=state_ids_after_event, ) - return context + return EventContext.with_state( + state_group=state_group_after_event, + state_group_before_event=state_group_before_event, + current_state_ids=state_ids_after_event, + prev_state_ids=state_ids_before_event, + prev_group=state_group_before_event, + delta_ids=delta_ids, + ) + @measure_func() @defer.inlineCallbacks def resolve_state_groups_for_events(self, room_id, event_ids): """ Given a list of event_ids this method fetches the state at each @@ -364,14 +374,16 @@ class StateHandler(object): # map from state group id to the state in that state group (where # 'state' is a map from state key to event id) # dict[int, dict[(str, str), str]] - state_groups_ids = yield self.store.get_state_groups_ids(room_id, event_ids) + state_groups_ids = yield self.state_store.get_state_groups_ids( + room_id, event_ids + ) if len(state_groups_ids) == 0: return _StateCacheEntry(state={}, state_group=None) elif len(state_groups_ids) == 1: name, state_list = list(state_groups_ids.items()).pop() - prev_group, delta_ids = yield self.store.get_state_group_delta(name) + prev_group, delta_ids = yield self.state_store.get_state_group_delta(name) return _StateCacheEntry( state=state_list, @@ -380,7 +392,7 @@ class StateHandler(object): delta_ids=delta_ids, ) - room_version = yield self.store.get_room_version(room_id) + room_version = yield self.store.get_room_version_id(room_id) result = yield self._state_resolution_handler.resolve_state_groups( room_id, @@ -404,6 +416,7 @@ class StateHandler(object): with Measure(self.clock, "state._resolve_events"): new_state = yield resolve_events_with_store( + event.room_id, room_version, state_set_ids, event_map=state_map, @@ -432,7 +445,7 @@ class StateResolutionHandler(object): self._state_cache = ExpiringCache( cache_name="state_cache", clock=self.clock, - max_len=SIZE_OF_CACHE, + max_len=100000, expiry_ms=EVICTION_TIMEOUT_SECONDS * 1000, iterable=True, reset_expiry_on_get=True, @@ -449,7 +462,7 @@ class StateResolutionHandler(object): not be called for a single state group Args: - room_id (str): room we are resolving for (used for logging) + room_id (str): room we are resolving for (used for logging and sanity checks) room_version (str): version of the room state_groups_ids (dict[int, dict[(str, str), str]]): map from state group id to the state in that state group @@ -505,6 +518,7 @@ class StateResolutionHandler(object): logger.info("Resolving conflicted state for %r", room_id) with Measure(self.clock, "state._resolve_events"): new_state = yield resolve_events_with_store( + room_id, room_version, list(itervalues(state_groups_ids)), event_map=event_map, @@ -576,36 +590,44 @@ def _make_state_cache_entry(new_state, state_groups_ids): ) -def resolve_events_with_store(room_version, state_sets, event_map, state_res_store): +def resolve_events_with_store( + room_id: str, + room_version: str, + state_sets: List[StateMap[str]], + event_map: Optional[Dict[str, EventBase]], + state_res_store: "StateResolutionStore", +): """ Args: - room_version(str): Version of the room + room_id: the room we are working in - state_sets(list): List of dicts of (type, state_key) -> event_id, + room_version: Version of the room + + state_sets: List of dicts of (type, state_key) -> event_id, which are the different state groups to resolve. - event_map(dict[str,FrozenEvent]|None): + event_map: a dict from event_id to event, for any events that we happen to have in flight (eg, those currently being persisted). This will be used as a starting point fof finding the state we need; any missing events will be requested via state_map_factory. - If None, all events will be fetched via state_map_factory. + If None, all events will be fetched via state_res_store. - state_res_store (StateResolutionStore) + state_res_store: a place to fetch events from - Returns + Returns: Deferred[dict[(str, str), str]]: a map from (type, state_key) to event_id. """ v = KNOWN_ROOM_VERSIONS[room_version] if v.state_res == StateResolutionVersions.V1: return v1.resolve_events_with_store( - state_sets, event_map, state_res_store.get_events + room_id, state_sets, event_map, state_res_store.get_events ) else: return v2.resolve_events_with_store( - room_version, state_sets, event_map, state_res_store + room_id, room_version, state_sets, event_map, state_res_store ) @@ -633,28 +655,21 @@ class StateResolutionStore(object): return self.store.get_events( event_ids, - check_redacted=False, + redact_behaviour=EventRedactBehaviour.AS_IS, get_prev_content=False, allow_rejected=allow_rejected, ) - def get_auth_chain(self, event_ids): - """Gets the full auth chain for a set of events (including rejected - events). - - Includes the given event IDs in the result. + def get_auth_chain_difference(self, state_sets: List[Set[str]]): + """Given sets of state events figure out the auth chain difference (as + per state res v2 algorithm). - Note that: - 1. All events must be state events. - 2. For v1 rooms this may not have the full auth chain in the - presence of rejected events - - Args: - event_ids (list): The event IDs of the events to fetch the auth - chain for. Must be state events. + This equivalent to fetching the full auth chain for each set of state + and returning the events that don't appear in each and every auth + chain. Returns: - Deferred[list[str]]: List of event IDs of the auth chain. + Deferred[Set[str]]: Set of event IDs. """ - return self.store.get_auth_chain_ids(event_ids, include_given=True) + return self.store.get_auth_chain_difference(state_sets) |