forked-synapse/synapse/state/v2.py
Erik Johnston 67f9e5293e
Ensure a long state res does not starve CPU (#15960)
We do this by yielding the reactor in hot loops.
2023-07-19 17:00:33 +00:00

847 lines
27 KiB
Python

# 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.
import heapq
import itertools
import logging
from typing import (
Any,
Awaitable,
Callable,
Collection,
Dict,
Generator,
Iterable,
List,
Mapping,
Optional,
Sequence,
Set,
Tuple,
overload,
)
from typing_extensions import Literal, Protocol
from synapse import event_auth
from synapse.api.constants import EventTypes
from synapse.api.errors import AuthError
from synapse.api.room_versions import RoomVersion
from synapse.events import EventBase
from synapse.types import MutableStateMap, StateMap
logger = logging.getLogger(__name__)
class Clock(Protocol):
# This is usually synapse.util.Clock, but it's replaced with a FakeClock in tests.
# We only ever sleep(0) though, so that other async functions can make forward
# progress without waiting for stateres to complete.
def sleep(self, duration_ms: float) -> Awaitable[None]:
...
class StateResolutionStore(Protocol):
# This is usually synapse.state.StateResolutionStore, but it's replaced with a
# TestStateResolutionStore in tests.
def get_events(
self, event_ids: Collection[str], allow_rejected: bool = False
) -> Awaitable[Dict[str, EventBase]]:
...
def get_auth_chain_difference(
self, room_id: str, state_sets: List[Set[str]]
) -> Awaitable[Set[str]]:
...
# We want to await to the reactor occasionally during state res when dealing
# with large data sets, so that we don't exhaust the reactor. This is done by
# awaiting to reactor during loops every N iterations.
_AWAIT_AFTER_ITERATIONS = 100
__all__ = [
"resolve_events_with_store",
]
async def resolve_events_with_store(
clock: Clock,
room_id: str,
room_version: RoomVersion,
state_sets: Sequence[StateMap[str]],
event_map: Optional[Dict[str, EventBase]],
state_res_store: StateResolutionStore,
) -> StateMap[str]:
"""Resolves the state using the v2 state resolution algorithm
Args:
clock
room_id: the room we are working in
room_version: The room version
state_sets: List of dicts of (type, state_key) -> event_id,
which are the different state groups to resolve.
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 for finding the state we need; any missing
events will be requested via state_res_store.
If None, all events will be fetched via state_res_store.
state_res_store:
Returns:
A map from (type, state_key) to event_id.
"""
logger.debug("Computing conflicted state")
# We use event_map as a cache, so if its None we need to initialize it
if event_map is None:
event_map = {}
# First split up the un/conflicted state
unconflicted_state, conflicted_state = _seperate(state_sets)
if not conflicted_state:
return unconflicted_state
logger.debug("%d conflicted state entries", len(conflicted_state))
logger.debug("Calculating auth chain difference")
# Also fetch all auth events that appear in only some of the state sets'
# auth chains.
auth_diff = await _get_auth_chain_difference(
room_id, state_sets, event_map, state_res_store
)
full_conflicted_set = set(
itertools.chain(
itertools.chain.from_iterable(conflicted_state.values()), auth_diff
)
)
events = await state_res_store.get_events(
[eid for eid in full_conflicted_set if eid not in event_map],
allow_rejected=True,
)
event_map.update(events)
# everything in the event map should be in the right room
for event in event_map.values():
if event.room_id != room_id:
raise Exception(
"Attempting to state-resolve for room %s with event %s which is in %s"
% (
room_id,
event.event_id,
event.room_id,
)
)
full_conflicted_set = {eid for eid in full_conflicted_set if eid in event_map}
logger.debug("%d full_conflicted_set entries", len(full_conflicted_set))
# Get and sort all the power events (kicks/bans/etc)
power_events = (
eid for eid in full_conflicted_set if _is_power_event(event_map[eid])
)
sorted_power_events = await _reverse_topological_power_sort(
clock, room_id, power_events, event_map, state_res_store, full_conflicted_set
)
logger.debug("sorted %d power events", len(sorted_power_events))
# Now sequentially auth each one
resolved_state = await _iterative_auth_checks(
clock,
room_id,
room_version,
sorted_power_events,
unconflicted_state,
event_map,
state_res_store,
)
logger.debug("resolved power events")
# OK, so we've now resolved the power events. Now sort the remaining
# events using the mainline of the resolved power level.
set_power_events = set(sorted_power_events)
leftover_events = [
ev_id for ev_id in full_conflicted_set if ev_id not in set_power_events
]
logger.debug("sorting %d remaining events", len(leftover_events))
pl = resolved_state.get((EventTypes.PowerLevels, ""), None)
leftover_events = await _mainline_sort(
clock, room_id, leftover_events, pl, event_map, state_res_store
)
logger.debug("resolving remaining events")
resolved_state = await _iterative_auth_checks(
clock,
room_id,
room_version,
leftover_events,
resolved_state,
event_map,
state_res_store,
)
logger.debug("resolved")
# We make sure that unconflicted state always still applies.
resolved_state.update(unconflicted_state)
logger.debug("done")
return resolved_state
async def _get_power_level_for_sender(
room_id: str,
event_id: str,
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
) -> int:
"""Return the power level of the sender of the given event according to
their auth events.
Args:
room_id
event_id
event_map
state_res_store
Returns:
The power level.
"""
event = await _get_event(room_id, event_id, event_map, state_res_store)
pl = None
for aid in event.auth_event_ids():
aev = await _get_event(
room_id, aid, event_map, state_res_store, allow_none=True
)
if aev and (aev.type, aev.state_key) == (EventTypes.PowerLevels, ""):
pl = aev
break
if pl is None:
# Couldn't find power level. Check if they're the creator of the room
for aid in event.auth_event_ids():
aev = await _get_event(
room_id, aid, event_map, state_res_store, allow_none=True
)
if aev and (aev.type, aev.state_key) == (EventTypes.Create, ""):
if aev.content.get("creator") == event.sender:
return 100
break
return 0
level = pl.content.get("users", {}).get(event.sender)
if level is None:
level = pl.content.get("users_default", 0)
if level is None:
return 0
else:
return int(level)
async def _get_auth_chain_difference(
room_id: str,
state_sets: Sequence[Mapping[Any, str]],
unpersisted_events: Dict[str, EventBase],
state_res_store: StateResolutionStore,
) -> Set[str]:
"""Compare the auth chains of each state set and return the set of events
that only appear in some, but not all of the auth chains.
Args:
state_sets: The input state sets we are trying to resolve across.
unpersisted_events: A map from event ID to EventBase containing all unpersisted
events involved in this resolution.
state_res_store:
Returns:
The auth difference of the given state sets, as a set of event IDs.
"""
# The `StateResolutionStore.get_auth_chain_difference` function assumes that
# all events passed to it (and their auth chains) have been persisted
# previously. We need to manually handle any other events that are yet to be
# persisted.
#
# We do this in three steps:
# 1. Compute the set of unpersisted events belonging to the auth difference.
# 2. Replacing any unpersisted events in the state_sets with their auth events,
# recursively, until the state_sets contain only persisted events.
# Then we call `store.get_auth_chain_difference` as normal, which computes
# the set of persisted events belonging to the auth difference.
# 3. Adding the results of 1 and 2 together.
# Map from event ID in `unpersisted_events` to their auth event IDs, and their auth
# event IDs if they appear in the `unpersisted_events`. This is the intersection of
# the event's auth chain with the events in `unpersisted_events` *plus* their
# auth event IDs.
events_to_auth_chain: Dict[str, Set[str]] = {}
for event in unpersisted_events.values():
chain = {event.event_id}
events_to_auth_chain[event.event_id] = chain
to_search = [event]
while to_search:
for auth_id in to_search.pop().auth_event_ids():
chain.add(auth_id)
auth_event = unpersisted_events.get(auth_id)
if auth_event:
to_search.append(auth_event)
# We now 1) calculate the auth chain difference for the unpersisted events
# and 2) work out the state sets to pass to the store.
#
# Note: If there are no `unpersisted_events` (which is the common case), we can do a
# much simpler calculation.
if unpersisted_events:
# The list of state sets to pass to the store, where each state set is a set
# of the event ids making up the state. This is similar to `state_sets`,
# except that (a) we only have event ids, not the complete
# ((type, state_key)->event_id) mappings; and (b) we have stripped out
# unpersisted events and replaced them with the persisted events in
# their auth chain.
state_sets_ids: List[Set[str]] = []
# For each state set, the unpersisted event IDs reachable (by their auth
# chain) from the events in that set.
unpersisted_set_ids: List[Set[str]] = []
for state_set in state_sets:
set_ids: Set[str] = set()
state_sets_ids.append(set_ids)
unpersisted_ids: Set[str] = set()
unpersisted_set_ids.append(unpersisted_ids)
for event_id in state_set.values():
event_chain = events_to_auth_chain.get(event_id)
if event_chain is not None:
# We have an unpersisted event. We add all the auth
# events that it references which are also unpersisted.
set_ids.update(
e for e in event_chain if e not in unpersisted_events
)
# We also add the full chain of unpersisted event IDs
# referenced by this state set, so that we can work out the
# auth chain difference of the unpersisted events.
unpersisted_ids.update(
e for e in event_chain if e in unpersisted_events
)
else:
set_ids.add(event_id)
# The auth chain difference of the unpersisted events of the state sets
# is calculated by taking the difference between the union and
# intersections.
union = unpersisted_set_ids[0].union(*unpersisted_set_ids[1:])
intersection = unpersisted_set_ids[0].intersection(*unpersisted_set_ids[1:])
auth_difference_unpersisted_part: Collection[str] = union - intersection
else:
auth_difference_unpersisted_part = ()
state_sets_ids = [set(state_set.values()) for state_set in state_sets]
difference = await state_res_store.get_auth_chain_difference(
room_id, state_sets_ids
)
difference.update(auth_difference_unpersisted_part)
return difference
def _seperate(
state_sets: Iterable[StateMap[str]],
) -> Tuple[StateMap[str], StateMap[Set[str]]]:
"""Return the unconflicted and conflicted state. This is different than in
the original algorithm, as this defines a key to be conflicted if one of
the state sets doesn't have that key.
Args:
state_sets
Returns:
A tuple of unconflicted and conflicted state. The conflicted state dict
is a map from type/state_key to set of event IDs
"""
unconflicted_state = {}
conflicted_state = {}
for key in set(itertools.chain.from_iterable(state_sets)):
event_ids = {state_set.get(key) for state_set in state_sets}
if len(event_ids) == 1:
unconflicted_state[key] = event_ids.pop()
else:
event_ids.discard(None)
conflicted_state[key] = event_ids
# mypy doesn't understand that discarding None above means that conflicted
# state is StateMap[Set[str]], not StateMap[Set[Optional[Str]]].
return unconflicted_state, conflicted_state # type: ignore
def _is_power_event(event: EventBase) -> bool:
"""Return whether or not the event is a "power event", as defined by the
v2 state resolution algorithm
Args:
event
Returns:
True if the event is a power event.
"""
if (event.type, event.state_key) in (
(EventTypes.PowerLevels, ""),
(EventTypes.JoinRules, ""),
(EventTypes.Create, ""),
):
return True
if event.type == EventTypes.Member:
if event.membership in ("leave", "ban"):
return event.sender != event.state_key
return False
async def _add_event_and_auth_chain_to_graph(
graph: Dict[str, Set[str]],
room_id: str,
event_id: str,
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
full_conflicted_set: Set[str],
) -> None:
"""Helper function for _reverse_topological_power_sort that add the event
and its auth chain (that is in the auth diff) to the graph
Args:
graph: A map from event ID to the events auth event IDs
room_id: the room we are working in
event_id: Event to add to the graph
event_map
state_res_store
full_conflicted_set: Set of event IDs that are in the full conflicted set.
"""
state = [event_id]
while state:
eid = state.pop()
graph.setdefault(eid, set())
event = await _get_event(room_id, eid, event_map, state_res_store)
for aid in event.auth_event_ids():
if aid in full_conflicted_set:
if aid not in graph:
state.append(aid)
graph.setdefault(eid, set()).add(aid)
async def _reverse_topological_power_sort(
clock: Clock,
room_id: str,
event_ids: Iterable[str],
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
full_conflicted_set: Set[str],
) -> List[str]:
"""Returns a list of the event_ids sorted by reverse topological ordering,
and then by power level and origin_server_ts
Args:
clock
room_id: the room we are working in
event_ids: The events to sort
event_map
state_res_store
full_conflicted_set: Set of event IDs that are in the full conflicted set.
Returns:
The sorted list
"""
graph: Dict[str, Set[str]] = {}
for idx, event_id in enumerate(event_ids, start=1):
await _add_event_and_auth_chain_to_graph(
graph, room_id, event_id, event_map, state_res_store, full_conflicted_set
)
# We await occasionally when we're working with large data sets to
# ensure that we don't block the reactor loop for too long.
if idx % _AWAIT_AFTER_ITERATIONS == 0:
await clock.sleep(0)
event_to_pl = {}
for idx, event_id in enumerate(graph, start=1):
pl = await _get_power_level_for_sender(
room_id, event_id, event_map, state_res_store
)
event_to_pl[event_id] = pl
# We await occasionally when we're working with large data sets to
# ensure that we don't block the reactor loop for too long.
if idx % _AWAIT_AFTER_ITERATIONS == 0:
await clock.sleep(0)
def _get_power_order(event_id: str) -> Tuple[int, int, str]:
ev = event_map[event_id]
pl = event_to_pl[event_id]
return -pl, ev.origin_server_ts, event_id
# Note: graph is modified during the sort
it = lexicographical_topological_sort(graph, key=_get_power_order)
sorted_events = list(it)
return sorted_events
async def _iterative_auth_checks(
clock: Clock,
room_id: str,
room_version: RoomVersion,
event_ids: List[str],
base_state: StateMap[str],
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
) -> MutableStateMap[str]:
"""Sequentially apply auth checks to each event in given list, updating the
state as it goes along.
Args:
clock
room_id
room_version
event_ids: Ordered list of events to apply auth checks to
base_state: The set of state to start with
event_map
state_res_store
Returns:
Returns the final updated state
"""
resolved_state = dict(base_state)
for idx, event_id in enumerate(event_ids, start=1):
event = event_map[event_id]
auth_events = {}
for aid in event.auth_event_ids():
ev = await _get_event(
room_id, aid, event_map, state_res_store, allow_none=True
)
if not ev:
logger.warning(
"auth_event id %s for event %s is missing", aid, event_id
)
else:
if ev.rejected_reason is None:
auth_events[(ev.type, ev.state_key)] = ev
for key in event_auth.auth_types_for_event(room_version, event):
if key in resolved_state:
ev_id = resolved_state[key]
ev = await _get_event(room_id, ev_id, event_map, state_res_store)
if ev.rejected_reason is None:
auth_events[key] = event_map[ev_id]
if event.rejected_reason is not None:
# Do not admit previously rejected events into state.
# TODO: This isn't spec compliant. Events that were previously rejected due
# to failing auth checks at their state, but pass auth checks during
# state resolution should be accepted. Synapse does not handle the
# change of rejection status well, so we preserve the previous
# rejection status for now.
#
# Note that events rejected for non-state reasons, such as having the
# wrong auth events, should remain rejected.
#
# https://spec.matrix.org/v1.2/rooms/v9/#rejected-events
# https://github.com/matrix-org/synapse/issues/13797
continue
try:
event_auth.check_state_dependent_auth_rules(
event,
auth_events.values(),
)
resolved_state[(event.type, event.state_key)] = event_id
except AuthError:
pass
# We await occasionally when we're working with large data sets to
# ensure that we don't block the reactor loop for too long.
if idx % _AWAIT_AFTER_ITERATIONS == 0:
await clock.sleep(0)
return resolved_state
async def _mainline_sort(
clock: Clock,
room_id: str,
event_ids: List[str],
resolved_power_event_id: Optional[str],
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
) -> List[str]:
"""Returns a sorted list of event_ids sorted by mainline ordering based on
the given event resolved_power_event_id
Args:
clock
room_id: room we're working in
event_ids: Events to sort
resolved_power_event_id: The final resolved power level event ID
event_map
state_res_store
Returns:
The sorted list
"""
if not event_ids:
# It's possible for there to be no event IDs here to sort, so we can
# skip calculating the mainline in that case.
return []
mainline = []
pl = resolved_power_event_id
idx = 0
while pl:
mainline.append(pl)
pl_ev = await _get_event(room_id, pl, event_map, state_res_store)
auth_events = pl_ev.auth_event_ids()
pl = None
for aid in auth_events:
ev = await _get_event(
room_id, aid, event_map, state_res_store, allow_none=True
)
if ev and (ev.type, ev.state_key) == (EventTypes.PowerLevels, ""):
pl = aid
break
# We await occasionally when we're working with large data sets to
# ensure that we don't block the reactor loop for too long.
if idx != 0 and idx % _AWAIT_AFTER_ITERATIONS == 0:
await clock.sleep(0)
idx += 1
mainline_map = {ev_id: i + 1 for i, ev_id in enumerate(reversed(mainline))}
event_ids = list(event_ids)
order_map = {}
for idx, ev_id in enumerate(event_ids, start=1):
depth = await _get_mainline_depth_for_event(
clock, event_map[ev_id], mainline_map, event_map, state_res_store
)
order_map[ev_id] = (depth, event_map[ev_id].origin_server_ts, ev_id)
# We await occasionally when we're working with large data sets to
# ensure that we don't block the reactor loop for too long.
if idx % _AWAIT_AFTER_ITERATIONS == 0:
await clock.sleep(0)
event_ids.sort(key=lambda ev_id: order_map[ev_id])
return event_ids
async def _get_mainline_depth_for_event(
clock: Clock,
event: EventBase,
mainline_map: Dict[str, int],
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
) -> int:
"""Get the mainline depths for the given event based on the mainline map
Args:
event
mainline_map: Map from event_id to mainline depth for events in the mainline.
event_map
state_res_store
Returns:
The mainline depth
"""
room_id = event.room_id
tmp_event: Optional[EventBase] = event
# We do an iterative search, replacing `event with the power level in its
# auth events (if any)
idx = 0
while tmp_event:
depth = mainline_map.get(tmp_event.event_id)
if depth is not None:
return depth
auth_events = tmp_event.auth_event_ids()
tmp_event = None
for aid in auth_events:
aev = await _get_event(
room_id, aid, event_map, state_res_store, allow_none=True
)
if aev and (aev.type, aev.state_key) == (EventTypes.PowerLevels, ""):
tmp_event = aev
break
idx += 1
if idx % _AWAIT_AFTER_ITERATIONS == 0:
await clock.sleep(0)
# Didn't find a power level auth event, so we just return 0
return 0
@overload
async def _get_event(
room_id: str,
event_id: str,
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
allow_none: Literal[False] = False,
) -> EventBase:
...
@overload
async def _get_event(
room_id: str,
event_id: str,
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
allow_none: Literal[True],
) -> Optional[EventBase]:
...
async def _get_event(
room_id: str,
event_id: str,
event_map: Dict[str, EventBase],
state_res_store: StateResolutionStore,
allow_none: bool = False,
) -> Optional[EventBase]:
"""Helper function to look up event in event_map, falling back to looking
it up in the store
Args:
room_id
event_id
event_map
state_res_store
allow_none: if the event is not found, return None rather than raising
an exception
Returns:
The event, or none if the event does not exist (and allow_none is True).
"""
if event_id not in event_map:
events = await state_res_store.get_events([event_id], allow_rejected=True)
event_map.update(events)
event = event_map.get(event_id)
if event is None:
if allow_none:
return None
raise Exception("Unknown event %s" % (event_id,))
if event.room_id != room_id:
raise Exception(
"In state res for room %s, event %s is in %s"
% (room_id, event_id, event.room_id)
)
return event
def lexicographical_topological_sort(
graph: Dict[str, Set[str]], key: Callable[[str], Any]
) -> Generator[str, None, None]:
"""Performs a lexicographic reverse topological sort on the graph.
This returns a reverse topological sort (i.e. if node A references B then B
appears before A in the sort), with ties broken lexicographically based on
return value of the `key` function.
NOTE: `graph` is modified during the sort.
Args:
graph: A representation of the graph where each node is a key in the
dict and its value are the nodes edges.
key: A function that takes a node and returns a value that is comparable
and used to order nodes
Yields:
The next node in the topological sort
"""
# Note, this is basically Kahn's algorithm except we look at nodes with no
# outgoing edges, c.f.
# https://en.wikipedia.org/wiki/Topological_sorting#Kahn's_algorithm
outdegree_map = graph
reverse_graph: Dict[str, Set[str]] = {}
# Lists of nodes with zero out degree. Is actually a tuple of
# `(key(node), node)` so that sorting does the right thing
zero_outdegree = []
for node, edges in graph.items():
if len(edges) == 0:
zero_outdegree.append((key(node), node))
reverse_graph.setdefault(node, set())
for edge in edges:
reverse_graph.setdefault(edge, set()).add(node)
# heapq is a built in implementation of a sorted queue.
heapq.heapify(zero_outdegree)
while zero_outdegree:
_, node = heapq.heappop(zero_outdegree)
for parent in reverse_graph[node]:
out = outdegree_map[parent]
out.discard(node)
if len(out) == 0:
heapq.heappush(zero_outdegree, (key(parent), parent))
yield node