MatrixSynapse/synapse/storage/events.py

1061 lines
36 KiB
Python
Raw Normal View History

2015-03-20 14:52:56 +01:00
# -*- coding: utf-8 -*-
2016-01-07 05:26:29 +01:00
# Copyright 2014-2016 OpenMarket Ltd
2015-03-20 14:52:56 +01:00
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from _base import SQLBaseStore, _RollbackButIsFineException
from twisted.internet import defer, reactor
2015-03-20 14:52:56 +01:00
from synapse.events import FrozenEvent, USE_FROZEN_DICTS
from synapse.events.utils import prune_event
2016-02-04 11:22:44 +01:00
from synapse.util.logcontext import preserve_fn, PreserveLoggingContext
2015-03-20 14:52:56 +01:00
from synapse.util.logutils import log_function
from synapse.api.constants import EventTypes
from canonicaljson import encode_canonical_json
from contextlib import contextmanager
2015-03-20 14:52:56 +01:00
import logging
import math
import ujson as json
2015-03-20 14:52:56 +01:00
logger = logging.getLogger(__name__)
def encode_json(json_object):
if USE_FROZEN_DICTS:
2015-08-24 18:18:58 +02:00
# ujson doesn't like frozen_dicts
return encode_canonical_json(json_object)
else:
return json.dumps(json_object, ensure_ascii=False)
2015-05-21 12:18:22 +02:00
# These values are used in the `enqueus_event` and `_do_fetch` methods to
# control how we batch/bulk fetch events from the database.
# The values are plucked out of thing air to make initial sync run faster
# on jki.re
# TODO: Make these configurable.
EVENT_QUEUE_THREADS = 3 # Max number of threads that will fetch events
2015-05-21 15:52:23 +02:00
EVENT_QUEUE_ITERATIONS = 3 # No. times we block waiting for requests for events
2015-05-21 15:50:57 +02:00
EVENT_QUEUE_TIMEOUT_S = 0.1 # Timeout when waiting for requests for events
2015-05-21 12:18:22 +02:00
2015-03-20 14:52:56 +01:00
class EventsStore(SQLBaseStore):
EVENT_ORIGIN_SERVER_TS_NAME = "event_origin_server_ts"
def __init__(self, hs):
super(EventsStore, self).__init__(hs)
self.register_background_update_handler(
self.EVENT_ORIGIN_SERVER_TS_NAME, self._background_reindex_origin_server_ts
)
2015-06-25 18:18:19 +02:00
@defer.inlineCallbacks
def persist_events(self, events_and_contexts, backfilled=False,
is_new_state=True):
if not events_and_contexts:
return
if backfilled:
start = self.min_stream_token - 1
self.min_stream_token -= len(events_and_contexts) + 1
stream_orderings = range(start, self.min_stream_token, -1)
2015-06-25 18:18:19 +02:00
@contextmanager
def stream_ordering_manager():
yield stream_orderings
stream_ordering_manager = stream_ordering_manager()
else:
stream_ordering_manager = yield self._stream_id_gen.get_next_mult(
self, len(events_and_contexts)
)
with stream_ordering_manager as stream_orderings:
for (event, _), stream in zip(events_and_contexts, stream_orderings):
event.internal_metadata.stream_ordering = stream
chunks = [
2016-02-02 18:18:50 +01:00
events_and_contexts[x:x + 100]
2015-06-25 18:18:19 +02:00
for x in xrange(0, len(events_and_contexts), 100)
]
for chunk in chunks:
# We can't easily parallelize these since different chunks
# might contain the same event. :(
yield self.runInteraction(
"persist_events",
self._persist_events_txn,
events_and_contexts=chunk,
backfilled=backfilled,
is_new_state=is_new_state,
)
2015-03-20 14:52:56 +01:00
@defer.inlineCallbacks
@log_function
def persist_event(self, event, context, backfilled=False,
is_new_state=True, current_state=None):
stream_ordering = None
if backfilled:
self.min_stream_token -= 1
stream_ordering = self.min_stream_token
2015-03-20 14:52:56 +01:00
if stream_ordering is None:
stream_ordering_manager = yield self._stream_id_gen.get_next(self)
else:
@contextmanager
def stream_ordering_manager():
yield stream_ordering
2015-05-12 17:12:37 +02:00
stream_ordering_manager = stream_ordering_manager()
2015-03-20 14:52:56 +01:00
try:
with stream_ordering_manager as stream_ordering:
2015-06-25 18:18:19 +02:00
event.internal_metadata.stream_ordering = stream_ordering
yield self.runInteraction(
"persist_event",
self._persist_event_txn,
event=event,
context=context,
backfilled=backfilled,
is_new_state=is_new_state,
current_state=current_state,
)
2015-03-20 14:52:56 +01:00
except _RollbackButIsFineException:
pass
max_persisted_id = yield self._stream_id_gen.get_max_token(self)
defer.returnValue((stream_ordering, max_persisted_id))
2015-03-20 14:52:56 +01:00
@defer.inlineCallbacks
def get_event(self, event_id, check_redacted=True,
get_prev_content=False, allow_rejected=False,
allow_none=False):
"""Get an event from the database by event_id.
Args:
event_id (str): The event_id of the event to fetch
check_redacted (bool): If True, check if event has been redacted
and redact it.
get_prev_content (bool): If True and event is a state event,
include the previous states content in the unsigned field.
allow_rejected (bool): If True return rejected events.
allow_none (bool): If True, return None if no event found, if
False throw an exception.
Returns:
Deferred : A FrozenEvent.
"""
events = yield self._get_events(
[event_id],
2015-03-20 14:52:56 +01:00
check_redacted=check_redacted,
get_prev_content=get_prev_content,
allow_rejected=allow_rejected,
)
if not events and not allow_none:
2015-03-20 14:52:56 +01:00
raise RuntimeError("Could not find event %s" % (event_id,))
defer.returnValue(events[0] if events else None)
2015-03-20 14:52:56 +01:00
@log_function
def _persist_event_txn(self, txn, event, context, backfilled,
2015-06-25 18:18:19 +02:00
is_new_state=True, current_state=None):
2015-03-20 14:52:56 +01:00
# We purposefully do this first since if we include a `current_state`
# key, we *want* to update the `current_state_events` table
if current_state:
txn.call_after(self.get_current_state_for_key.invalidate_all)
txn.call_after(self.get_rooms_for_user.invalidate_all)
txn.call_after(self.get_users_in_room.invalidate, (event.room_id,))
txn.call_after(self.get_joined_hosts_for_room.invalidate, (event.room_id,))
txn.call_after(self.get_room_name_and_aliases, event.room_id)
self._simple_delete_txn(
txn,
table="current_state_events",
keyvalues={"room_id": event.room_id},
2015-03-20 14:52:56 +01:00
)
for s in current_state:
self._simple_insert_txn(
txn,
"current_state_events",
{
"event_id": s.event_id,
"room_id": s.room_id,
"type": s.type,
"state_key": s.state_key,
2015-05-05 16:13:25 +02:00
}
2015-03-20 14:52:56 +01:00
)
2015-06-25 18:18:19 +02:00
return self._persist_events_txn(
txn,
[(event, context)],
backfilled=backfilled,
is_new_state=is_new_state,
)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
@log_function
def _persist_events_txn(self, txn, events_and_contexts, backfilled,
is_new_state=True):
# Remove the any existing cache entries for the event_ids
for event, _ in events_and_contexts:
txn.call_after(self._invalidate_get_event_cache, event.event_id)
if not backfilled:
txn.call_after(
2016-01-28 17:37:41 +01:00
self._events_stream_cache.entity_has_changed,
event.room_id, event.internal_metadata.stream_ordering,
)
2015-06-25 18:18:19 +02:00
depth_updates = {}
for event, _ in events_and_contexts:
if event.internal_metadata.is_outlier():
continue
depth_updates[event.room_id] = max(
event.depth, depth_updates.get(event.room_id, event.depth)
2015-03-20 14:52:56 +01:00
)
2015-06-25 18:18:19 +02:00
for room_id, depth in depth_updates.items():
self._update_min_depth_for_room_txn(txn, room_id, depth)
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 14:52:56 +01:00
)
2015-06-25 18:18:19 +02:00
have_persisted = {
event_id: outlier
for event_id, outlier in txn.fetchall()
}
event_map = {}
to_remove = set()
for event, context in events_and_contexts:
# Handle the case of the list including the same event multiple
# times. The tricky thing here is when they differ by whether
# they are an outlier.
if event.event_id in event_map:
other = event_map[event.event_id]
if not other.internal_metadata.is_outlier():
to_remove.add(event)
continue
elif not event.internal_metadata.is_outlier():
to_remove.add(event)
continue
else:
to_remove.add(other)
event_map[event.event_id] = event
if event.event_id not in have_persisted:
continue
to_remove.add(event)
outlier_persisted = have_persisted[event.event_id]
if not event.internal_metadata.is_outlier() and outlier_persisted:
self._store_state_groups_txn(
txn, event, context,
)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
metadata_json = encode_json(
event.internal_metadata.get_dict()
2015-06-25 18:18:19 +02:00
).decode("UTF-8")
2015-03-20 14:52:56 +01:00
sql = (
"UPDATE event_json SET internal_metadata = ?"
" WHERE event_id = ?"
)
txn.execute(
sql,
2015-04-16 12:17:52 +02:00
(metadata_json, event.event_id,)
2015-03-20 14:52:56 +01:00
)
sql = (
"UPDATE events SET outlier = ?"
2015-03-20 14:52:56 +01:00
" WHERE event_id = ?"
)
txn.execute(
sql,
(False, event.event_id,)
2015-03-20 14:52:56 +01:00
)
self._update_extremeties(txn, [event])
2015-06-25 18:18:19 +02:00
events_and_contexts = filter(
lambda ec: ec[0] not in to_remove,
events_and_contexts
2015-03-24 17:20:26 +01:00
)
2015-06-25 18:18:19 +02:00
if not events_and_contexts:
return
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
self._store_mult_state_groups_txn(txn, [
(event, context)
for event, context in events_and_contexts
if not event.internal_metadata.is_outlier()
])
self._handle_mult_prev_events(
2015-03-20 14:52:56 +01:00
txn,
2015-06-25 18:18:19 +02:00
events=[event for event, _ in events_and_contexts],
2015-03-20 14:52:56 +01:00
)
2015-06-25 18:18:19 +02:00
for event, _ in events_and_contexts:
if event.type == EventTypes.Name:
self._store_room_name_txn(txn, event)
elif event.type == EventTypes.Topic:
self._store_room_topic_txn(txn, event)
elif event.type == EventTypes.Message:
self._store_room_message_txn(txn, event)
2015-06-25 18:18:19 +02:00
elif event.type == EventTypes.Redaction:
self._store_redaction(txn, event)
elif event.type == EventTypes.RoomHistoryVisibility:
self._store_history_visibility_txn(txn, event)
elif event.type == EventTypes.GuestAccess:
self._store_guest_access_txn(txn, event)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
self._store_room_members_txn(
txn,
[
event
for event, _ in events_and_contexts
if event.type == EventTypes.Member
2015-03-20 14:52:56 +01:00
]
2015-06-25 18:18:19 +02:00
)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
def event_dict(event):
return {
k: v
for k, v in event.get_dict().items()
if k not in [
"redacted",
"redacted_because",
]
}
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
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()
2015-06-25 18:18:19 +02:00
).decode("UTF-8"),
"json": encode_json(event_dict(event)).decode("UTF-8"),
2015-06-25 18:18:19 +02:00
}
for event, _ in events_and_contexts
],
2015-04-15 11:24:24 +02:00
)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
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(),
"content": encode_json(event.content).decode("UTF-8"),
"origin_server_ts": int(event.origin_server_ts),
2015-06-25 18:18:19 +02:00
}
for event, _ in events_and_contexts
],
2015-03-20 14:52:56 +01:00
)
if context.rejected:
self._store_rejections_txn(
txn, event.event_id, context.rejected
)
2015-03-20 14:52:56 +01:00
2015-05-05 16:13:25 +02:00
self._simple_insert_many_txn(
txn,
table="event_auth",
values=[
{
"event_id": event.event_id,
"room_id": event.room_id,
"auth_id": auth_id,
2015-05-05 16:13:25 +02:00
}
2015-06-25 18:18:19 +02:00
for event, _ in events_and_contexts
2015-05-05 16:13:25 +02:00
for auth_id, _ in event.auth_events
],
)
2015-06-25 18:18:19 +02:00
self._store_event_reference_hashes_txn(
txn, [event for event, _ in events_and_contexts]
)
state_events_and_contexts = filter(
lambda i: i[0].is_state(),
events_and_contexts,
)
2015-06-25 18:18:19 +02:00
state_values = []
for event, context in state_events_and_contexts:
2015-03-20 14:52:56 +01:00
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 18:18:19 +02:00
state_values.append(vals)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
self._simple_insert_many_txn(
txn,
table="state_events",
values=state_values,
)
2015-03-20 14:52:56 +01:00
2015-06-25 18:18:19 +02:00
self._simple_insert_many_txn(
txn,
table="event_edges",
values=[
{
"event_id": event.event_id,
"prev_event_id": prev_id,
"room_id": event.room_id,
"is_state": True,
}
for event, _ in state_events_and_contexts
for prev_id, _ in event.prev_state
],
)
2015-06-25 18:18:19 +02:00
if is_new_state:
for event, _ in state_events_and_contexts:
if not context.rejected:
txn.call_after(
2015-06-25 18:18:19 +02:00
self.get_current_state_for_key.invalidate,
(event.room_id, event.type, event.state_key,)
)
2015-06-25 18:18:19 +02:00
if event.type in [EventTypes.Name, EventTypes.Aliases]:
txn.call_after(
self.get_room_name_and_aliases.invalidate,
(event.room_id,)
2015-06-25 18:18:19 +02:00
)
self._simple_upsert_txn(
txn,
"current_state_events",
keyvalues={
"room_id": event.room_id,
"type": event.type,
"state_key": event.state_key,
},
values={
"event_id": event.event_id,
}
)
return
2015-03-20 14:52:56 +01:00
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 14:52:56 +01:00
txn.execute(
"INSERT INTO redactions (event_id, redacts) VALUES (?,?)",
(event.event_id, event.redacts)
)
def have_events(self, event_ids):
"""Given a list of event ids, check if we have already processed them.
Returns:
dict: Has an entry for each event id we already have seen. Maps to
the rejected reason string if we rejected the event, else maps to
None.
"""
if not event_ids:
return defer.succeed({})
def f(txn):
sql = (
"SELECT e.event_id, reason FROM events as e "
"LEFT JOIN rejections as r ON e.event_id = r.event_id "
"WHERE e.event_id = ?"
)
res = {}
for event_id in event_ids:
txn.execute(sql, (event_id,))
row = txn.fetchone()
if row:
_, rejected = row
res[event_id] = rejected
return res
return self.runInteraction(
"have_events", f,
)
2015-05-14 15:45:22 +02:00
@defer.inlineCallbacks
def _get_events(self, event_ids, check_redacted=True,
get_prev_content=False, allow_rejected=False):
2015-05-14 15:45:22 +02:00
if not event_ids:
defer.returnValue([])
event_map = self._get_events_from_cache(
event_ids,
check_redacted=check_redacted,
get_prev_content=get_prev_content,
allow_rejected=allow_rejected,
)
missing_events_ids = [e for e in event_ids if e not in event_map]
if not missing_events_ids:
defer.returnValue([
event_map[e_id] for e_id in event_ids
if e_id in event_map and event_map[e_id]
])
missing_events = yield self._enqueue_events(
missing_events_ids,
check_redacted=check_redacted,
get_prev_content=get_prev_content,
allow_rejected=allow_rejected,
)
2015-05-14 15:45:22 +02:00
event_map.update(missing_events)
defer.returnValue([
event_map[e_id] for e_id in event_ids
if e_id in event_map and event_map[e_id]
])
def _get_events_txn(self, txn, event_ids, check_redacted=True,
2015-05-14 15:45:22 +02:00
get_prev_content=False, allow_rejected=False):
if not event_ids:
return []
event_map = self._get_events_from_cache(
2015-05-14 15:45:22 +02:00
event_ids,
check_redacted=check_redacted,
get_prev_content=get_prev_content,
allow_rejected=allow_rejected,
)
missing_events_ids = [e for e in event_ids if e not in event_map]
if not missing_events_ids:
return [
event_map[e_id] for e_id in event_ids
if e_id in event_map and event_map[e_id]
]
missing_events = self._fetch_events_txn(
txn,
missing_events_ids,
check_redacted=check_redacted,
get_prev_content=get_prev_content,
allow_rejected=allow_rejected,
)
event_map.update(missing_events)
return [
event_map[e_id] for e_id in event_ids
if e_id in event_map and event_map[e_id]
]
def _invalidate_get_event_cache(self, event_id):
for check_redacted in (False, True):
for get_prev_content in (False, True):
self._get_event_cache.invalidate(
(event_id, check_redacted, get_prev_content)
)
def _get_event_txn(self, txn, event_id, check_redacted=True,
get_prev_content=False, allow_rejected=False):
2015-05-14 15:45:22 +02:00
events = self._get_events_txn(
txn, [event_id],
check_redacted=check_redacted,
get_prev_content=get_prev_content,
allow_rejected=allow_rejected,
)
return events[0] if events else None
def _get_events_from_cache(self, events, check_redacted, get_prev_content,
allow_rejected):
event_map = {}
for event_id in events:
try:
ret = self._get_event_cache.get(
(event_id, check_redacted, get_prev_content,)
2015-05-14 15:45:22 +02:00
)
2015-05-14 15:45:22 +02:00
if allow_rejected or not ret.rejected_reason:
event_map[event_id] = ret
else:
event_map[event_id] = None
except KeyError:
pass
return event_map
2015-05-15 11:54:04 +02:00
def _do_fetch(self, conn):
2015-05-18 16:18:41 +02:00
"""Takes a database connection and waits for requests for events from
the _event_fetch_list queue.
"""
2015-05-15 11:54:04 +02:00
event_list = []
2015-05-15 12:35:04 +02:00
i = 0
2015-05-15 12:00:50 +02:00
while True:
try:
2015-05-15 11:54:04 +02:00
with self._event_fetch_lock:
event_list = self._event_fetch_list
self._event_fetch_list = []
2015-05-15 17:34:17 +02:00
2015-05-15 11:54:04 +02:00
if not event_list:
2015-05-21 12:18:22 +02:00
single_threaded = self.database_engine.single_threaded
if single_threaded or i > EVENT_QUEUE_ITERATIONS:
2015-05-15 12:35:04 +02:00
self._event_fetch_ongoing -= 1
return
else:
2015-05-21 12:18:22 +02:00
self._event_fetch_lock.wait(EVENT_QUEUE_TIMEOUT_S)
2015-05-15 12:35:04 +02:00
i += 1
continue
i = 0
2015-05-15 11:54:04 +02:00
event_id_lists = zip(*event_list)[0]
event_ids = [
item for sublist in event_id_lists for item in sublist
]
rows = self._new_transaction(
conn, "do_fetch", [], None, self._fetch_event_rows, event_ids
2015-05-15 11:54:04 +02:00
)
2015-05-15 11:54:04 +02:00
row_dict = {
r["event_id"]: r
for r in rows
}
2015-05-18 16:18:41 +02:00
# We only want to resolve deferreds from the main thread
2015-05-15 12:26:00 +02:00
def fire(lst, res):
for ids, d in lst:
2015-05-15 11:54:04 +02:00
if not d.called:
try:
2016-02-04 11:22:44 +01:00
with PreserveLoggingContext():
d.callback([
res[i]
for i in ids
if i in res
])
2015-05-15 11:54:04 +02:00
except:
logger.exception("Failed to callback")
2016-02-04 11:22:44 +01:00
with PreserveLoggingContext():
reactor.callFromThread(fire, event_list, row_dict)
2015-05-15 12:00:50 +02:00
except Exception as e:
logger.exception("do_fetch")
2015-05-18 16:18:41 +02:00
# We only want to resolve deferreds from the main thread
2015-05-15 12:00:50 +02:00
def fire(evs):
for _, d in evs:
if not d.called:
2016-02-04 11:22:44 +01:00
with PreserveLoggingContext():
d.errback(e)
2015-05-15 11:54:04 +02:00
2015-05-15 12:00:50 +02:00
if event_list:
2016-02-04 11:22:44 +01:00
with PreserveLoggingContext():
reactor.callFromThread(fire, event_list)
2015-05-15 11:54:04 +02:00
@defer.inlineCallbacks
def _enqueue_events(self, events, check_redacted=True,
get_prev_content=False, allow_rejected=False):
2015-05-18 16:18:41 +02:00
"""Fetches events from the database using the _event_fetch_list. This
allows batch and bulk fetching of events - it allows us to fetch events
without having to create a new transaction for each request for events.
"""
2015-05-15 11:54:04 +02:00
if not events:
defer.returnValue({})
2015-05-14 17:54:35 +02:00
events_d = defer.Deferred()
2015-05-15 12:00:50 +02:00
with self._event_fetch_lock:
self._event_fetch_list.append(
(events, events_d)
)
2015-05-14 17:54:35 +02:00
2015-05-15 15:06:42 +02:00
self._event_fetch_lock.notify()
2015-05-15 12:35:04 +02:00
2015-05-21 12:18:22 +02:00
if self._event_fetch_ongoing < EVENT_QUEUE_THREADS:
2015-05-15 11:54:04 +02:00
self._event_fetch_ongoing += 1
2015-05-15 12:00:50 +02:00
should_start = True
else:
2015-05-15 12:00:50 +02:00
should_start = False
2015-05-15 12:00:50 +02:00
if should_start:
2016-02-04 11:22:44 +01:00
with PreserveLoggingContext():
self.runWithConnection(
self._do_fetch
)
2015-05-14 17:54:35 +02:00
2016-02-04 11:22:44 +01:00
with PreserveLoggingContext():
rows = yield events_d
2015-05-14 17:54:35 +02:00
2015-05-18 16:18:41 +02:00
if not allow_rejected:
rows[:] = [r for r in rows if not r["rejects"]]
2015-05-14 17:54:35 +02:00
res = yield defer.gatherResults(
[
2016-02-04 11:22:44 +01:00
preserve_fn(self._get_event_from_row)(
row["internal_metadata"], row["json"], row["redacts"],
check_redacted=check_redacted,
get_prev_content=get_prev_content,
rejected_reason=row["rejects"],
)
for row in rows
2015-05-14 17:54:35 +02:00
],
consumeErrors=True
)
defer.returnValue({
e.event_id: e
for e in res if e
})
def _fetch_event_rows(self, txn, events):
2015-05-14 15:45:22 +02:00
rows = []
N = 200
for i in range(1 + len(events) / N):
2016-02-02 18:18:50 +01:00
evs = events[i * N:(i + 1) * N]
2015-05-14 15:45:22 +02:00
if not evs:
break
2015-05-14 15:45:22 +02:00
sql = (
"SELECT "
" e.event_id as event_id, "
" e.internal_metadata,"
" e.json,"
" r.redacts as redacts,"
" rej.event_id as rejects "
2015-05-14 15:45:22 +02:00
" FROM event_json 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)"
2016-02-02 18:18:50 +01:00
) % (",".join(["?"] * len(evs)),)
2015-05-14 15:45:22 +02:00
txn.execute(sql, evs)
rows.extend(self.cursor_to_dict(txn))
return rows
2015-05-18 16:18:41 +02:00
def _fetch_events_txn(self, txn, events, check_redacted=True,
get_prev_content=False, allow_rejected=False):
if not events:
2015-05-18 16:18:41 +02:00
return {}
2015-05-18 16:18:41 +02:00
rows = self._fetch_event_rows(
txn, events,
)
if not allow_rejected:
rows[:] = [r for r in rows if not r["rejects"]]
2015-05-18 16:18:41 +02:00
res = [
self._get_event_from_row_txn(
2015-05-18 16:18:41 +02:00
txn,
row["internal_metadata"], row["json"], row["redacts"],
check_redacted=check_redacted,
get_prev_content=get_prev_content,
rejected_reason=row["rejects"],
)
2015-05-18 16:18:41 +02:00
for row in rows
]
2015-05-14 15:45:22 +02:00
2015-05-18 16:18:41 +02:00
return {
r.event_id: r
for r in res
2015-05-18 16:18:41 +02:00
}
2015-05-14 15:45:22 +02:00
@defer.inlineCallbacks
def _get_event_from_row(self, internal_metadata, js, redacted,
2015-05-14 15:45:22 +02:00
check_redacted=True, get_prev_content=False,
rejected_reason=None):
d = json.loads(js)
internal_metadata = json.loads(internal_metadata)
2015-05-18 16:18:41 +02:00
if rejected_reason:
rejected_reason = yield self._simple_select_one_onecol(
table="rejections",
keyvalues={"event_id": rejected_reason},
retcol="reason",
desc="_get_event_from_row",
)
ev = FrozenEvent(
d,
internal_metadata_dict=internal_metadata,
rejected_reason=rejected_reason,
)
if check_redacted and redacted:
ev = prune_event(ev)
redaction_id = yield self._simple_select_one_onecol(
table="redactions",
keyvalues={"redacts": ev.event_id},
retcol="event_id",
desc="_get_event_from_row",
)
ev.unsigned["redacted_by"] = redaction_id
# Get the redaction event.
because = yield self.get_event(
redaction_id,
check_redacted=False,
allow_none=True,
)
if because:
# It's fine to do add the event directly, since get_pdu_json
# will serialise this field correctly
ev.unsigned["redacted_because"] = because
if get_prev_content and "replaces_state" in ev.unsigned:
prev = yield self.get_event(
ev.unsigned["replaces_state"],
get_prev_content=False,
allow_none=True,
)
if prev:
ev.unsigned["prev_content"] = prev.content
ev.unsigned["prev_sender"] = prev.sender
self._get_event_cache.prefill(
(ev.event_id, check_redacted, get_prev_content), ev
)
defer.returnValue(ev)
def _get_event_from_row_txn(self, txn, internal_metadata, js, redacted,
check_redacted=True, get_prev_content=False,
rejected_reason=None):
d = json.loads(js)
internal_metadata = json.loads(internal_metadata)
2015-05-14 15:45:22 +02:00
if rejected_reason:
rejected_reason = self._simple_select_one_onecol_txn(
2015-05-14 15:45:22 +02:00
txn,
table="rejections",
keyvalues={"event_id": rejected_reason},
retcol="reason",
)
ev = FrozenEvent(
d,
internal_metadata_dict=internal_metadata,
rejected_reason=rejected_reason,
)
if check_redacted and redacted:
ev = prune_event(ev)
redaction_id = self._simple_select_one_onecol_txn(
2015-05-14 15:45:22 +02:00
txn,
table="redactions",
keyvalues={"redacts": ev.event_id},
retcol="event_id",
)
ev.unsigned["redacted_by"] = redaction_id
# Get the redaction event.
because = self._get_event_txn(
txn,
2015-05-14 15:45:22 +02:00
redaction_id,
check_redacted=False
)
if because:
ev.unsigned["redacted_because"] = because
if get_prev_content and "replaces_state" in ev.unsigned:
prev = self._get_event_txn(
txn,
ev.unsigned["replaces_state"],
get_prev_content=False,
)
if prev:
ev.unsigned["prev_content"] = prev.content
ev.unsigned["prev_sender"] = prev.sender
self._get_event_cache.prefill(
(ev.event_id, check_redacted, get_prev_content), ev
)
return ev
def _parse_events_txn(self, txn, rows):
event_ids = [r["event_id"] for r in rows]
return self._get_events_txn(txn, event_ids)
@defer.inlineCallbacks
def count_daily_messages(self):
2015-09-22 14:29:36 +02:00
"""
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):
now = self.hs.get_clock().time()
txn.execute(
"SELECT reported_stream_token, reported_time FROM stats_reporting"
)
last_reported = self.cursor_to_dict(txn)
txn.execute(
"SELECT stream_ordering"
" FROM events"
" ORDER BY stream_ordering DESC"
" LIMIT 1"
)
now_reporting = self.cursor_to_dict(txn)
if not now_reporting:
2016-01-06 14:52:16 +01:00
logger.info("Calculating daily messages skipped; no now_reporting")
return None
now_reporting = now_reporting[0]["stream_ordering"]
txn.execute("DELETE FROM stats_reporting")
txn.execute(
"INSERT INTO stats_reporting"
" (reported_stream_token, reported_time)"
" VALUES (?, ?)",
(now_reporting, now,)
)
if not last_reported:
2016-01-06 14:52:16 +01:00
logger.info("Calculating daily messages skipped; no last_reported")
return None
# Close enough to correct for our purposes.
yesterday = (now - 24 * 60 * 60)
2016-01-06 14:52:16 +01:00
since_yesterday_seconds = yesterday - last_reported[0]["reported_time"]
any_since_yesterday = math.fabs(since_yesterday_seconds) > 60 * 60
if any_since_yesterday:
logger.info(
"Calculating daily messages skipped; since_yesterday_seconds: %d" %
(since_yesterday_seconds,)
)
return None
txn.execute(
"SELECT COUNT(*) as messages"
" FROM events NATURAL JOIN event_json"
" WHERE json like '%m.room.message%'"
" AND stream_ordering > ?"
" AND stream_ordering <= ?",
(
last_reported[0]["reported_stream_token"],
now_reporting,
)
)
rows = self.cursor_to_dict(txn)
if not rows:
2016-01-06 14:52:16 +01:00
logger.info("Calculating daily messages skipped; messages count missing")
return None
return rows[0]["messages"]
ret = yield self.runInteraction("count_messages", _count_messages)
defer.returnValue(ret)
@defer.inlineCallbacks
def _background_reindex_origin_server_ts(self, progress, batch_size):
target_min_stream_id = progress["target_min_stream_id_inclusive"]
max_stream_id = progress["max_stream_id_exclusive"]
rows_inserted = progress.get("rows_inserted", 0)
INSERT_CLUMP_SIZE = 1000
def reindex_search_txn(txn):
sql = (
"SELECT stream_ordering, event_id FROM events"
" WHERE ? <= stream_ordering AND stream_ordering < ?"
" ORDER BY stream_ordering DESC"
" LIMIT ?"
)
txn.execute(sql, (target_min_stream_id, max_stream_id, batch_size))
rows = txn.fetchall()
if not rows:
return 0
min_stream_id = rows[-1][0]
event_ids = [row[1] for row in rows]
events = self._get_events_txn(txn, event_ids)
rows = []
for event in events:
try:
event_id = event.event_id
origin_server_ts = event.origin_server_ts
except (KeyError, AttributeError):
# If the event is missing a necessary field then
# skip over it.
continue
rows.append((origin_server_ts, event_id))
sql = (
"UPDATE events SET origin_server_ts = ? WHERE event_id = ?"
)
for index in range(0, len(rows), INSERT_CLUMP_SIZE):
clump = rows[index:index + INSERT_CLUMP_SIZE]
txn.executemany(sql, clump)
progress = {
"target_min_stream_id_inclusive": target_min_stream_id,
"max_stream_id_exclusive": min_stream_id,
"rows_inserted": rows_inserted + len(rows)
}
self._background_update_progress_txn(
txn, self.EVENT_ORIGIN_SERVER_TS_NAME, progress
)
return len(rows)
result = yield self.runInteraction(
self.EVENT_ORIGIN_SERVER_TS_NAME, reindex_search_txn
)
if not result:
yield self._end_background_update(self.EVENT_ORIGIN_SERVER_TS_NAME)
defer.returnValue(result)