anonymousland-synapse/synapse/storage/events.py

2471 lines
94 KiB
Python
Raw Normal View History

2015-03-20 09:52:56 -04:00
# -*- coding: utf-8 -*-
2016-01-06 23:26:29 -05:00
# Copyright 2014-2016 OpenMarket Ltd
# Copyright 2018-2019 New Vector Ltd
# Copyright 2019 The Matrix.org Foundation C.I.C.
2015-03-20 09:52:56 -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.
import itertools
import logging
from collections import Counter as c_counter, OrderedDict, deque, namedtuple
2018-07-09 02:09:20 -04:00
from functools import wraps
2018-08-30 10:19:58 -04:00
from six import iteritems, text_type
2018-07-09 02:09:20 -04:00
from six.moves import range
2015-03-20 09:52:56 -04:00
2019-08-29 12:38:51 -04:00
from canonicaljson import encode_canonical_json, json
from prometheus_client import Counter, Histogram
from twisted.internet import defer
2018-07-09 02:09:20 -04:00
import synapse.metrics
from synapse.api.constants import EventTypes
from synapse.api.errors import SynapseError
from synapse.events import EventBase # noqa: F401
from synapse.events.snapshot import EventContext # noqa: F401
2019-08-29 12:38:51 -04:00
from synapse.events.utils import prune_event_dict
from synapse.logging.context import PreserveLoggingContext, make_deferred_yieldable
from synapse.logging.utils import log_function
from synapse.metrics import BucketCollector
from synapse.metrics.background_process_metrics import run_as_background_process
from synapse.state import StateResolutionStore
from synapse.storage.background_updates import BackgroundUpdateStore
from synapse.storage.event_federation import EventFederationStore
from synapse.storage.events_worker import EventsWorkerStore
2018-10-12 15:43:18 -04:00
from synapse.storage.state import StateGroupWorkerStore
2018-07-09 02:09:20 -04:00
from synapse.types import RoomStreamToken, get_domain_from_id
from synapse.util import batch_iter
from synapse.util.async_helpers import ObservableDeferred
2018-07-09 02:09:20 -04:00
from synapse.util.caches.descriptors import cached, cachedInlineCallbacks
from synapse.util.frozenutils import frozendict_json_encoder
2016-08-19 12:40:23 -04:00
from synapse.util.metrics import Measure
2015-03-20 09:52:56 -04:00
2018-05-21 20:48:57 -04:00
logger = logging.getLogger(__name__)
2015-03-20 09:52:56 -04:00
2018-05-21 20:48:57 -04:00
persist_event_counter = Counter("synapse_storage_events_persisted_events", "")
event_counter = Counter(
"synapse_storage_events_persisted_events_sep",
"",
["type", "origin_type", "origin_entity"],
)
2018-03-27 08:13:38 -04:00
# The number of times we are recalculating the current state
2018-05-21 20:48:57 -04:00
state_delta_counter = Counter("synapse_storage_events_state_delta", "")
2018-03-27 08:13:38 -04:00
# The number of times we are recalculating state when there is only a
# single forward extremity
2018-05-22 18:32:57 -04:00
state_delta_single_event_counter = Counter(
"synapse_storage_events_state_delta_single_event", ""
)
2018-05-21 20:48:57 -04:00
2018-03-27 08:13:38 -04:00
# The number of times we are reculating state when we could have resonably
# calculated the delta when we calculated the state for an event we were
# persisting.
2018-05-22 18:32:57 -04:00
state_delta_reuse_delta_counter = Counter(
"synapse_storage_events_state_delta_reuse_delta", ""
)
# The number of forward extremities for each new event.
forward_extremities_counter = Histogram(
"synapse_storage_events_forward_extremities_persisted",
"Number of forward extremities for each new event",
2019-06-20 08:10:06 -04:00
buckets=(1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"),
)
# The number of stale forward extremities for each new event. Stale extremities
# are those that were in the previous set of extremities as well as the new.
stale_forward_extremities_counter = Histogram(
"synapse_storage_events_stale_forward_extremities_persisted",
"Number of unchanged forward extremities for each new event",
2019-06-20 08:10:06 -04:00
buckets=(0, 1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"),
)
def encode_json(json_object):
"""
Encode a Python object as JSON and return it in a Unicode string.
"""
out = frozendict_json_encoder.encode(json_object)
if isinstance(out, bytes):
out = out.decode("utf8")
return out
2016-11-15 06:22:29 -05:00
2016-05-06 09:31:38 -04:00
class _EventPeristenceQueue(object):
"""Queues up events so that they can be persisted in bulk with only one
concurrent transaction per room.
"""
_EventPersistQueueItem = namedtuple(
"_EventPersistQueueItem", ("events_and_contexts", "backfilled", "deferred")
)
2016-05-06 09:31:38 -04:00
def __init__(self):
self._event_persist_queues = {}
self._currently_persisting_rooms = set()
def add_to_queue(self, room_id, events_and_contexts, backfilled):
2016-05-06 09:31:38 -04:00
"""Add events to the queue, with the given persist_event options.
NB: due to the normal usage pattern of this method, it does *not*
follow the synapse logcontext rules, and leaves the logcontext in
place whether or not the returned deferred is ready.
Args:
room_id (str):
events_and_contexts (list[(EventBase, EventContext)]):
backfilled (bool):
Returns:
defer.Deferred: a deferred which will resolve once the events are
persisted. Runs its callbacks *without* a logcontext.
2016-05-06 09:31:38 -04:00
"""
queue = self._event_persist_queues.setdefault(room_id, deque())
if queue:
# if the last item in the queue has the same `backfilled` setting,
# we can just add these new events to that item.
2016-05-06 09:31:38 -04:00
end_item = queue[-1]
if end_item.backfilled == backfilled:
end_item.events_and_contexts.extend(events_and_contexts)
return end_item.deferred.observe()
deferred = ObservableDeferred(defer.Deferred(), consumeErrors=True)
2016-05-06 09:31:38 -04:00
queue.append(
self._EventPersistQueueItem(
events_and_contexts=events_and_contexts,
backfilled=backfilled,
deferred=deferred,
)
)
2016-05-06 09:31:38 -04:00
return deferred.observe()
2016-05-06 10:38:42 -04:00
def handle_queue(self, room_id, per_item_callback):
2016-05-06 09:31:38 -04:00
"""Attempts to handle the queue for a room if not already being handled.
The given callback will be invoked with for each item in the queue,
2016-05-06 10:38:42 -04:00
of type _EventPersistQueueItem. The per_item_callback will continuously
be called with new items, unless the queue becomnes empty. The return
value of the function will be given to the deferreds waiting on the item,
exceptions will be passed to the deferreds as well.
2016-05-06 09:31:38 -04:00
This function should therefore be called whenever anything is added
to the queue.
If another callback is currently handling the queue then it will not be
invoked.
"""
if room_id in self._currently_persisting_rooms:
return
self._currently_persisting_rooms.add(room_id)
2016-05-06 10:38:42 -04:00
@defer.inlineCallbacks
def handle_queue_loop():
try:
queue = self._get_drainining_queue(room_id)
for item in queue:
try:
ret = yield per_item_callback(item)
except Exception:
with PreserveLoggingContext():
item.deferred.errback()
else:
with PreserveLoggingContext():
item.deferred.callback(ret)
2016-05-06 10:38:42 -04:00
finally:
queue = self._event_persist_queues.pop(room_id, None)
if queue:
self._event_persist_queues[room_id] = queue
self._currently_persisting_rooms.discard(room_id)
# set handle_queue_loop off in the background
run_as_background_process("persist_events", handle_queue_loop)
2016-05-06 09:31:38 -04:00
def _get_drainining_queue(self, room_id):
2016-05-06 10:38:42 -04:00
queue = self._event_persist_queues.setdefault(room_id, deque())
2016-05-06 09:31:38 -04:00
try:
while True:
yield queue.popleft()
except IndexError:
# Queue has been drained.
pass
2016-06-03 12:55:32 -04:00
_EventCacheEntry = namedtuple("_EventCacheEntry", ("event", "redacted_event"))
def _retry_on_integrity_error(func):
"""Wraps a database function so that it gets retried on IntegrityError,
with `delete_existing=True` passed in.
Args:
func: function that returns a Deferred and accepts a `delete_existing` arg
"""
@wraps(func)
@defer.inlineCallbacks
def f(self, *args, **kwargs):
try:
res = yield func(self, *args, **kwargs)
except self.database_engine.module.IntegrityError:
logger.exception("IntegrityError, retrying.")
res = yield func(self, *args, delete_existing=True, **kwargs)
return res
return f
# inherits from EventFederationStore so that we can call _update_backward_extremities
# and _handle_mult_prev_events (though arguably those could both be moved in here)
class EventsStore(
StateGroupWorkerStore,
EventFederationStore,
EventsWorkerStore,
BackgroundUpdateStore,
):
def __init__(self, db_conn, hs):
super(EventsStore, self).__init__(db_conn, hs)
2016-05-06 09:31:38 -04:00
self._event_persist_queue = _EventPeristenceQueue()
self._state_resolution_handler = hs.get_state_resolution_handler()
# Collect metrics on the number of forward extremities that exist.
2019-06-14 09:07:32 -04:00
# Counter of number of extremities to count
self._current_forward_extremities_amount = c_counter()
BucketCollector(
"synapse_forward_extremities",
lambda: self._current_forward_extremities_amount,
2019-06-20 05:32:02 -04:00
buckets=[1, 2, 3, 5, 7, 10, 15, 20, 50, 100, 200, 500, "+Inf"],
)
# Read the extrems every 60 minutes
def read_forward_extremities():
# run as a background process to make sure that the database transactions
# have a logcontext to report to
return run_as_background_process(
"read_forward_extremities", self._read_forward_extremities
)
hs.get_clock().looping_call(read_forward_extremities, 60 * 60 * 1000)
2019-08-29 12:38:51 -04:00
def _censor_redactions():
return run_as_background_process(
"_censor_redactions", self._censor_redactions
)
if self.hs.config.redaction_retention_period is not None:
2019-09-09 08:31:00 -04:00
hs.get_clock().looping_call(_censor_redactions, 5 * 60 * 1000)
2019-08-29 12:38:51 -04:00
@defer.inlineCallbacks
def _read_forward_extremities(self):
def fetch(txn):
txn.execute(
"""
select count(*) c from event_forward_extremities
group by room_id
"""
)
return txn.fetchall()
res = yield self.runInteraction("read_forward_extremities", fetch)
self._current_forward_extremities_amount = c_counter(list(x[0] for x in res))
@defer.inlineCallbacks
def persist_events(self, events_and_contexts, backfilled=False):
"""
Write events to the database
Args:
events_and_contexts: list of tuples of (event, context)
backfilled (bool): Whether the results are retrieved from federation
via backfill or not. Used to determine if they're "new" events
which might update the current state etc.
Returns:
2018-08-01 08:39:14 -04:00
Deferred[int]: the stream ordering of the latest persisted event
2016-05-06 09:31:38 -04:00
"""
partitioned = {}
for event, ctx in events_and_contexts:
partitioned.setdefault(event.room_id, []).append((event, ctx))
2016-05-06 09:31:38 -04:00
deferreds = []
for room_id, evs_ctxs in iteritems(partitioned):
d = self._event_persist_queue.add_to_queue(
room_id, evs_ctxs, backfilled=backfilled
2016-05-06 09:31:38 -04:00
)
deferreds.append(d)
2017-03-24 06:57:02 -04:00
for room_id in partitioned:
2016-05-06 09:31:38 -04:00
self._maybe_start_persisting(room_id)
yield make_deferred_yieldable(
2016-08-23 10:23:39 -04:00
defer.gatherResults(deferreds, consumeErrors=True)
)
2016-05-06 09:31:38 -04:00
max_persisted_id = yield self._stream_id_gen.get_current_token()
return max_persisted_id
2016-05-06 09:31:38 -04:00
@defer.inlineCallbacks
@log_function
def persist_event(self, event, context, backfilled=False):
"""
Args:
event (EventBase):
context (EventContext):
backfilled (bool):
Returns:
Deferred: resolves to (int, int): the stream ordering of ``event``,
and the stream ordering of the latest persisted event
"""
2016-05-06 09:31:38 -04:00
deferred = self._event_persist_queue.add_to_queue(
event.room_id, [(event, context)], backfilled=backfilled
2016-05-06 09:31:38 -04:00
)
self._maybe_start_persisting(event.room_id)
yield make_deferred_yieldable(deferred)
2016-05-06 09:31:38 -04:00
max_persisted_id = yield self._stream_id_gen.get_current_token()
return (event.internal_metadata.stream_ordering, max_persisted_id)
2016-05-06 09:31:38 -04:00
def _maybe_start_persisting(self, room_id):
@defer.inlineCallbacks
2016-05-06 10:38:42 -04:00
def persisting_queue(item):
with Measure(self._clock, "persist_events"):
yield self._persist_events(
item.events_and_contexts, backfilled=item.backfilled
)
2016-05-06 09:31:38 -04:00
self._event_persist_queue.handle_queue(room_id, persisting_queue)
@_retry_on_integrity_error
2016-05-06 09:31:38 -04:00
@defer.inlineCallbacks
def _persist_events(
self, events_and_contexts, backfilled=False, delete_existing=False
):
"""Persist events to db
Args:
events_and_contexts (list[(EventBase, EventContext)]):
backfilled (bool):
delete_existing (bool):
Returns:
Deferred: resolves when the events have been persisted
"""
2015-06-25 12:18:19 -04:00
if not events_and_contexts:
return
chunks = [
events_and_contexts[x : x + 100]
for x in range(0, len(events_and_contexts), 100)
]
for chunk in chunks:
# We can't easily parallelize these since different chunks
# might contain the same event. :(
# NB: Assumes that we are only persisting events for one room
# at a time.
# map room_id->list[event_ids] giving the new forward
# extremities in each room
new_forward_extremeties = {}
# map room_id->(type,state_key)->event_id tracking the full
# state in each room after adding these events.
# This is simply used to prefill the get_current_state_ids
# cache
current_state_for_room = {}
# map room_id->(to_delete, to_insert) where to_delete is a list
# of type/state keys to remove from current state, and to_insert
# is a map (type,key)->event_id giving the state delta in each
# room
state_delta_for_room = {}
if not backfilled:
with Measure(self._clock, "_calculate_state_and_extrem"):
# Work out the new "current state" for each room.
# We do this by working out what the new extremities are and then
# calculating the state from that.
events_by_room = {}
for event, context in chunk:
events_by_room.setdefault(event.room_id, []).append(
(event, context)
)
for room_id, ev_ctx_rm in iteritems(events_by_room):
latest_event_ids = yield self.get_latest_event_ids_in_room(
room_id
)
new_latest_event_ids = yield self._calculate_new_extremities(
room_id, ev_ctx_rm, latest_event_ids
)
latest_event_ids = set(latest_event_ids)
if new_latest_event_ids == latest_event_ids:
# No change in extremities, so no change in state
continue
# there should always be at least one forward extremity.
# (except during the initial persistence of the send_join
# results, in which case there will be no existing
# extremities, so we'll `continue` above and skip this bit.)
assert new_latest_event_ids, "No forward extremities left!"
new_forward_extremeties[room_id] = new_latest_event_ids
len_1 = (
len(latest_event_ids) == 1
and len(new_latest_event_ids) == 1
)
if len_1:
all_single_prev_not_state = all(
len(event.prev_event_ids()) == 1
and not event.is_state()
for event, ctx in ev_ctx_rm
)
# Don't bother calculating state if they're just
# a long chain of single ancestor non-state events.
if all_single_prev_not_state:
continue
state_delta_counter.inc()
if len(new_latest_event_ids) == 1:
state_delta_single_event_counter.inc()
# This is a fairly handwavey check to see if we could
# have guessed what the delta would have been when
# processing one of these events.
# What we're interested in is if the latest extremities
# were the same when we created the event as they are
# now. When this server creates a new event (as opposed
# to receiving it over federation) it will use the
# forward extremities as the prev_events, so we can
# guess this by looking at the prev_events and checking
# if they match the current forward extremities.
for ev, _ in ev_ctx_rm:
prev_event_ids = set(ev.prev_event_ids())
if latest_event_ids == prev_event_ids:
state_delta_reuse_delta_counter.inc()
break
logger.info("Calculating state delta for room %s", room_id)
with Measure(
self._clock, "persist_events.get_new_state_after_events"
):
res = yield self._get_new_state_after_events(
room_id,
ev_ctx_rm,
latest_event_ids,
new_latest_event_ids,
)
current_state, delta_ids = res
# If either are not None then there has been a change,
# and we need to work out the delta (or use that
# given)
if delta_ids is not None:
# If there is a delta we know that we've
# only added or replaced state, never
# removed keys entirely.
state_delta_for_room[room_id] = ([], delta_ids)
elif current_state is not None:
2018-07-23 19:04:44 -04:00
with Measure(
self._clock, "persist_events.calculate_state_delta"
2018-07-23 19:04:44 -04:00
):
delta = yield self._calculate_state_delta(
room_id, current_state
)
state_delta_for_room[room_id] = delta
# If we have the current_state then lets prefill
# the cache with it.
if current_state is not None:
current_state_for_room[room_id] = current_state
# We want to calculate the stream orderings as late as possible, as
# we only notify after all events with a lesser stream ordering have
# been persisted. I.e. if we spend 10s inside the with block then
# that will delay all subsequent events from being notified about.
# Hence why we do it down here rather than wrapping the entire
# function.
#
# Its safe to do this after calculating the state deltas etc as we
# only need to protect the *persistence* of the events. This is to
# ensure that queries of the form "fetch events since X" don't
# return events and stream positions after events that are still in
# flight, as otherwise subsequent requests "fetch event since Y"
# will not return those events.
#
# Note: Multiple instances of this function cannot be in flight at
# the same time for the same room.
if backfilled:
stream_ordering_manager = self._backfill_id_gen.get_next_mult(
len(chunk)
)
else:
stream_ordering_manager = self._stream_id_gen.get_next_mult(len(chunk))
with stream_ordering_manager as stream_orderings:
for (event, context), stream in zip(chunk, stream_orderings):
event.internal_metadata.stream_ordering = stream
yield self.runInteraction(
"persist_events",
self._persist_events_txn,
events_and_contexts=chunk,
backfilled=backfilled,
delete_existing=delete_existing,
state_delta_for_room=state_delta_for_room,
2017-01-20 09:28:53 -05:00
new_forward_extremeties=new_forward_extremeties,
)
2018-05-21 20:48:57 -04:00
persist_event_counter.inc(len(chunk))
if not backfilled:
# backfilled events have negative stream orderings, so we don't
# want to set the event_persisted_position to that.
synapse.metrics.event_persisted_position.set(
chunk[-1][0].internal_metadata.stream_ordering
)
2017-05-02 06:36:11 -04:00
for event, context in chunk:
if context.app_service:
origin_type = "local"
origin_entity = context.app_service.id
elif self.hs.is_mine_id(event.sender):
origin_type = "local"
origin_entity = "*client*"
else:
origin_type = "remote"
origin_entity = get_domain_from_id(event.sender)
2018-05-21 20:48:57 -04:00
event_counter.labels(event.type, origin_type, origin_entity).inc()
2015-06-25 12:18:19 -04:00
for room_id, new_state in iteritems(current_state_for_room):
self.get_current_state_ids.prefill((room_id,), new_state)
2017-05-02 05:40:31 -04:00
for room_id, latest_event_ids in iteritems(new_forward_extremeties):
self.get_latest_event_ids_in_room.prefill(
2017-06-29 10:47:37 -04:00
(room_id,), list(latest_event_ids)
)
2017-01-20 09:28:53 -05:00
@defer.inlineCallbacks
def _calculate_new_extremities(self, room_id, event_contexts, latest_event_ids):
"""Calculates the new forward extremities for a room given events to
2017-01-20 09:38:13 -05:00
persist.
Assumes that we are only persisting events for one room at a time.
"""
# we're only interested in new events which aren't outliers and which aren't
# being rejected.
new_events = [
event
for event, ctx in event_contexts
if not event.internal_metadata.is_outlier()
and not ctx.rejected
2019-02-12 05:31:21 -05:00
and not event.internal_metadata.is_soft_failed()
]
latest_event_ids = set(latest_event_ids)
# start with the existing forward extremities
result = set(latest_event_ids)
# add all the new events to the list
result.update(event.event_id for event in new_events)
# Now remove all events which are prev_events of any of the new events
result.difference_update(
e_id for event in new_events for e_id in event.prev_event_ids()
2017-01-20 09:28:53 -05:00
)
# Remove any events which are prev_events of any existing events.
existing_prevs = yield self._get_events_which_are_prevs(result)
result.difference_update(existing_prevs)
2017-01-20 09:28:53 -05:00
# Finally handle the case where the new events have soft-failed prev
# events. If they do we need to remove them and their prev events,
# otherwise we end up with dangling extremities.
existing_prevs = yield self._get_prevs_before_rejected(
e_id for event in new_events for e_id in event.prev_event_ids()
)
result.difference_update(existing_prevs)
# We only update metrics for events that change forward extremities
# (e.g. we ignore backfill/outliers/etc)
if result != latest_event_ids:
forward_extremities_counter.observe(len(result))
stale = latest_event_ids & result
stale_forward_extremities_counter.observe(len(stale))
return result
2017-01-20 09:28:53 -05:00
@defer.inlineCallbacks
def _get_events_which_are_prevs(self, event_ids):
"""Filter the supplied list of event_ids to get those which are prev_events of
2018-10-03 05:19:41 -04:00
existing (non-outlier/rejected) events.
Args:
event_ids (Iterable[str]): event ids to filter
Returns:
Deferred[List[str]]: filtered event ids
"""
results = []
def _get_events_which_are_prevs_txn(txn, batch):
sql = """
SELECT prev_event_id, internal_metadata
FROM event_edges
INNER JOIN events USING (event_id)
LEFT JOIN rejections USING (event_id)
LEFT JOIN event_json USING (event_id)
WHERE
prev_event_id IN (%s)
2018-10-03 05:19:41 -04:00
AND NOT events.outlier
AND rejections.event_id IS NULL
""" % (
",".join("?" for _ in batch),
)
2017-01-20 09:28:53 -05:00
txn.execute(sql, batch)
results.extend(r[0] for r in txn if not json.loads(r[1]).get("soft_failed"))
for chunk in batch_iter(event_ids, 100):
yield self.runInteraction(
"_get_events_which_are_prevs", _get_events_which_are_prevs_txn, chunk
)
return results
2017-01-20 09:28:53 -05:00
@defer.inlineCallbacks
def _get_prevs_before_rejected(self, event_ids):
"""Get soft-failed ancestors to remove from the extremities.
Given a set of events, find all those that have been soft-failed or
rejected. Returns those soft failed/rejected events and their prev
events (whether soft-failed/rejected or not), and recurses up the
prev-event graph until it finds no more soft-failed/rejected events.
This is used to find extremities that are ancestors of new events, but
are separated by soft failed events.
Args:
event_ids (Iterable[str]): Events to find prev events for. Note
that these must have already been persisted.
Returns:
Deferred[set[str]]
"""
# The set of event_ids to return. This includes all soft-failed events
# and their prev events.
existing_prevs = set()
def _get_prevs_before_rejected_txn(txn, batch):
to_recursively_check = batch
while to_recursively_check:
sql = """
SELECT
event_id, prev_event_id, internal_metadata,
rejections.event_id IS NOT NULL
FROM event_edges
INNER JOIN events USING (event_id)
LEFT JOIN rejections USING (event_id)
LEFT JOIN event_json USING (event_id)
WHERE
event_id IN (%s)
AND NOT events.outlier
""" % (
",".join("?" for _ in to_recursively_check),
)
txn.execute(sql, to_recursively_check)
to_recursively_check = []
for event_id, prev_event_id, metadata, rejected in txn:
if prev_event_id in existing_prevs:
continue
soft_failed = json.loads(metadata).get("soft_failed")
if soft_failed or rejected:
to_recursively_check.append(prev_event_id)
existing_prevs.add(prev_event_id)
for chunk in batch_iter(event_ids, 100):
yield self.runInteraction(
"_get_prevs_before_rejected", _get_prevs_before_rejected_txn, chunk
)
return existing_prevs
@defer.inlineCallbacks
def _get_new_state_after_events(
self, room_id, events_context, old_latest_event_ids, new_latest_event_ids
):
"""Calculate the current state dict after adding some new events to
a room
Args:
room_id (str):
room to which the events are being added. Used for logging etc
events_context (list[(EventBase, EventContext)]):
events and contexts which are being added to the room
old_latest_event_ids (iterable[str]):
the old forward extremities for the room.
new_latest_event_ids (iterable[str]):
the new forward extremities for the room.
Returns:
Deferred[tuple[dict[(str,str), str]|None, dict[(str,str), str]|None]]:
Returns a tuple of two state maps, the first being the full new current
state and the second being the delta to the existing current state.
If both are None then there has been no change.
If there has been a change then we only return the delta if its
already been calculated. Conversely if we do know the delta then
the new current state is only returned if we've already calculated
it.
"""
# map from state_group to ((type, key) -> event_id) state map
state_groups_map = {}
2018-07-24 09:31:38 -04:00
# Map from (prev state group, new state group) -> delta state dict
state_group_deltas = {}
for ev, ctx in events_context:
if ctx.state_group is None:
# This should only happen for outlier events.
2018-07-25 04:48:01 -04:00
if not ev.internal_metadata.is_outlier():
raise Exception(
"Context for new event %s has no state "
"group" % (ev.event_id,)
)
2018-07-25 06:13:20 -04:00
continue
if ctx.state_group in state_groups_map:
continue
2018-07-23 12:43:01 -04:00
# We're only interested in pulling out state that has already
# been cached in the context. We'll pull stuff out of the DB later
# if necessary.
current_state_ids = ctx.get_cached_current_state_ids()
if current_state_ids is not None:
state_groups_map[ctx.state_group] = current_state_ids
if ctx.prev_group:
state_group_deltas[(ctx.prev_group, ctx.state_group)] = ctx.delta_ids
2018-05-03 05:38:08 -04:00
# We need to map the event_ids to their state groups. First, let's
# check if the event is one we're persisting, in which case we can
# pull the state group from its context.
# Otherwise we need to pull the state group from the database.
# Set of events we need to fetch groups for. (We know none of the old
# extremities are going to be in events_context).
missing_event_ids = set(old_latest_event_ids)
event_id_to_state_group = {}
for event_id in new_latest_event_ids:
# First search in the list of new events we're adding.
for ev, ctx in events_context:
if event_id == ev.event_id and ctx.state_group is not None:
event_id_to_state_group[event_id] = ctx.state_group
break
else:
# If we couldn't find it, then we'll need to pull
# the state from the database
missing_event_ids.add(event_id)
if missing_event_ids:
# Now pull out the state groups for any missing events from DB
event_to_groups = yield self._get_state_group_for_events(missing_event_ids)
event_id_to_state_group.update(event_to_groups)
# State groups of old_latest_event_ids
old_state_groups = set(
event_id_to_state_group[evid] for evid in old_latest_event_ids
)
# State groups of new_latest_event_ids
new_state_groups = set(
event_id_to_state_group[evid] for evid in new_latest_event_ids
)
# If they old and new groups are the same then we don't need to do
# anything.
if old_state_groups == new_state_groups:
return None, None
2018-07-24 09:31:38 -04:00
if len(new_state_groups) == 1 and len(old_state_groups) == 1:
# If we're going from one state group to another, lets check if
# we have a delta for that transition. If we do then we can just
# return that.
2018-07-24 09:31:38 -04:00
new_state_group = next(iter(new_state_groups))
old_state_group = next(iter(old_state_groups))
delta_ids = state_group_deltas.get((old_state_group, new_state_group), None)
2018-07-24 09:31:38 -04:00
if delta_ids is not None:
# We have a delta from the existing to new current state,
# so lets just return that. If we happen to already have
# the current state in memory then lets also return that,
# but it doesn't matter if we don't.
new_state = state_groups_map.get(new_state_group)
return new_state, delta_ids
# Now that we have calculated new_state_groups we need to get
# their state IDs so we can resolve to a single state set.
missing_state = new_state_groups - set(state_groups_map)
if missing_state:
group_to_state = yield self._get_state_for_groups(missing_state)
state_groups_map.update(group_to_state)
if len(new_state_groups) == 1:
2018-07-24 09:31:38 -04:00
# If there is only one state group, then we know what the current
# state is.
return state_groups_map[new_state_groups.pop()], None
# Ok, we need to defer to the state handler to resolve our state sets.
state_groups = {sg: state_groups_map[sg] for sg in new_state_groups}
events_map = {ev.event_id: ev for ev, _ in events_context}
# We need to get the room version, which is in the create event.
# Normally that'd be in the database, but its also possible that we're
# currently trying to persist it.
room_version = None
for ev, _ in events_context:
if ev.type == EventTypes.Create and ev.state_key == "":
room_version = ev.content.get("room_version", "1")
break
if not room_version:
room_version = yield self.get_room_version(room_id)
logger.debug("calling resolve_state_groups from preserve_events")
res = yield self._state_resolution_handler.resolve_state_groups(
room_id,
room_version,
state_groups,
events_map,
state_res_store=StateResolutionStore(self),
)
return res.state, None
@defer.inlineCallbacks
def _calculate_state_delta(self, room_id, current_state):
"""Calculate the new state deltas for a room.
Assumes that we are only persisting events for one room at a time.
Returns:
tuple[list, dict] (to_delete, to_insert): where to_delete are the
type/state_keys to remove from current_state_events and `to_insert`
are the updates to current_state_events.
"""
existing_state = yield self.get_current_state_ids(room_id)
to_delete = [key for key in existing_state if key not in current_state]
2018-07-24 05:55:11 -04:00
to_insert = {
key: ev_id
for key, ev_id in iteritems(current_state)
2018-07-24 05:55:11 -04:00
if ev_id != existing_state.get(key)
}
return to_delete, to_insert
2015-06-25 12:18:19 -04:00
@log_function
def _persist_events_txn(
self,
txn,
events_and_contexts,
backfilled,
delete_existing=False,
state_delta_for_room={},
new_forward_extremeties={},
):
"""Insert some number of room events into the necessary database tables.
Rejected events are only inserted into the events table, the events_json table,
and the rejections table. Things reading from those table will need to check
whether the event was rejected.
Args:
txn (twisted.enterprise.adbapi.Connection): db connection
events_and_contexts (list[(EventBase, EventContext)]):
events to persist
backfilled (bool): True if the events were backfilled
delete_existing (bool): True to purge existing table rows for the
events from the database. This is useful when retrying due to
IntegrityError.
state_delta_for_room (dict[str, (list, dict)]):
The current-state delta for each room. For each room, a tuple
(to_delete, to_insert), being a list of type/state keys to be
removed from the current state, and a state set to be added to
the current state.
new_forward_extremeties (dict[str, list[str]]):
The new forward extremities for each room. For each room, a
list of the event ids which are the forward extremities.
"""
all_events_and_contexts = events_and_contexts
min_stream_order = events_and_contexts[0][0].internal_metadata.stream_ordering
2017-01-20 09:28:53 -05:00
max_stream_order = events_and_contexts[-1][0].internal_metadata.stream_ordering
2017-05-30 09:41:42 -04:00
self._update_forward_extremities_txn(
txn,
new_forward_extremities=new_forward_extremeties,
max_stream_order=max_stream_order,
)
# Ensure that we don't have the same event twice.
events_and_contexts = self._filter_events_and_contexts_for_duplicates(
events_and_contexts
)
self._update_room_depths_txn(
txn, events_and_contexts=events_and_contexts, backfilled=backfilled
)
# _update_outliers_txn filters out any events which have already been
# persisted, and returns the filtered list.
events_and_contexts = self._update_outliers_txn(
txn, events_and_contexts=events_and_contexts
)
# From this point onwards the events are only events that we haven't
# seen before.
if delete_existing:
# For paranoia reasons, we go and delete all the existing entries
# for these events so we can reinsert them.
# This gets around any problems with some tables already having
# entries.
self._delete_existing_rows_txn(txn, events_and_contexts=events_and_contexts)
self._store_event_txn(txn, events_and_contexts=events_and_contexts)
# Insert into event_to_state_groups.
self._store_event_state_mappings_txn(txn, events_and_contexts)
# We want to store event_auth mappings for rejected events, as they're
# used in state res v2.
# This is only necessary if the rejected event appears in an accepted
# event's auth chain, but its easier for now just to store them (and
# it doesn't take much storage compared to storing the entire event
# anyway).
self._simple_insert_many_txn(
txn,
table="event_auth",
values=[
{
"event_id": event.event_id,
"room_id": event.room_id,
"auth_id": auth_id,
}
for event, _ in events_and_contexts
for auth_id in event.auth_event_ids()
if event.is_state()
],
)
# _store_rejected_events_txn filters out any events which were
# rejected, and returns the filtered list.
events_and_contexts = self._store_rejected_events_txn(
txn, events_and_contexts=events_and_contexts
)
# From this point onwards the events are only ones that weren't
# rejected.
self._update_metadata_tables_txn(
txn,
events_and_contexts=events_and_contexts,
all_events_and_contexts=all_events_and_contexts,
backfilled=backfilled,
)
# We call this last as it assumes we've inserted the events into
# room_memberships, where applicable.
self._update_current_state_txn(txn, state_delta_for_room, min_stream_order)
def _update_current_state_txn(self, txn, state_delta_by_room, stream_id):
for room_id, current_state_tuple in iteritems(state_delta_by_room):
2019-01-30 05:53:17 -05:00
to_delete, to_insert = current_state_tuple
2019-01-30 05:53:17 -05:00
# First we add entries to the current_state_delta_stream. We
# do this before updating the current_state_events table so
# that we can use it to calculate the `prev_event_id`. (This
# allows us to not have to pull out the existing state
# unnecessarily).
#
# The stream_id for the update is chosen to be the minimum of the stream_ids
# for the batch of the events that we are persisting; that means we do not
# end up in a situation where workers see events before the
# current_state_delta updates.
#
2019-01-30 05:53:17 -05:00
sql = """
INSERT INTO current_state_delta_stream
(stream_id, room_id, type, state_key, event_id, prev_event_id)
SELECT ?, ?, ?, ?, ?, (
SELECT event_id FROM current_state_events
WHERE room_id = ? AND type = ? AND state_key = ?
)
2019-01-30 05:53:17 -05:00
"""
txn.executemany(
sql,
2019-01-30 05:53:17 -05:00
(
(
stream_id,
room_id,
etype,
state_key,
None,
room_id,
etype,
state_key,
)
for etype, state_key in to_delete
# We sanity check that we're deleting rather than updating
if (etype, state_key) not in to_insert
),
)
txn.executemany(
sql,
2019-01-30 05:53:17 -05:00
(
(
stream_id,
room_id,
etype,
state_key,
ev_id,
room_id,
etype,
state_key,
)
for (etype, state_key), ev_id in iteritems(to_insert)
),
)
2019-01-30 05:53:17 -05:00
# Now we actually update the current_state_events table
txn.executemany(
"DELETE FROM current_state_events"
" WHERE room_id = ? AND type = ? AND state_key = ?",
(
(room_id, etype, state_key)
for etype, state_key in itertools.chain(to_delete, to_insert)
),
)
# We include the membership in the current state table, hence we do
# a lookup when we insert. This assumes that all events have already
# been inserted into room_memberships.
txn.executemany(
"""INSERT INTO current_state_events
(room_id, type, state_key, event_id, membership)
VALUES (?, ?, ?, ?, (SELECT membership FROM room_memberships WHERE event_id = ?))
""",
[
(room_id, key[0], key[1], ev_id, ev_id)
2019-01-30 05:53:17 -05:00
for key, ev_id in iteritems(to_insert)
],
)
2019-01-30 05:53:17 -05:00
txn.call_after(
self._curr_state_delta_stream_cache.entity_has_changed,
room_id,
stream_id,
2019-01-30 05:53:17 -05:00
)
2017-06-13 04:56:18 -04:00
2019-01-30 05:53:17 -05:00
# Invalidate the various caches
# Figure out the changes of membership to invalidate the
# `get_rooms_for_user` cache.
# We find out which membership events we may have deleted
# and which we have added, then we invlidate the caches for all
# those users.
members_changed = set(
state_key
for ev_type, state_key in itertools.chain(to_delete, to_insert)
if ev_type == EventTypes.Member
)
for member in members_changed:
txn.call_after(
self.get_rooms_for_user_with_stream_ordering.invalidate, (member,)
)
self._invalidate_state_caches_and_stream(txn, room_id, members_changed)
2019-01-30 05:53:17 -05:00
def _update_forward_extremities_txn(
self, txn, new_forward_extremities, max_stream_order
):
for room_id, new_extrem in iteritems(new_forward_extremities):
2017-01-20 09:28:53 -05:00
self._simple_delete_txn(
txn, table="event_forward_extremities", keyvalues={"room_id": room_id}
2017-01-20 09:28:53 -05:00
)
txn.call_after(self.get_latest_event_ids_in_room.invalidate, (room_id,))
2017-01-20 09:28:53 -05:00
self._simple_insert_many_txn(
txn,
table="event_forward_extremities",
values=[
{"event_id": ev_id, "room_id": room_id}
for room_id, new_extrem in iteritems(new_forward_extremities)
2017-01-20 09:28:53 -05:00
for ev_id in new_extrem
],
)
# We now insert into stream_ordering_to_exterm a mapping from room_id,
# new stream_ordering to new forward extremeties in the room.
# This allows us to later efficiently look up the forward extremeties
# for a room before a given stream_ordering
self._simple_insert_many_txn(
txn,
table="stream_ordering_to_exterm",
values=[
{
"room_id": room_id,
"event_id": event_id,
"stream_ordering": max_stream_order,
}
for room_id, new_extrem in iteritems(new_forward_extremities)
2017-01-20 09:28:53 -05:00
for event_id in new_extrem
],
2017-01-20 09:28:53 -05:00
)
@classmethod
def _filter_events_and_contexts_for_duplicates(cls, events_and_contexts):
"""Ensure that we don't have the same event twice.
Pick the earliest non-outlier if there is one, else the earliest one.
Args:
events_and_contexts (list[(EventBase, EventContext)]):
Returns:
list[(EventBase, EventContext)]: filtered list
"""
new_events_and_contexts = OrderedDict()
for event, context in events_and_contexts:
prev_event_context = new_events_and_contexts.get(event.event_id)
if prev_event_context:
if not event.internal_metadata.is_outlier():
if prev_event_context[0].internal_metadata.is_outlier():
# To ensure correct ordering we pop, as OrderedDict is
# ordered by first insertion.
new_events_and_contexts.pop(event.event_id, None)
new_events_and_contexts[event.event_id] = (event, context)
else:
new_events_and_contexts[event.event_id] = (event, context)
return list(new_events_and_contexts.values())
def _update_room_depths_txn(self, txn, events_and_contexts, backfilled):
"""Update min_depth for each room
Args:
txn (twisted.enterprise.adbapi.Connection): db connection
events_and_contexts (list[(EventBase, EventContext)]): events
we are persisting
backfilled (bool): True if the events were backfilled
"""
depth_updates = {}
for event, context in events_and_contexts:
# Remove the any existing cache entries for the event_ids
2015-06-25 12:18:19 -04:00
txn.call_after(self._invalidate_get_event_cache, event.event_id)
if not backfilled:
txn.call_after(
2016-01-28 11:37:41 -05:00
self._events_stream_cache.entity_has_changed,
event.room_id,
event.internal_metadata.stream_ordering,
)
if not event.internal_metadata.is_outlier() and not context.rejected:
depth_updates[event.room_id] = max(
event.depth, depth_updates.get(event.room_id, event.depth)
)
for room_id, depth in iteritems(depth_updates):
2015-06-25 12:18:19 -04:00
self._update_min_depth_for_room_txn(txn, room_id, depth)
def _update_outliers_txn(self, txn, events_and_contexts):
"""Update any outliers with new event info.
This turns outliers into ex-outliers (unless the new event was
rejected).
Args:
txn (twisted.enterprise.adbapi.Connection): db connection
events_and_contexts (list[(EventBase, EventContext)]): events
we are persisting
Returns:
list[(EventBase, EventContext)] new list, without events which
are already in the events table.
"""
2015-06-25 12:18:19 -04:00
txn.execute(
"SELECT event_id, outlier FROM events WHERE event_id in (%s)"
% (",".join(["?"] * len(events_and_contexts)),),
[event.event_id for event, _ in events_and_contexts],
2015-03-20 09:52:56 -04:00
)
have_persisted = {event_id: outlier for event_id, outlier in txn}
2015-06-25 12:18:19 -04:00
to_remove = set()
for event, context in events_and_contexts:
if event.event_id not in have_persisted:
continue
to_remove.add(event)
if context.rejected:
# If the event is rejected then we don't care if the event
# was an outlier or not.
continue
2015-06-25 12:18:19 -04:00
outlier_persisted = have_persisted[event.event_id]
if not event.internal_metadata.is_outlier() and outlier_persisted:
# We received a copy of an event that we had already stored as
# 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 event_to_state_groups.
2016-09-02 05:41:38 -04:00
try:
self._store_event_state_mappings_txn(txn, ((event, context),))
2016-09-02 05:41:38 -04:00
except Exception:
logger.exception("")
raise
2015-03-20 09:52:56 -04:00
metadata_json = encode_json(event.internal_metadata.get_dict())
2015-03-20 09:52:56 -04:00
sql = (
"UPDATE event_json SET internal_metadata = ?" " WHERE event_id = ?"
2015-03-20 09:52:56 -04:00
)
txn.execute(sql, (metadata_json, event.event_id))
2015-03-20 09:52:56 -04:00
# Add an entry to the ex_outlier_stream table to replicate the
# change in outlier status to our workers.
stream_order = event.internal_metadata.stream_ordering
state_group_id = context.state_group
self._simple_insert_txn(
txn,
table="ex_outlier_stream",
values={
"event_stream_ordering": stream_order,
"event_id": event.event_id,
"state_group": state_group_id,
},
)
sql = "UPDATE events SET outlier = ?" " WHERE event_id = ?"
txn.execute(sql, (False, event.event_id))
# Update the event_backward_extremities table now that this
# event isn't an outlier any more.
2017-01-20 09:40:31 -05:00
self._update_backward_extremeties(txn, [event])
return [ec for ec in events_and_contexts if ec[0] not in to_remove]
2015-03-24 12:20:26 -04:00
@classmethod
def _delete_existing_rows_txn(cls, txn, events_and_contexts):
2015-06-25 12:18:19 -04:00
if not events_and_contexts:
# nothing to do here
2015-06-25 12:18:19 -04:00
return
2015-03-20 09:52:56 -04:00
logger.info("Deleting existing")
2015-03-20 09:52:56 -04:00
for table in (
"events",
"event_auth",
"event_json",
"event_edges",
"event_forward_extremities",
"event_reference_hashes",
"event_search",
"event_to_state_groups",
"local_invites",
"state_events",
"rejections",
"redactions",
"room_memberships",
):
txn.executemany(
"DELETE FROM %s WHERE event_id = ?" % (table,),
[(ev.event_id,) for ev, _ in events_and_contexts],
)
for table in ("event_push_actions",):
txn.executemany(
"DELETE FROM %s WHERE room_id = ? AND event_id = ?" % (table,),
[(ev.room_id, ev.event_id) for ev, _ in events_and_contexts],
)
def _store_event_txn(self, txn, events_and_contexts):
"""Insert new events into the event and event_json tables
Args:
txn (twisted.enterprise.adbapi.Connection): db connection
events_and_contexts (list[(EventBase, EventContext)]): events
we are persisting
"""
if not events_and_contexts:
# nothing to do here
return
def event_dict(event):
2017-03-24 06:57:02 -04:00
d = event.get_dict()
d.pop("redacted", None)
d.pop("redacted_because", None)
return d
2015-06-25 12:18:19 -04:00
self._simple_insert_many_txn(
txn,
table="event_json",
values=[
{
"event_id": event.event_id,
"room_id": event.room_id,
"internal_metadata": encode_json(
event.internal_metadata.get_dict()
),
"json": encode_json(event_dict(event)),
"format_version": event.format_version,
2015-06-25 12:18:19 -04:00
}
for event, _ in events_and_contexts
],
2015-04-15 05:24:24 -04:00
)
2015-03-20 09:52:56 -04:00
2015-06-25 12:18:19 -04:00
self._simple_insert_many_txn(
txn,
table="events",
values=[
{
"stream_ordering": event.internal_metadata.stream_ordering,
"topological_ordering": event.depth,
"depth": event.depth,
"event_id": event.event_id,
"room_id": event.room_id,
"type": event.type,
"processed": True,
"outlier": event.internal_metadata.is_outlier(),
"origin_server_ts": int(event.origin_server_ts),
"received_ts": self._clock.time_msec(),
"sender": event.sender,
"contains_url": (
"url" in event.content
2018-08-30 10:19:58 -04:00
and isinstance(event.content["url"], text_type)
),
2015-06-25 12:18:19 -04:00
}
for event, _ in events_and_contexts
],
2015-03-20 09:52:56 -04:00
)
def _store_rejected_events_txn(self, txn, events_and_contexts):
"""Add rows to the 'rejections' table for received events which were
rejected
Args:
txn (twisted.enterprise.adbapi.Connection): db connection
events_and_contexts (list[(EventBase, EventContext)]): events
we are persisting
Returns:
list[(EventBase, EventContext)] new list, without the rejected
events.
"""
# Remove the rejected events from the list now that we've added them
# to the events table and the events_json table.
to_remove = set()
for event, context in events_and_contexts:
if context.rejected:
# Insert the event_id into the rejections table
self._store_rejections_txn(txn, event.event_id, context.rejected)
to_remove.add(event)
return [ec for ec in events_and_contexts if ec[0] not in to_remove]
def _update_metadata_tables_txn(
self, txn, events_and_contexts, all_events_and_contexts, backfilled
):
"""Update all the miscellaneous tables for new events
Args:
txn (twisted.enterprise.adbapi.Connection): db connection
events_and_contexts (list[(EventBase, EventContext)]): events
we are persisting
all_events_and_contexts (list[(EventBase, EventContext)]): all
events that we were going to persist. This includes events
we've already persisted, etc, that wouldn't appear in
events_and_context.
backfilled (bool): True if the events were backfilled
"""
# Insert all the push actions into the event_push_actions table.
self._set_push_actions_for_event_and_users_txn(
txn,
events_and_contexts=events_and_contexts,
all_events_and_contexts=all_events_and_contexts,
)
if not events_and_contexts:
# nothing to do here
return
for event, context in events_and_contexts:
2016-07-26 06:05:39 -04:00
if event.type == EventTypes.Redaction and event.redacts is not None:
# Remove the entries in the event_push_actions table for the
# redacted event.
self._remove_push_actions_for_event_id_txn(
txn, event.room_id, event.redacts
)
# Remove from relations table.
self._handle_redaction(txn, event.redacts)
# Update the event_forward_extremities, event_backward_extremities and
# event_edges tables.
self._handle_mult_prev_events(
txn, events=[event for event, _ in events_and_contexts]
)
for event, _ in events_and_contexts:
if event.type == EventTypes.Name:
# Insert into the event_search table.
self._store_room_name_txn(txn, event)
elif event.type == EventTypes.Topic:
# Insert into the event_search table.
self._store_room_topic_txn(txn, event)
elif event.type == EventTypes.Message:
# Insert into the event_search table.
self._store_room_message_txn(txn, event)
elif event.type == EventTypes.Redaction:
# Insert into the redactions table.
self._store_redaction(txn, event)
self._handle_event_relations(txn, event)
# Insert into the room_memberships table.
self._store_room_members_txn(
txn,
[
event
for event, _ in events_and_contexts
if event.type == EventTypes.Member
],
backfilled=backfilled,
)
# Insert event_reference_hashes table.
2015-06-25 12:18:19 -04:00
self._store_event_reference_hashes_txn(
txn, [event for event, _ in events_and_contexts]
)
state_events_and_contexts = [
ec for ec in events_and_contexts if ec[0].is_state()
]
2015-06-25 12:18:19 -04:00
state_values = []
for event, context in state_events_and_contexts:
2015-03-20 09:52:56 -04:00
vals = {
"event_id": event.event_id,
"room_id": event.room_id,
"type": event.type,
"state_key": event.state_key,
}
# TODO: How does this work with backfilling?
if hasattr(event, "replaces_state"):
vals["prev_state"] = event.replaces_state
2015-06-25 12:18:19 -04:00
state_values.append(vals)
2015-03-20 09:52:56 -04:00
self._simple_insert_many_txn(txn, table="state_events", values=state_values)
2015-03-20 09:52:56 -04:00
2016-07-26 06:05:39 -04:00
# Prefill the event cache
self._add_to_cache(txn, events_and_contexts)
def _add_to_cache(self, txn, events_and_contexts):
to_prefill = []
rows = []
N = 200
for i in range(0, len(events_and_contexts), N):
ev_map = {e[0].event_id: e[0] for e in events_and_contexts[i : i + N]}
if not ev_map:
break
sql = (
"SELECT "
" e.event_id as event_id, "
" r.redacts as redacts,"
" rej.event_id as rejects "
" FROM events as e"
" LEFT JOIN rejections as rej USING (event_id)"
" LEFT JOIN redactions as r ON e.event_id = r.redacts"
" WHERE e.event_id IN (%s)"
) % (",".join(["?"] * len(ev_map)),)
txn.execute(sql, list(ev_map))
rows = self.cursor_to_dict(txn)
for row in rows:
event = ev_map[row["event_id"]]
if not row["rejects"] and not row["redacts"]:
to_prefill.append(
_EventCacheEntry(event=event, redacted_event=None)
)
def prefill():
for cache_entry in to_prefill:
self._get_event_cache.prefill((cache_entry[0].event_id,), cache_entry)
txn.call_after(prefill)
2015-03-20 09:52:56 -04:00
def _store_redaction(self, txn, event):
# invalidate the cache for the redacted event
txn.call_after(self._invalidate_get_event_cache, event.redacts)
2015-03-20 09:52:56 -04:00
txn.execute(
"INSERT INTO redactions (event_id, redacts) VALUES (?,?)",
(event.event_id, event.redacts),
2015-03-20 09:52:56 -04:00
)
2019-08-29 12:38:51 -04:00
@defer.inlineCallbacks
def _censor_redactions(self):
2019-09-09 10:08:24 -04:00
"""Censors all redactions older than the configured period that haven't
been censored yet.
2019-08-29 12:38:51 -04:00
By censor we mean update the event_json table with the redacted event.
Returns:
Deferred
"""
2019-09-09 08:23:41 -04:00
if self.hs.config.redaction_retention_period is None:
2019-08-29 12:38:51 -04:00
return
max_pos = yield self.find_first_stream_ordering_after_ts(
self._clock.time_msec() - self.hs.config.redaction_retention_period
)
2019-08-29 12:38:51 -04:00
2019-09-09 08:31:00 -04:00
# We fetch all redactions that:
2019-09-09 10:08:24 -04:00
# 1. point to an event we have,
2019-09-09 08:31:00 -04:00
# 2. has a stream ordering from before the cut off, and
# 3. we haven't yet censored.
#
# This is limited to 100 events to ensure that we don't try and do too
# much at once. We'll get called again so this should eventually catch
# up.
#
# We use the range [-max_pos, max_pos] to handle backfilled events,
# which are given negative stream ordering.
2019-08-29 12:38:51 -04:00
sql = """
2019-09-09 08:33:38 -04:00
SELECT redact_event.event_id, redacts FROM redactions
INNER JOIN events AS redact_event USING (event_id)
INNER JOIN events AS original_event ON (
redact_event.room_id = original_event.room_id
AND redacts = original_event.event_id
)
2019-08-29 12:38:51 -04:00
WHERE NOT have_censored
2019-09-09 08:33:38 -04:00
AND ? <= redact_event.stream_ordering AND redact_event.stream_ordering <= ?
ORDER BY redact_event.stream_ordering ASC
2019-08-29 12:38:51 -04:00
LIMIT ?
"""
rows = yield self._execute(
"_censor_redactions_fetch", None, sql, -max_pos, max_pos, 100
)
updates = []
for redaction_id, event_id in rows:
redaction_event = yield self.get_event(redaction_id, allow_none=True)
original_event = yield self.get_event(
event_id, allow_rejected=True, allow_none=True
)
# The SQL above ensures that we have both the redaction and
# original event, so if the `get_event` calls return None it
# means that the redaction wasn't allowed. Either way we know that
# the result won't change so we mark the fact that we've checked.
if (
redaction_event
and original_event
and original_event.internal_metadata.is_redacted()
):
# Redaction was allowed
pruned_json = encode_canonical_json(
prune_event_dict(original_event.get_dict())
)
else:
# Redaction wasn't allowed
pruned_json = None
updates.append((redaction_id, event_id, pruned_json))
def _update_censor_txn(txn):
for redaction_id, event_id, pruned_json in updates:
if pruned_json:
self._simple_update_one_txn(
txn,
table="event_json",
keyvalues={"event_id": event_id},
updatevalues={"json": pruned_json},
)
self._simple_update_one_txn(
txn,
table="redactions",
keyvalues={"event_id": redaction_id},
updatevalues={"have_censored": True},
)
yield self.runInteraction("_update_censor_txn", _update_censor_txn)
@defer.inlineCallbacks
def count_daily_messages(self):
2015-09-22 08:29:36 -04:00
"""
Returns an estimate of the number of messages sent in the last day.
If it has been significantly less or more than one day since the last
call to this function, it will return None.
"""
def _count_messages(txn):
2017-06-14 14:37:17 -04:00
sql = """
SELECT COALESCE(COUNT(*), 0) FROM events
WHERE type = 'm.room.message'
AND stream_ordering > ?
"""
txn.execute(sql, (self.stream_ordering_day_ago,))
count, = txn.fetchone()
return count
ret = yield self.runInteraction("count_messages", _count_messages)
return ret
2017-06-14 14:37:17 -04:00
@defer.inlineCallbacks
def count_daily_sent_messages(self):
def _count_messages(txn):
# This is good enough as if you have silly characters in your own
# hostname then thats your own fault.
like_clause = "%:" + self.hs.hostname
sql = """
SELECT COALESCE(COUNT(*), 0) FROM events
WHERE type = 'm.room.message'
AND sender LIKE ?
AND stream_ordering > ?
"""
txn.execute(sql, (like_clause, self.stream_ordering_day_ago))
2017-06-14 14:37:17 -04:00
count, = txn.fetchone()
return count
ret = yield self.runInteraction("count_daily_sent_messages", _count_messages)
return ret
2017-06-15 04:39:39 -04:00
@defer.inlineCallbacks
def count_daily_active_rooms(self):
def _count(txn):
sql = """
SELECT COALESCE(COUNT(DISTINCT room_id), 0) FROM events
WHERE type = 'm.room.message'
AND stream_ordering > ?
"""
txn.execute(sql, (self.stream_ordering_day_ago,))
count, = txn.fetchone()
return count
ret = yield self.runInteraction("count_daily_active_rooms", _count)
return ret
def get_current_backfill_token(self):
"""The current minimum token that backfilled events have reached"""
return -self._backfill_id_gen.get_current_token()
def get_current_events_token(self):
"""The current maximum token that events have reached"""
return self._stream_id_gen.get_current_token()
def get_all_new_forward_event_rows(self, last_id, current_id, limit):
if last_id == current_id:
return defer.succeed([])
def get_all_new_forward_event_rows(txn):
sql = (
"SELECT e.stream_ordering, e.event_id, e.room_id, e.type,"
2019-05-16 05:18:53 -04:00
" state_key, redacts, relates_to_id"
" FROM events AS e"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
2019-05-16 05:18:53 -04:00
" LEFT JOIN event_relations USING (event_id)"
" WHERE ? < stream_ordering AND stream_ordering <= ?"
" ORDER BY stream_ordering ASC"
" LIMIT ?"
)
txn.execute(sql, (last_id, current_id, limit))
new_event_updates = txn.fetchall()
if len(new_event_updates) == limit:
upper_bound = new_event_updates[-1][0]
else:
upper_bound = current_id
sql = (
"SELECT event_stream_ordering, e.event_id, e.room_id, e.type,"
2019-05-16 05:18:53 -04:00
" state_key, redacts, relates_to_id"
" FROM events AS e"
" INNER JOIN ex_outlier_stream USING (event_id)"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
2019-05-16 05:18:53 -04:00
" LEFT JOIN event_relations USING (event_id)"
" WHERE ? < event_stream_ordering"
" AND event_stream_ordering <= ?"
" ORDER BY event_stream_ordering DESC"
)
txn.execute(sql, (last_id, upper_bound))
new_event_updates.extend(txn)
return new_event_updates
return self.runInteraction(
"get_all_new_forward_event_rows", get_all_new_forward_event_rows
)
def get_all_new_backfill_event_rows(self, last_id, current_id, limit):
if last_id == current_id:
return defer.succeed([])
def get_all_new_backfill_event_rows(txn):
sql = (
"SELECT -e.stream_ordering, e.event_id, e.room_id, e.type,"
2019-05-16 05:18:53 -04:00
" state_key, redacts, relates_to_id"
" FROM events AS e"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
2019-05-16 05:18:53 -04:00
" LEFT JOIN event_relations USING (event_id)"
" WHERE ? > stream_ordering AND stream_ordering >= ?"
" ORDER BY stream_ordering ASC"
" LIMIT ?"
)
txn.execute(sql, (-last_id, -current_id, limit))
new_event_updates = txn.fetchall()
if len(new_event_updates) == limit:
upper_bound = new_event_updates[-1][0]
else:
upper_bound = current_id
sql = (
"SELECT -event_stream_ordering, e.event_id, e.room_id, e.type,"
2019-05-16 05:18:53 -04:00
" state_key, redacts, relates_to_id"
" FROM events AS e"
" INNER JOIN ex_outlier_stream USING (event_id)"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
2019-05-16 05:18:53 -04:00
" LEFT JOIN event_relations USING (event_id)"
" WHERE ? > event_stream_ordering"
" AND event_stream_ordering >= ?"
" ORDER BY event_stream_ordering DESC"
)
txn.execute(sql, (-last_id, -upper_bound))
new_event_updates.extend(txn.fetchall())
return new_event_updates
return self.runInteraction(
"get_all_new_backfill_event_rows", get_all_new_backfill_event_rows
)
2017-02-01 05:42:37 -05:00
@cached(num_args=5, max_entries=10)
def get_all_new_events(
self,
last_backfill_id,
last_forward_id,
current_backfill_id,
current_forward_id,
limit,
):
"""Get all the new events that have arrived at the server either as
new events or as backfilled events"""
have_backfill_events = last_backfill_id != current_backfill_id
have_forward_events = last_forward_id != current_forward_id
if not have_backfill_events and not have_forward_events:
return defer.succeed(AllNewEventsResult([], [], [], [], []))
def get_all_new_events_txn(txn):
sql = (
"SELECT e.stream_ordering, e.event_id, e.room_id, e.type,"
" state_key, redacts"
" FROM events AS e"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
" WHERE ? < stream_ordering AND stream_ordering <= ?"
" ORDER BY stream_ordering ASC"
" LIMIT ?"
)
if have_forward_events:
txn.execute(sql, (last_forward_id, current_forward_id, limit))
new_forward_events = txn.fetchall()
if len(new_forward_events) == limit:
upper_bound = new_forward_events[-1][0]
else:
upper_bound = current_forward_id
sql = (
"SELECT event_stream_ordering, event_id, state_group"
" FROM ex_outlier_stream"
" WHERE ? > event_stream_ordering"
" AND event_stream_ordering >= ?"
" ORDER BY event_stream_ordering DESC"
)
txn.execute(sql, (last_forward_id, upper_bound))
forward_ex_outliers = txn.fetchall()
else:
new_forward_events = []
forward_ex_outliers = []
sql = (
"SELECT -e.stream_ordering, e.event_id, e.room_id, e.type,"
" state_key, redacts"
" FROM events AS e"
" LEFT JOIN redactions USING (event_id)"
" LEFT JOIN state_events USING (event_id)"
" WHERE ? > stream_ordering AND stream_ordering >= ?"
" ORDER BY stream_ordering DESC"
" LIMIT ?"
)
if have_backfill_events:
txn.execute(sql, (-last_backfill_id, -current_backfill_id, limit))
new_backfill_events = txn.fetchall()
if len(new_backfill_events) == limit:
upper_bound = new_backfill_events[-1][0]
else:
upper_bound = current_backfill_id
sql = (
"SELECT -event_stream_ordering, event_id, state_group"
" FROM ex_outlier_stream"
" WHERE ? > event_stream_ordering"
" AND event_stream_ordering >= ?"
" ORDER BY event_stream_ordering DESC"
)
txn.execute(sql, (-last_backfill_id, -upper_bound))
backward_ex_outliers = txn.fetchall()
else:
new_backfill_events = []
backward_ex_outliers = []
return AllNewEventsResult(
new_forward_events,
new_backfill_events,
forward_ex_outliers,
backward_ex_outliers,
)
return self.runInteraction("get_all_new_events", get_all_new_events_txn)
def purge_history(self, room_id, token, delete_local_events):
"""Deletes room history before a certain point
Args:
room_id (str):
token (str): A topological token to delete events before
2016-07-05 05:28:51 -04:00
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).
"""
2016-07-05 05:28:51 -04:00
return self.runInteraction(
"purge_history",
self._purge_history_txn,
room_id,
token,
delete_local_events,
2016-07-05 05:28:51 -04:00
)
def _purge_history_txn(self, txn, room_id, token_str, delete_local_events):
2018-05-16 05:52:06 -04:00
token = RoomStreamToken.parse(token_str)
# Tables that should be pruned:
# event_auth
# event_backward_extremities
# event_edges
# event_forward_extremities
# event_json
# event_push_actions
# event_reference_hashes
# event_search
# event_to_state_groups
# events
# rejections
# room_depth
# state_groups
# state_groups_state
# we will build a temporary table listing the events so that we don't
# have to keep shovelling the list back and forth across the
# connection. Annoyingly the python sqlite driver commits the
# transaction on CREATE, so let's do this first.
#
# furthermore, we might already have the table from a previous (failed)
# purge attempt, so let's drop the table first.
txn.execute("DROP TABLE IF EXISTS events_to_purge")
txn.execute(
"CREATE TEMPORARY TABLE events_to_purge ("
" event_id TEXT NOT NULL,"
" should_delete BOOLEAN NOT NULL"
")"
)
# First ensure that we're not about to delete all the forward extremeties
txn.execute(
"SELECT e.event_id, e.depth FROM events as e "
"INNER JOIN event_forward_extremities as f "
"ON e.event_id = f.event_id "
"AND e.room_id = f.room_id "
"WHERE f.room_id = ?",
(room_id,),
)
rows = txn.fetchall()
2018-08-30 10:19:58 -04:00
max_depth = max(row[1] for row in rows)
2018-08-30 10:19:58 -04:00
if max_depth < token.topological:
# We need to ensure we don't delete all the events from the database
2016-07-07 06:42:15 -04:00
# otherwise we wouldn't be able to send any events (due to not
# having any backwards extremeties)
2016-07-07 06:41:07 -04:00
raise SynapseError(
400, "topological_ordering is greater than forward extremeties"
)
logger.info("[purge] looking for events to delete")
should_delete_expr = "state_key IS NULL"
should_delete_params = ()
if not delete_local_events:
should_delete_expr += " AND event_id NOT LIKE ?"
2018-09-13 11:10:56 -04:00
# We include the parameter twice since we use the expression twice
should_delete_params += ("%:" + self.hs.hostname, "%:" + self.hs.hostname)
should_delete_params += (room_id, token.topological)
# Note that we insert events that are outliers and aren't going to be
# deleted, as nothing will happen to them.
txn.execute(
"INSERT INTO events_to_purge"
" SELECT event_id, %s"
" FROM events AS e LEFT JOIN state_events USING (event_id)"
" WHERE (NOT outlier OR (%s)) AND e.room_id = ? AND topological_ordering < ?"
% (should_delete_expr, should_delete_expr),
should_delete_params,
)
2018-09-13 10:05:52 -04:00
# We create the indices *after* insertion as that's a lot faster.
# create an index on should_delete because later we'll be looking for
# the should_delete / shouldn't_delete subsets
txn.execute(
"CREATE INDEX events_to_purge_should_delete"
" ON events_to_purge(should_delete)"
2018-09-13 10:05:52 -04:00
)
# We do joins against events_to_purge for e.g. calculating state
# groups to purge, etc., so lets make an index.
txn.execute("CREATE INDEX events_to_purge_id" " ON events_to_purge(event_id)")
2018-09-13 10:05:52 -04:00
txn.execute("SELECT event_id, should_delete FROM events_to_purge")
event_rows = txn.fetchall()
logger.info(
"[purge] found %i events before cutoff, of which %i can be deleted",
len(event_rows),
sum(1 for e in event_rows if e[1]),
)
2016-09-02 05:41:38 -04:00
logger.info("[purge] Finding new backward extremities")
2017-05-10 12:46:41 -04:00
# We calculate the new entries for the backward extremeties by finding
2018-05-16 06:47:26 -04:00
# events to be purged that are pointed to by events we're not going to
# purge.
txn.execute(
"SELECT DISTINCT e.event_id FROM events_to_purge AS e"
" INNER JOIN event_edges AS ed ON e.event_id = ed.prev_event_id"
" LEFT JOIN events_to_purge AS ep2 ON ed.event_id = ep2.event_id"
" WHERE ep2.event_id IS NULL"
)
new_backwards_extrems = txn.fetchall()
logger.info("[purge] replacing backward extremities: %r", new_backwards_extrems)
2017-05-10 12:46:41 -04:00
txn.execute(
"DELETE FROM event_backward_extremities WHERE room_id = ?", (room_id,)
)
# Update backward extremeties
txn.executemany(
"INSERT INTO event_backward_extremities (room_id, event_id)"
" VALUES (?, ?)",
[(room_id, event_id) for event_id, in new_backwards_extrems],
)
logger.info("[purge] finding redundant state groups")
2017-05-10 12:46:41 -04:00
2018-10-12 15:43:18 -04:00
# Get all state groups that are referenced by events that are to be
# deleted. We then go and check if they are referenced by other events
# or state groups, and if not we delete them.
txn.execute(
"""
2018-10-12 15:43:18 -04:00
SELECT DISTINCT state_group FROM events_to_purge
INNER JOIN event_to_state_groups USING (event_id)
"""
)
2018-10-19 11:06:59 -04:00
referenced_state_groups = set(sg for sg, in txn)
2018-10-12 15:43:18 -04:00
logger.info(
"[purge] found %i referenced state groups", len(referenced_state_groups)
2018-10-12 15:43:18 -04:00
)
2018-10-12 15:43:18 -04:00
logger.info("[purge] finding state groups that can be deleted")
_ = self._find_unreferenced_groups_during_purge(txn, referenced_state_groups)
state_groups_to_delete, remaining_state_groups = _
logger.info(
"[purge] found %i state groups to delete", len(state_groups_to_delete)
)
logger.info(
"[purge] de-delta-ing %i remaining state groups",
len(remaining_state_groups),
)
# Now we turn the state groups that reference to-be-deleted state
# groups to non delta versions.
for sg in remaining_state_groups:
logger.info("[purge] de-delta-ing remaining state group %s", sg)
curr_state = self._get_state_groups_from_groups_txn(txn, [sg])
curr_state = curr_state[sg]
self._simple_delete_txn(
txn, table="state_groups_state", keyvalues={"state_group": sg}
)
self._simple_delete_txn(
txn, table="state_group_edges", keyvalues={"state_group": sg}
)
self._simple_insert_many_txn(
txn,
table="state_groups_state",
values=[
{
"state_group": sg,
"room_id": room_id,
"type": key[0],
"state_key": key[1],
"event_id": state_id,
}
for key, state_id in iteritems(curr_state)
],
)
logger.info("[purge] removing redundant state groups")
txn.executemany(
"DELETE FROM state_groups_state WHERE state_group = ?",
2018-10-12 15:43:18 -04:00
((sg,) for sg in state_groups_to_delete),
)
txn.executemany(
"DELETE FROM state_groups WHERE id = ?",
2018-10-12 15:43:18 -04:00
((sg,) for sg in state_groups_to_delete),
)
2017-05-10 12:46:41 -04:00
logger.info("[purge] removing events from event_to_state_groups")
txn.execute(
"DELETE FROM event_to_state_groups "
"WHERE event_id IN (SELECT event_id from events_to_purge)"
)
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 (
"events",
"event_json",
"event_auth",
"event_edges",
"event_forward_extremities",
"event_reference_hashes",
"event_search",
"rejections",
):
logger.info("[purge] removing events from %s", table)
2017-05-10 12:46:41 -04:00
txn.execute(
"DELETE FROM %s WHERE event_id IN ("
" SELECT event_id FROM events_to_purge WHERE should_delete"
")" % (table,)
)
# event_push_actions lacks an index on event_id, and has one on
# (room_id, event_id) instead.
for table in ("event_push_actions",):
logger.info("[purge] removing events from %s", table)
txn.execute(
"DELETE FROM %s WHERE room_id = ? AND event_id IN ("
" SELECT event_id FROM events_to_purge WHERE should_delete"
")" % (table,),
(room_id,),
)
# Mark all state and own events as outliers
logger.info("[purge] marking remaining events as outliers")
txn.execute(
"UPDATE events SET outlier = ?"
" WHERE event_id IN ("
" SELECT event_id FROM events_to_purge "
" WHERE NOT should_delete"
")",
(True,),
)
# 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.
2018-05-16 06:13:31 -04:00
#
# We do this by calculating the minimum depth of the backwards
# extremities. However, the events in event_backward_extremities
# are ones we don't have yet so we need to look at the events that
# point to it via event_edges table.
txn.execute(
"""
SELECT COALESCE(MIN(depth), 0)
FROM event_backward_extremities AS eb
INNER JOIN event_edges AS eg ON eg.prev_event_id = eb.event_id
INNER JOIN events AS e ON e.event_id = eg.event_id
WHERE eb.room_id = ?
""",
(room_id,),
)
min_depth, = txn.fetchone()
logger.info("[purge] updating room_depth to %d", min_depth)
txn.execute(
"UPDATE room_depth SET min_depth = ? WHERE room_id = ?",
(min_depth, room_id),
)
# finally, drop the temp table. this will commit the txn in sqlite,
# so make sure to keep this actually last.
txn.execute("DROP TABLE events_to_purge")
logger.info("[purge] done")
2017-05-10 12:46:41 -04:00
2018-10-29 10:23:34 -04:00
def _find_unreferenced_groups_during_purge(self, txn, state_groups):
"""Used when purging history to figure out which state groups can be
deleted and which need to be de-delta'ed (due to one of its prev groups
being scheduled for deletion).
Args:
txn
state_groups (set[int]): Set of state groups referenced by events
that are going to be deleted.
Returns:
tuple[set[int], set[int]]: The set of state groups that can be
deleted and the set of state groups that need to be de-delta'ed
"""
# Graph of state group -> previous group
graph = {}
# Set of events that we have found to be referenced by events
referenced_groups = set()
# Set of state groups we've already seen
state_groups_seen = set(state_groups)
# Set of state groups to handle next.
next_to_search = set(state_groups)
while next_to_search:
# We bound size of groups we're looking up at once, to stop the
# SQL query getting too big
if len(next_to_search) < 100:
current_search = next_to_search
next_to_search = set()
else:
current_search = set(itertools.islice(next_to_search, 100))
next_to_search -= current_search
# Check if state groups are referenced
sql = """
SELECT DISTINCT state_group FROM event_to_state_groups
LEFT JOIN events_to_purge AS ep USING (event_id)
WHERE state_group IN (%s) AND ep.event_id IS NULL
""" % (
",".join("?" for _ in current_search),
)
2018-10-29 10:23:34 -04:00
txn.execute(sql, list(current_search))
referenced = set(sg for sg, in txn)
referenced_groups |= referenced
# We don't continue iterating up the state group graphs for state
# groups that are referenced.
current_search -= referenced
rows = self._simple_select_many_txn(
txn,
table="state_group_edges",
column="prev_state_group",
iterable=current_search,
keyvalues={},
retcols=("prev_state_group", "state_group"),
2018-10-29 10:23:34 -04:00
)
prevs = set(row["state_group"] for row in rows)
# We don't bother re-handling groups we've already seen
prevs -= state_groups_seen
next_to_search |= prevs
state_groups_seen |= prevs
for row in rows:
# Note: Each state group can have at most one prev group
graph[row["state_group"]] = row["prev_state_group"]
to_delete = state_groups_seen - referenced_groups
to_dedelta = set()
for sg in referenced_groups:
prev_sg = graph.get(sg)
if prev_sg and prev_sg in to_delete:
to_dedelta.add(sg)
return to_delete, to_dedelta
2019-08-22 05:42:59 -04:00
def purge_room(self, room_id):
"""Deletes all record of a room
Args:
room_id (str):
"""
return self.runInteraction("purge_room", self._purge_room_txn, room_id)
def _purge_room_txn(self, txn, room_id):
# first we have to delete the state groups states
logger.info("[purge] removing %s from state_groups_state", room_id)
txn.execute(
"""
DELETE FROM state_groups_state WHERE state_group IN (
SELECT state_group FROM events JOIN event_to_state_groups USING(event_id)
WHERE events.room_id=?
)
""",
(room_id,),
)
# ... and the state group edges
logger.info("[purge] removing %s from state_group_edges", room_id)
txn.execute(
"""
DELETE FROM state_group_edges WHERE state_group IN (
SELECT state_group FROM events JOIN event_to_state_groups USING(event_id)
WHERE events.room_id=?
)
""",
(room_id,),
)
# ... and the state groups
logger.info("[purge] removing %s from state_groups", room_id)
txn.execute(
"""
DELETE FROM state_groups WHERE id IN (
SELECT state_group FROM events JOIN event_to_state_groups USING(event_id)
WHERE events.room_id=?
)
""",
(room_id,),
)
# and then tables which lack an index on room_id but have one on event_id
for table in (
"event_auth",
"event_edges",
"event_push_actions_staging",
"event_reference_hashes",
"event_relations",
"event_to_state_groups",
"redactions",
"rejections",
"state_events",
):
logger.info("[purge] removing %s from %s", room_id, table)
txn.execute(
"""
DELETE FROM %s WHERE event_id IN (
SELECT event_id FROM events WHERE room_id=?
)
"""
% (table,),
(room_id,),
)
# and finally, the tables with an index on room_id (or no useful index)
for table in (
"current_state_events",
"event_backward_extremities",
"event_forward_extremities",
"event_json",
"event_push_actions",
"event_search",
"events",
"group_rooms",
"public_room_list_stream",
"receipts_graph",
"receipts_linearized",
"room_aliases",
"room_depth",
"room_memberships",
"room_stats_state",
"room_stats_current",
"room_stats_historical",
2019-08-22 05:42:59 -04:00
"room_stats_earliest_token",
"rooms",
"stream_ordering_to_exterm",
"topics",
"users_in_public_rooms",
"users_who_share_private_rooms",
# no useful index, but let's clear them anyway
"appservice_room_list",
"e2e_room_keys",
"event_push_summary",
"pusher_throttle",
"group_summary_rooms",
"local_invites",
"room_account_data",
"room_tags",
):
logger.info("[purge] removing %s from %s", room_id, table)
txn.execute("DELETE FROM %s WHERE room_id=?" % (table,), (room_id,))
# Other tables we do NOT need to clear out:
#
# - blocked_rooms
# This is important, to make sure that we don't accidentally rejoin a blocked
# room after it was purged
#
# - user_directory
# This has a room_id column, but it is unused
#
# Other tables that we might want to consider clearing out include:
#
# - event_reports
# Given that these are intended for abuse management my initial
# inclination is to leave them in place.
#
# - current_state_delta_stream
# - ex_outlier_stream
# - room_tags_revisions
# The problem with these is that they are largeish and there is no room_id
# index on them. In any case we should be clearing out 'stream' tables
# periodically anyway (#5888)
# TODO: we could probably usefully do a bunch of cache invalidation here
logger.info("[purge] done")
@defer.inlineCallbacks
def is_event_after(self, event_id1, event_id2):
2017-04-12 09:36:20 -04:00
"""Returns True if event_id1 is after event_id2 in the stream
"""
to_1, so_1 = yield self._get_event_ordering(event_id1)
to_2, so_2 = yield self._get_event_ordering(event_id2)
return (to_1, so_1) > (to_2, so_2)
@cachedInlineCallbacks(max_entries=5000)
def _get_event_ordering(self, event_id):
res = yield self._simple_select_one(
table="events",
retcols=["topological_ordering", "stream_ordering"],
keyvalues={"event_id": event_id},
allow_none=True,
)
if not res:
raise SynapseError(404, "Could not find event %s" % (event_id,))
return (int(res["topological_ordering"]), int(res["stream_ordering"]))
2017-06-15 07:47:05 -04:00
def get_all_updated_current_state_deltas(self, from_token, to_token, limit):
def get_all_updated_current_state_deltas_txn(txn):
sql = """
SELECT stream_id, room_id, type, state_key, event_id
FROM current_state_delta_stream
WHERE ? < stream_id AND stream_id <= ?
ORDER BY stream_id ASC LIMIT ?
"""
txn.execute(sql, (from_token, to_token, limit))
return txn.fetchall()
2017-06-15 07:47:05 -04:00
return self.runInteraction(
"get_all_updated_current_state_deltas",
get_all_updated_current_state_deltas_txn,
)
2017-04-11 13:35:45 -04:00
AllNewEventsResult = namedtuple(
"AllNewEventsResult",
[
"new_forward_events",
"new_backfill_events",
"forward_ex_outliers",
"backward_ex_outliers",
],
)