class SlavedClientIpStore(BaseSlavedStore): def __init__(self, db_conn, hs): super(SlavedClientIpStore, self).__init__(db_conn, hs) self.client_ip_last_seen = Cache( name="client_ip_last_seen", keylen=4, max_entries=50000 * CACHE_SIZE_FACTOR, ) def insert_client_ip(self, user_id, access_token, ip, user_agent, device_id): now = int(self._clock.time_msec()) key = (user_id, access_token, ip) try: last_seen = self.client_ip_last_seen.get(key) except KeyError: last_seen = None # Rate-limited inserts if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY: return self.client_ip_last_seen.prefill(key, now) self.hs.get_tcp_replication().send_user_ip( user_id, access_token, ip, user_agent, device_id, now )
class SlavedClientIpStore(BaseSlavedStore): def __init__(self, database: DatabasePool, db_conn, hs): super().__init__(database, db_conn, hs) self.client_ip_last_seen = Cache( name="client_ip_last_seen", keylen=4, max_entries=50000 ) async def insert_client_ip(self, user_id, access_token, ip, user_agent, device_id): now = int(self._clock.time_msec()) key = (user_id, access_token, ip) try: last_seen = self.client_ip_last_seen.get(key) except KeyError: last_seen = None # Rate-limited inserts if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY: return self.client_ip_last_seen.prefill(key, now) self.hs.get_tcp_replication().send_user_ip( user_id, access_token, ip, user_agent, device_id, now )
def test_cache_metric(self): """ Caches produce metrics reflecting their state when scraped. """ CACHE_NAME = "cache_metrics_test_fgjkbdfg" cache = Cache(CACHE_NAME, max_entries=777) items = { x.split(b"{")[0].decode("ascii"): x.split(b" ")[1].decode("ascii") for x in filter( lambda x: b"cache_metrics_test_fgjkbdfg" in x, generate_latest(REGISTRY).split(b"\n"), ) } self.assertEqual(items["synapse_util_caches_cache_size"], "0.0") self.assertEqual(items["synapse_util_caches_cache_max_size"], "777.0") cache.prefill("1", "hi") items = { x.split(b"{")[0].decode("ascii"): x.split(b" ")[1].decode("ascii") for x in filter( lambda x: b"cache_metrics_test_fgjkbdfg" in x, generate_latest(REGISTRY).split(b"\n"), ) } self.assertEqual(items["synapse_util_caches_cache_size"], "1.0") self.assertEqual(items["synapse_util_caches_cache_max_size"], "777.0")
def test_eviction(self): cache = Cache("test", max_entries=2) cache.prefill(1, "one") cache.prefill(2, "two") cache.prefill(3, "three") # 1 will be evicted failed = False try: cache.get(1) except KeyError: failed = True self.assertTrue(failed) cache.get(2) cache.get(3)
def test_eviction_lru(self): cache = Cache("test", max_entries=2) cache.prefill(1, "one") cache.prefill(2, "two") # Now access 1 again, thus causing 2 to be least-recently used cache.get(1) cache.prefill(3, "three") failed = False try: cache.get(2) except KeyError: failed = True self.assertTrue(failed) cache.get(1) cache.get(3)
def test_eviction_lru(self): cache = Cache("test", max_entries=2, lru=True) cache.prefill(1, "one") cache.prefill(2, "two") # Now access 1 again, thus causing 2 to be least-recently used cache.get(1) cache.prefill(3, "three") failed = False try: cache.get(2) except KeyError: failed = True self.assertTrue(failed) cache.get(1) cache.get(3)
class CacheTestCase(unittest.TestCase): def setUp(self): self.cache = Cache("test") def test_empty(self): failed = False try: self.cache.get("foo") except KeyError: failed = True self.assertTrue(failed) def test_hit(self): self.cache.prefill("foo", 123) self.assertEquals(self.cache.get("foo"), 123) def test_invalidate(self): self.cache.prefill(("foo", ), 123) self.cache.invalidate(("foo", )) failed = False try: self.cache.get(("foo", )) except KeyError: failed = True self.assertTrue(failed) def test_eviction(self): cache = Cache("test", max_entries=2) cache.prefill(1, "one") cache.prefill(2, "two") cache.prefill(3, "three") # 1 will be evicted failed = False try: cache.get(1) except KeyError: failed = True self.assertTrue(failed) cache.get(2) cache.get(3) def test_eviction_lru(self): cache = Cache("test", max_entries=2) cache.prefill(1, "one") cache.prefill(2, "two") # Now access 1 again, thus causing 2 to be least-recently used cache.get(1) cache.prefill(3, "three") failed = False try: cache.get(2) except KeyError: failed = True self.assertTrue(failed) cache.get(1) cache.get(3)
class DeviceStore(DeviceWorkerStore, DeviceBackgroundUpdateStore): def __init__(self, database: Database, db_conn, hs): super(DeviceStore, self).__init__(database, db_conn, hs) # Map of (user_id, device_id) -> bool. If there is an entry that implies # the device exists. self.device_id_exists_cache = Cache( name="device_id_exists", keylen=2, max_entries=10000 ) self._clock.looping_call(self._prune_old_outbound_device_pokes, 60 * 60 * 1000) @defer.inlineCallbacks def store_device(self, user_id, device_id, initial_device_display_name): """Ensure the given device is known; add it to the store if not Args: user_id (str): id of user associated with the device device_id (str): id of device initial_device_display_name (str): initial displayname of the device. Ignored if device exists. Returns: defer.Deferred: boolean whether the device was inserted or an existing device existed with that ID. Raises: StoreError: if the device is already in use """ key = (user_id, device_id) if self.device_id_exists_cache.get(key, None): return False try: inserted = yield self.db.simple_insert( "devices", values={ "user_id": user_id, "device_id": device_id, "display_name": initial_device_display_name, "hidden": False, }, desc="store_device", or_ignore=True, ) if not inserted: # if the device already exists, check if it's a real device, or # if the device ID is reserved by something else hidden = yield self.db.simple_select_one_onecol( "devices", keyvalues={"user_id": user_id, "device_id": device_id}, retcol="hidden", ) if hidden: raise StoreError(400, "The device ID is in use", Codes.FORBIDDEN) self.device_id_exists_cache.prefill(key, True) return inserted except StoreError: raise except Exception as e: logger.error( "store_device with device_id=%s(%r) user_id=%s(%r)" " display_name=%s(%r) failed: %s", type(device_id).__name__, device_id, type(user_id).__name__, user_id, type(initial_device_display_name).__name__, initial_device_display_name, e, ) raise StoreError(500, "Problem storing device.") @defer.inlineCallbacks def delete_device(self, user_id, device_id): """Delete a device. Args: user_id (str): The ID of the user which owns the device device_id (str): The ID of the device to delete Returns: defer.Deferred """ yield self.db.simple_delete_one( table="devices", keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False}, desc="delete_device", ) self.device_id_exists_cache.invalidate((user_id, device_id)) @defer.inlineCallbacks def delete_devices(self, user_id, device_ids): """Deletes several devices. Args: user_id (str): The ID of the user which owns the devices device_ids (list): The IDs of the devices to delete Returns: defer.Deferred """ yield self.db.simple_delete_many( table="devices", column="device_id", iterable=device_ids, keyvalues={"user_id": user_id, "hidden": False}, desc="delete_devices", ) for device_id in device_ids: self.device_id_exists_cache.invalidate((user_id, device_id)) def update_device(self, user_id, device_id, new_display_name=None): """Update a device. Only updates the device if it is not marked as hidden. Args: user_id (str): The ID of the user which owns the device device_id (str): The ID of the device to update new_display_name (str|None): new displayname for device; None to leave unchanged Raises: StoreError: if the device is not found Returns: defer.Deferred """ updates = {} if new_display_name is not None: updates["display_name"] = new_display_name if not updates: return defer.succeed(None) return self.db.simple_update_one( table="devices", keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False}, updatevalues=updates, desc="update_device", ) def update_remote_device_list_cache_entry( self, user_id, device_id, content, stream_id ): """Updates a single device in the cache of a remote user's devicelist. Note: assumes that we are the only thread that can be updating this user's device list. Args: user_id (str): User to update device list for device_id (str): ID of decivice being updated content (dict): new data on this device stream_id (int): the version of the device list Returns: Deferred[None] """ return self.db.runInteraction( "update_remote_device_list_cache_entry", self._update_remote_device_list_cache_entry_txn, user_id, device_id, content, stream_id, ) def _update_remote_device_list_cache_entry_txn( self, txn, user_id, device_id, content, stream_id ): if content.get("deleted"): self.db.simple_delete_txn( txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id, "device_id": device_id}, ) txn.call_after(self.device_id_exists_cache.invalidate, (user_id, device_id)) else: self.db.simple_upsert_txn( txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id, "device_id": device_id}, values={"content": json.dumps(content)}, # we don't need to lock, because we assume we are the only thread # updating this user's devices. lock=False, ) txn.call_after(self._get_cached_user_device.invalidate, (user_id, device_id)) txn.call_after(self.get_cached_devices_for_user.invalidate, (user_id,)) txn.call_after( self.get_device_list_last_stream_id_for_remote.invalidate, (user_id,) ) self.db.simple_upsert_txn( txn, table="device_lists_remote_extremeties", keyvalues={"user_id": user_id}, values={"stream_id": stream_id}, # again, we can assume we are the only thread updating this user's # extremity. lock=False, ) def update_remote_device_list_cache(self, user_id, devices, stream_id): """Replace the entire cache of the remote user's devices. Note: assumes that we are the only thread that can be updating this user's device list. Args: user_id (str): User to update device list for devices (list[dict]): list of device objects supplied over federation stream_id (int): the version of the device list Returns: Deferred[None] """ return self.db.runInteraction( "update_remote_device_list_cache", self._update_remote_device_list_cache_txn, user_id, devices, stream_id, ) def _update_remote_device_list_cache_txn(self, txn, user_id, devices, stream_id): self.db.simple_delete_txn( txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id} ) self.db.simple_insert_many_txn( txn, table="device_lists_remote_cache", values=[ { "user_id": user_id, "device_id": content["device_id"], "content": json.dumps(content), } for content in devices ], ) txn.call_after(self.get_cached_devices_for_user.invalidate, (user_id,)) txn.call_after(self._get_cached_user_device.invalidate_many, (user_id,)) txn.call_after( self.get_device_list_last_stream_id_for_remote.invalidate, (user_id,) ) self.db.simple_upsert_txn( txn, table="device_lists_remote_extremeties", keyvalues={"user_id": user_id}, values={"stream_id": stream_id}, # we don't need to lock, because we can assume we are the only thread # updating this user's extremity. lock=False, ) # If we're replacing the remote user's device list cache presumably # we've done a full resync, so we remove the entry that says we need # to resync self.db.simple_delete_txn( txn, table="device_lists_remote_resync", keyvalues={"user_id": user_id}, ) @defer.inlineCallbacks def add_device_change_to_streams(self, user_id, device_ids, hosts): """Persist that a user's devices have been updated, and which hosts (if any) should be poked. """ if not device_ids: return with self._device_list_id_gen.get_next_mult(len(device_ids)) as stream_ids: yield self.db.runInteraction( "add_device_change_to_stream", self._add_device_change_to_stream_txn, user_id, device_ids, stream_ids, ) if not hosts: return stream_ids[-1] context = get_active_span_text_map() with self._device_list_id_gen.get_next_mult( len(hosts) * len(device_ids) ) as stream_ids: yield self.db.runInteraction( "add_device_outbound_poke_to_stream", self._add_device_outbound_poke_to_stream_txn, user_id, device_ids, hosts, stream_ids, context, ) return stream_ids[-1] def _add_device_change_to_stream_txn( self, txn: LoggingTransaction, user_id: str, device_ids: Collection[str], stream_ids: List[str], ): txn.call_after( self._device_list_stream_cache.entity_has_changed, user_id, stream_ids[-1], ) min_stream_id = stream_ids[0] # Delete older entries in the table, as we really only care about # when the latest change happened. txn.executemany( """ DELETE FROM device_lists_stream WHERE user_id = ? AND device_id = ? AND stream_id < ? """, [(user_id, device_id, min_stream_id) for device_id in device_ids], ) self.db.simple_insert_many_txn( txn, table="device_lists_stream", values=[ {"stream_id": stream_id, "user_id": user_id, "device_id": device_id} for stream_id, device_id in zip(stream_ids, device_ids) ], ) def _add_device_outbound_poke_to_stream_txn( self, txn, user_id, device_ids, hosts, stream_ids, context, ): for host in hosts: txn.call_after( self._device_list_federation_stream_cache.entity_has_changed, host, stream_ids[-1], ) now = self._clock.time_msec() next_stream_id = iter(stream_ids) self.db.simple_insert_many_txn( txn, table="device_lists_outbound_pokes", values=[ { "destination": destination, "stream_id": next(next_stream_id), "user_id": user_id, "device_id": device_id, "sent": False, "ts": now, "opentracing_context": json.dumps(context) if whitelisted_homeserver(destination) else "{}", } for destination in hosts for device_id in device_ids ], ) def _prune_old_outbound_device_pokes(self, prune_age=24 * 60 * 60 * 1000): """Delete old entries out of the device_lists_outbound_pokes to ensure that we don't fill up due to dead servers. Normally, we try to send device updates as a delta since a previous known point: this is done by setting the prev_id in the m.device_list_update EDU. However, for that to work, we have to have a complete record of each change to each device, which can add up to quite a lot of data. An alternative mechanism is that, if the remote server sees that it has missed an entry in the stream_id sequence for a given user, it will request a full list of that user's devices. Hence, we can reduce the amount of data we have to store (and transmit in some future transaction), by clearing almost everything for a given destination out of the database, and having the remote server resync. All we need to do is make sure we keep at least one row for each (user, destination) pair, to remind us to send a m.device_list_update EDU for that user when the destination comes back. It doesn't matter which device we keep. """ yesterday = self._clock.time_msec() - prune_age def _prune_txn(txn): # look for (user, destination) pairs which have an update older than # the cutoff. # # For each pair, we also need to know the most recent stream_id, and # an arbitrary device_id at that stream_id. select_sql = """ SELECT dlop1.destination, dlop1.user_id, MAX(dlop1.stream_id) AS stream_id, (SELECT MIN(dlop2.device_id) AS device_id FROM device_lists_outbound_pokes dlop2 WHERE dlop2.destination = dlop1.destination AND dlop2.user_id=dlop1.user_id AND dlop2.stream_id=MAX(dlop1.stream_id) ) FROM device_lists_outbound_pokes dlop1 GROUP BY destination, user_id HAVING min(ts) < ? AND count(*) > 1 """ txn.execute(select_sql, (yesterday,)) rows = txn.fetchall() if not rows: return logger.info( "Pruning old outbound device list updates for %i users/destinations: %s", len(rows), shortstr((row[0], row[1]) for row in rows), ) # we want to keep the update with the highest stream_id for each user. # # there might be more than one update (with different device_ids) with the # same stream_id, so we also delete all but one rows with the max stream id. delete_sql = """ DELETE FROM device_lists_outbound_pokes WHERE destination = ? AND user_id = ? AND ( stream_id < ? OR (stream_id = ? AND device_id != ?) ) """ count = 0 for (destination, user_id, stream_id, device_id) in rows: txn.execute( delete_sql, (destination, user_id, stream_id, stream_id, device_id) ) count += txn.rowcount # Since we've deleted unsent deltas, we need to remove the entry # of last successful sent so that the prev_ids are correctly set. sql = """ DELETE FROM device_lists_outbound_last_success WHERE destination = ? AND user_id = ? """ txn.executemany(sql, ((row[0], row[1]) for row in rows)) logger.info("Pruned %d device list outbound pokes", count) return run_as_background_process( "prune_old_outbound_device_pokes", self.db.runInteraction, "_prune_old_outbound_device_pokes", _prune_txn, )
class DeviceStore(DeviceWorkerStore, DeviceBackgroundUpdateStore): def __init__(self, database: Database, db_conn, hs): super(DeviceStore, self).__init__(database, db_conn, hs) # Map of (user_id, device_id) -> bool. If there is an entry that implies # the device exists. self.device_id_exists_cache = Cache( name="device_id_exists", keylen=2, max_entries=10000 ) self._clock.looping_call(self._prune_old_outbound_device_pokes, 60 * 60 * 1000) @defer.inlineCallbacks def store_device(self, user_id, device_id, initial_device_display_name): """Ensure the given device is known; add it to the store if not Args: user_id (str): id of user associated with the device device_id (str): id of device initial_device_display_name (str): initial displayname of the device. Ignored if device exists. Returns: defer.Deferred: boolean whether the device was inserted or an existing device existed with that ID. Raises: StoreError: if the device is already in use """ key = (user_id, device_id) if self.device_id_exists_cache.get(key, None): return False try: inserted = yield self.db.simple_insert( "devices", values={ "user_id": user_id, "device_id": device_id, "display_name": initial_device_display_name, "hidden": False, }, desc="store_device", or_ignore=True, ) if not inserted: # if the device already exists, check if it's a real device, or # if the device ID is reserved by something else hidden = yield self.db.simple_select_one_onecol( "devices", keyvalues={"user_id": user_id, "device_id": device_id}, retcol="hidden", ) if hidden: raise StoreError(400, "The device ID is in use", Codes.FORBIDDEN) self.device_id_exists_cache.prefill(key, True) return inserted except StoreError: raise except Exception as e: logger.error( "store_device with device_id=%s(%r) user_id=%s(%r)" " display_name=%s(%r) failed: %s", type(device_id).__name__, device_id, type(user_id).__name__, user_id, type(initial_device_display_name).__name__, initial_device_display_name, e, ) raise StoreError(500, "Problem storing device.") @defer.inlineCallbacks def delete_device(self, user_id, device_id): """Delete a device. Args: user_id (str): The ID of the user which owns the device device_id (str): The ID of the device to delete Returns: defer.Deferred """ yield self.db.simple_delete_one( table="devices", keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False}, desc="delete_device", ) self.device_id_exists_cache.invalidate((user_id, device_id)) @defer.inlineCallbacks def delete_devices(self, user_id, device_ids): """Deletes several devices. Args: user_id (str): The ID of the user which owns the devices device_ids (list): The IDs of the devices to delete Returns: defer.Deferred """ yield self.db.simple_delete_many( table="devices", column="device_id", iterable=device_ids, keyvalues={"user_id": user_id, "hidden": False}, desc="delete_devices", ) for device_id in device_ids: self.device_id_exists_cache.invalidate((user_id, device_id)) def update_device(self, user_id, device_id, new_display_name=None): """Update a device. Only updates the device if it is not marked as hidden. Args: user_id (str): The ID of the user which owns the device device_id (str): The ID of the device to update new_display_name (str|None): new displayname for device; None to leave unchanged Raises: StoreError: if the device is not found Returns: defer.Deferred """ updates = {} if new_display_name is not None: updates["display_name"] = new_display_name if not updates: return defer.succeed(None) return self.db.simple_update_one( table="devices", keyvalues={"user_id": user_id, "device_id": device_id, "hidden": False}, updatevalues=updates, desc="update_device", ) @defer.inlineCallbacks def mark_remote_user_device_list_as_unsubscribed(self, user_id): """Mark that we no longer track device lists for remote user. """ yield self.db.simple_delete( table="device_lists_remote_extremeties", keyvalues={"user_id": user_id}, desc="mark_remote_user_device_list_as_unsubscribed", ) self.get_device_list_last_stream_id_for_remote.invalidate((user_id,)) def update_remote_device_list_cache_entry( self, user_id, device_id, content, stream_id ): """Updates a single device in the cache of a remote user's devicelist. Note: assumes that we are the only thread that can be updating this user's device list. Args: user_id (str): User to update device list for device_id (str): ID of decivice being updated content (dict): new data on this device stream_id (int): the version of the device list Returns: Deferred[None] """ return self.db.runInteraction( "update_remote_device_list_cache_entry", self._update_remote_device_list_cache_entry_txn, user_id, device_id, content, stream_id, ) def _update_remote_device_list_cache_entry_txn( self, txn, user_id, device_id, content, stream_id ): if content.get("deleted"): self.db.simple_delete_txn( txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id, "device_id": device_id}, ) txn.call_after(self.device_id_exists_cache.invalidate, (user_id, device_id)) else: self.db.simple_upsert_txn( txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id, "device_id": device_id}, values={"content": json.dumps(content)}, # we don't need to lock, because we assume we are the only thread # updating this user's devices. lock=False, ) txn.call_after(self._get_cached_user_device.invalidate, (user_id, device_id)) txn.call_after(self._get_cached_devices_for_user.invalidate, (user_id,)) txn.call_after( self.get_device_list_last_stream_id_for_remote.invalidate, (user_id,) ) self.db.simple_upsert_txn( txn, table="device_lists_remote_extremeties", keyvalues={"user_id": user_id}, values={"stream_id": stream_id}, # again, we can assume we are the only thread updating this user's # extremity. lock=False, ) def update_remote_device_list_cache(self, user_id, devices, stream_id): """Replace the entire cache of the remote user's devices. Note: assumes that we are the only thread that can be updating this user's device list. Args: user_id (str): User to update device list for devices (list[dict]): list of device objects supplied over federation stream_id (int): the version of the device list Returns: Deferred[None] """ return self.db.runInteraction( "update_remote_device_list_cache", self._update_remote_device_list_cache_txn, user_id, devices, stream_id, ) def _update_remote_device_list_cache_txn(self, txn, user_id, devices, stream_id): self.db.simple_delete_txn( txn, table="device_lists_remote_cache", keyvalues={"user_id": user_id} ) self.db.simple_insert_many_txn( txn, table="device_lists_remote_cache", values=[ { "user_id": user_id, "device_id": content["device_id"], "content": json.dumps(content), } for content in devices ], ) txn.call_after(self._get_cached_devices_for_user.invalidate, (user_id,)) txn.call_after(self._get_cached_user_device.invalidate_many, (user_id,)) txn.call_after( self.get_device_list_last_stream_id_for_remote.invalidate, (user_id,) ) self.db.simple_upsert_txn( txn, table="device_lists_remote_extremeties", keyvalues={"user_id": user_id}, values={"stream_id": stream_id}, # we don't need to lock, because we can assume we are the only thread # updating this user's extremity. lock=False, ) @defer.inlineCallbacks def add_device_change_to_streams(self, user_id, device_ids, hosts): """Persist that a user's devices have been updated, and which hosts (if any) should be poked. """ with self._device_list_id_gen.get_next() as stream_id: yield self.db.runInteraction( "add_device_change_to_streams", self._add_device_change_txn, user_id, device_ids, hosts, stream_id, ) return stream_id def _add_device_change_txn(self, txn, user_id, device_ids, hosts, stream_id): now = self._clock.time_msec() txn.call_after( self._device_list_stream_cache.entity_has_changed, user_id, stream_id ) for host in hosts: txn.call_after( self._device_list_federation_stream_cache.entity_has_changed, host, stream_id, ) # Delete older entries in the table, as we really only care about # when the latest change happened. txn.executemany( """ DELETE FROM device_lists_stream WHERE user_id = ? AND device_id = ? AND stream_id < ? """, [(user_id, device_id, stream_id) for device_id in device_ids], ) self.db.simple_insert_many_txn( txn, table="device_lists_stream", values=[ {"stream_id": stream_id, "user_id": user_id, "device_id": device_id} for device_id in device_ids ], ) context = get_active_span_text_map() self.db.simple_insert_many_txn( txn, table="device_lists_outbound_pokes", values=[ { "destination": destination, "stream_id": stream_id, "user_id": user_id, "device_id": device_id, "sent": False, "ts": now, "opentracing_context": json.dumps(context) if whitelisted_homeserver(destination) else "{}", } for destination in hosts for device_id in device_ids ], ) def _prune_old_outbound_device_pokes(self): """Delete old entries out of the device_lists_outbound_pokes to ensure that we don't fill up due to dead servers. We keep one entry per (destination, user_id) tuple to ensure that the prev_ids remain correct if the server does come back. """ yesterday = self._clock.time_msec() - 24 * 60 * 60 * 1000 def _prune_txn(txn): select_sql = """ SELECT destination, user_id, max(stream_id) as stream_id FROM device_lists_outbound_pokes GROUP BY destination, user_id HAVING min(ts) < ? AND count(*) > 1 """ txn.execute(select_sql, (yesterday,)) rows = txn.fetchall() if not rows: return delete_sql = """ DELETE FROM device_lists_outbound_pokes WHERE ts < ? AND destination = ? AND user_id = ? AND stream_id < ? """ txn.executemany( delete_sql, ((yesterday, row[0], row[1], row[2]) for row in rows) ) # Since we've deleted unsent deltas, we need to remove the entry # of last successful sent so that the prev_ids are correctly set. sql = """ DELETE FROM device_lists_outbound_last_success WHERE destination = ? AND user_id = ? """ txn.executemany(sql, ((row[0], row[1]) for row in rows)) logger.info("Pruned %d device list outbound pokes", txn.rowcount) return run_as_background_process( "prune_old_outbound_device_pokes", self.db.runInteraction, "_prune_old_outbound_device_pokes", _prune_txn, )
class ClientIpStore(ClientIpBackgroundUpdateStore): def __init__(self, database: Database, db_conn, hs): self.client_ip_last_seen = Cache(name="client_ip_last_seen", keylen=4, max_entries=50000 * CACHE_SIZE_FACTOR) super(ClientIpStore, self).__init__(database, db_conn, hs) self.user_ips_max_age = hs.config.user_ips_max_age # (user_id, access_token, ip,) -> (user_agent, device_id, last_seen) self._batch_row_update = {} self._client_ip_looper = self._clock.looping_call( self._update_client_ips_batch, 5 * 1000) self.hs.get_reactor().addSystemEventTrigger( "before", "shutdown", self._update_client_ips_batch) if self.user_ips_max_age: self._clock.looping_call(self._prune_old_user_ips, 5 * 1000) @defer.inlineCallbacks def insert_client_ip(self, user_id, access_token, ip, user_agent, device_id, now=None): if not now: now = int(self._clock.time_msec()) key = (user_id, access_token, ip) try: last_seen = self.client_ip_last_seen.get(key) except KeyError: last_seen = None yield self.populate_monthly_active_users(user_id) # Rate-limited inserts if last_seen is not None and (now - last_seen) < LAST_SEEN_GRANULARITY: return self.client_ip_last_seen.prefill(key, now) self._batch_row_update[key] = (user_agent, device_id, now) @wrap_as_background_process("update_client_ips") def _update_client_ips_batch(self): # If the DB pool has already terminated, don't try updating if not self.db.is_running(): return to_update = self._batch_row_update self._batch_row_update = {} return self.db.runInteraction("_update_client_ips_batch", self._update_client_ips_batch_txn, to_update) def _update_client_ips_batch_txn(self, txn, to_update): if "user_ips" in self.db._unsafe_to_upsert_tables or ( not self.database_engine.can_native_upsert): self.database_engine.lock_table(txn, "user_ips") for entry in iteritems(to_update): (user_id, access_token, ip), (user_agent, device_id, last_seen) = entry try: self.db.simple_upsert_txn( txn, table="user_ips", keyvalues={ "user_id": user_id, "access_token": access_token, "ip": ip, }, values={ "user_agent": user_agent, "device_id": device_id, "last_seen": last_seen, }, lock=False, ) # Technically an access token might not be associated with # a device so we need to check. if device_id: # this is always an update rather than an upsert: the row should # already exist, and if it doesn't, that may be because it has been # deleted, and we don't want to re-create it. self.db.simple_update_txn( txn, table="devices", keyvalues={ "user_id": user_id, "device_id": device_id }, updatevalues={ "user_agent": user_agent, "last_seen": last_seen, "ip": ip, }, ) except Exception as e: # Failed to upsert, log and continue logger.error("Failed to insert client IP %r: %r", entry, e) @defer.inlineCallbacks def get_last_client_ip_by_device(self, user_id, device_id): """For each device_id listed, give the user_ip it was last seen on Args: user_id (str) device_id (str): If None fetches all devices for the user Returns: defer.Deferred: resolves to a dict, where the keys are (user_id, device_id) tuples. The values are also dicts, with keys giving the column names """ keyvalues = {"user_id": user_id} if device_id is not None: keyvalues["device_id"] = device_id res = yield self.db.simple_select_list( table="devices", keyvalues=keyvalues, retcols=("user_id", "ip", "user_agent", "device_id", "last_seen"), ) ret = {(d["user_id"], d["device_id"]): d for d in res} for key in self._batch_row_update: uid, access_token, ip = key if uid == user_id: user_agent, did, last_seen = self._batch_row_update[key] if not device_id or did == device_id: ret[(user_id, device_id)] = { "user_id": user_id, "access_token": access_token, "ip": ip, "user_agent": user_agent, "device_id": did, "last_seen": last_seen, } return ret @defer.inlineCallbacks def get_user_ip_and_agents(self, user): user_id = user.to_string() results = {} for key in self._batch_row_update: uid, access_token, ip, = key if uid == user_id: user_agent, _, last_seen = self._batch_row_update[key] results[(access_token, ip)] = (user_agent, last_seen) rows = yield self.db.simple_select_list( table="user_ips", keyvalues={"user_id": user_id}, retcols=["access_token", "ip", "user_agent", "last_seen"], desc="get_user_ip_and_agents", ) results.update(((row["access_token"], row["ip"]), (row["user_agent"], row["last_seen"])) for row in rows) return [{ "access_token": access_token, "ip": ip, "user_agent": user_agent, "last_seen": last_seen, } for (access_token, ip), (user_agent, last_seen) in iteritems(results) ] @wrap_as_background_process("prune_old_user_ips") async def _prune_old_user_ips(self): """Removes entries in user IPs older than the configured period. """ if self.user_ips_max_age is None: # Nothing to do return if not await self.db.updates.has_completed_background_update( "devices_last_seen"): # Only start pruning if we have finished populating the devices # last seen info. return # We do a slightly funky SQL delete to ensure we don't try and delete # too much at once (as the table may be very large from before we # started pruning). # # This works by finding the max last_seen that is less than the given # time, but has no more than N rows before it, deleting all rows with # a lesser last_seen time. (We COALESCE so that the sub-SELECT always # returns exactly one row). sql = """ DELETE FROM user_ips WHERE last_seen <= ( SELECT COALESCE(MAX(last_seen), -1) FROM ( SELECT last_seen FROM user_ips WHERE last_seen <= ? ORDER BY last_seen ASC LIMIT 5000 ) AS u ) """ timestamp = self.clock.time_msec() - self.user_ips_max_age def _prune_old_user_ips_txn(txn): txn.execute(sql, (timestamp, )) await self.db.runInteraction("_prune_old_user_ips", _prune_old_user_ips_txn)
class EventsWorkerStore(SQLBaseStore): def __init__(self, database: Database, db_conn, hs): super(EventsWorkerStore, self).__init__(database, db_conn, hs) self._get_event_cache = Cache("*getEvent*", keylen=3, max_entries=hs.config.event_cache_size) self._event_fetch_lock = threading.Condition() self._event_fetch_list = [] self._event_fetch_ongoing = 0 def get_received_ts(self, event_id): """Get received_ts (when it was persisted) for the event. Raises an exception for unknown events. Args: event_id (str) Returns: Deferred[int|None]: Timestamp in milliseconds, or None for events that were persisted before received_ts was implemented. """ return self.db.simple_select_one_onecol( table="events", keyvalues={"event_id": event_id}, retcol="received_ts", desc="get_received_ts", ) def get_received_ts_by_stream_pos(self, stream_ordering): """Given a stream ordering get an approximate timestamp of when it happened. This is done by simply taking the received ts of the first event that has a stream ordering greater than or equal to the given stream pos. If none exists returns the current time, on the assumption that it must have happened recently. Args: stream_ordering (int) Returns: Deferred[int] """ def _get_approximate_received_ts_txn(txn): sql = """ SELECT received_ts FROM events WHERE stream_ordering >= ? LIMIT 1 """ txn.execute(sql, (stream_ordering, )) row = txn.fetchone() if row and row[0]: ts = row[0] else: ts = self.clock.time_msec() return ts return self.db.runInteraction("get_approximate_received_ts", _get_approximate_received_ts_txn) @defer.inlineCallbacks def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: bool = False, check_room_id: Optional[str] = None, ): """Get an event from the database by event_id. Args: event_id: The event_id of the event to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (behave as per allow_none if the event is redacted) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, behave as per allow_none. allow_none: If True, return None if no event found, if False throw a NotFoundError check_room_id: if not None, check the room of the found event. If there is a mismatch, behave as per allow_none. Returns: Deferred[EventBase|None] """ if not isinstance(event_id, str): raise TypeError("Invalid event event_id %r" % (event_id, )) events = yield self.get_events_as_list( [event_id], redact_behaviour=redact_behaviour, get_prev_content=get_prev_content, allow_rejected=allow_rejected, ) event = events[0] if events else None if event is not None and check_room_id is not None: if event.room_id != check_room_id: event = None if event is None and not allow_none: raise NotFoundError("Could not find event %s" % (event_id, )) return event @defer.inlineCallbacks def get_events( self, event_ids: List[str], redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, ): """Get events from the database Args: event_ids: The event_ids of the events to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (omit them from the response) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, omits rejeted events from the response. Returns: Deferred : Dict from event_id to event. """ events = yield self.get_events_as_list( event_ids, redact_behaviour=redact_behaviour, get_prev_content=get_prev_content, allow_rejected=allow_rejected, ) return {e.event_id: e for e in events} @defer.inlineCallbacks def get_events_as_list( self, event_ids: List[str], redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, ): """Get events from the database and return in a list in the same order as given by `event_ids` arg. Unknown events will be omitted from the response. Args: event_ids: The event_ids of the events to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (omit them from the response) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, omits rejected events from the response. Returns: Deferred[list[EventBase]]: List of events fetched from the database. The events are in the same order as `event_ids` arg. Note that the returned list may be smaller than the list of event IDs if not all events could be fetched. """ if not event_ids: return [] # there may be duplicates so we cast the list to a set event_entry_map = yield self._get_events_from_cache_or_db( set(event_ids), allow_rejected=allow_rejected) events = [] for event_id in event_ids: entry = event_entry_map.get(event_id, None) if not entry: continue if not allow_rejected: assert not entry.event.rejected_reason, ( "rejected event returned from _get_events_from_cache_or_db despite " "allow_rejected=False") # We may not have had the original event when we received a redaction, so # we have to recheck auth now. if not allow_rejected and entry.event.type == EventTypes.Redaction: if not hasattr(entry.event, "redacts"): # A redacted redaction doesn't have a `redacts` key, in # which case lets just withhold the event. # # Note: Most of the time if the redactions has been # redacted we still have the un-redacted event in the DB # and so we'll still see the `redacts` key. However, this # isn't always true e.g. if we have censored the event. logger.debug( "Withholding redaction event %s as we don't have redacts key", event_id, ) continue redacted_event_id = entry.event.redacts event_map = yield self._get_events_from_cache_or_db( [redacted_event_id]) original_event_entry = event_map.get(redacted_event_id) if not original_event_entry: # we don't have the redacted event (or it was rejected). # # We assume that the redaction isn't authorized for now; if the # redacted event later turns up, the redaction will be re-checked, # and if it is found valid, the original will get redacted before it # is served to the client. logger.debug( "Withholding redaction event %s since we don't (yet) have the " "original %s", event_id, redacted_event_id, ) continue original_event = original_event_entry.event if original_event.type == EventTypes.Create: # we never serve redactions of Creates to clients. logger.info( "Withholding redaction %s of create event %s", event_id, redacted_event_id, ) continue if original_event.room_id != entry.event.room_id: logger.info( "Withholding redaction %s of event %s from a different room", event_id, redacted_event_id, ) continue if entry.event.internal_metadata.need_to_check_redaction(): original_domain = get_domain_from_id(original_event.sender) redaction_domain = get_domain_from_id(entry.event.sender) if original_domain != redaction_domain: # the senders don't match, so this is forbidden logger.info( "Withholding redaction %s whose sender domain %s doesn't " "match that of redacted event %s %s", event_id, redaction_domain, redacted_event_id, original_domain, ) continue # Update the cache to save doing the checks again. entry.event.internal_metadata.recheck_redaction = False event = entry.event if entry.redacted_event: if redact_behaviour == EventRedactBehaviour.BLOCK: # Skip this event continue elif redact_behaviour == EventRedactBehaviour.REDACT: event = entry.redacted_event events.append(event) if get_prev_content: if "replaces_state" in event.unsigned: prev = yield self.get_event( event.unsigned["replaces_state"], get_prev_content=False, allow_none=True, ) if prev: event.unsigned = dict(event.unsigned) event.unsigned["prev_content"] = prev.content event.unsigned["prev_sender"] = prev.sender return events @defer.inlineCallbacks def _get_events_from_cache_or_db(self, event_ids, allow_rejected=False): """Fetch a bunch of events from the cache or the database. If events are pulled from the database, they will be cached for future lookups. Unknown events are omitted from the response. Args: event_ids (Iterable[str]): The event_ids of the events to fetch allow_rejected (bool): Whether to include rejected events. If False, rejected events are omitted from the response. Returns: Deferred[Dict[str, _EventCacheEntry]]: map from event id to result """ event_entry_map = self._get_events_from_cache( event_ids, allow_rejected=allow_rejected) missing_events_ids = [e for e in event_ids if e not in event_entry_map] if missing_events_ids: log_ctx = LoggingContext.current_context() log_ctx.record_event_fetch(len(missing_events_ids)) # Note that _get_events_from_db is also responsible for turning db rows # into FrozenEvents (via _get_event_from_row), which involves seeing if # the events have been redacted, and if so pulling the redaction event out # of the database to check it. # missing_events = yield self._get_events_from_db( missing_events_ids, allow_rejected=allow_rejected) event_entry_map.update(missing_events) return event_entry_map def _invalidate_get_event_cache(self, event_id): self._get_event_cache.invalidate((event_id, )) def _get_events_from_cache(self, events, allow_rejected, update_metrics=True): """Fetch events from the caches Args: events (Iterable[str]): list of event_ids to fetch allow_rejected (bool): Whether to return events that were rejected update_metrics (bool): Whether to update the cache hit ratio metrics Returns: dict of event_id -> _EventCacheEntry for each event_id in cache. If allow_rejected is `False` then there will still be an entry but it will be `None` """ event_map = {} for event_id in events: ret = self._get_event_cache.get((event_id, ), None, update_metrics=update_metrics) if not ret: continue if allow_rejected or not ret.event.rejected_reason: event_map[event_id] = ret else: event_map[event_id] = None return event_map def _do_fetch(self, conn): """Takes a database connection and waits for requests for events from the _event_fetch_list queue. """ i = 0 while True: with self._event_fetch_lock: event_list = self._event_fetch_list self._event_fetch_list = [] if not event_list: single_threaded = self.database_engine.single_threaded if single_threaded or i > EVENT_QUEUE_ITERATIONS: self._event_fetch_ongoing -= 1 return else: self._event_fetch_lock.wait(EVENT_QUEUE_TIMEOUT_S) i += 1 continue i = 0 self._fetch_event_list(conn, event_list) def _fetch_event_list(self, conn, event_list): """Handle a load of requests from the _event_fetch_list queue Args: conn (twisted.enterprise.adbapi.Connection): database connection event_list (list[Tuple[list[str], Deferred]]): The fetch requests. Each entry consists of a list of event ids to be fetched, and a deferred to be completed once the events have been fetched. The deferreds are callbacked with a dictionary mapping from event id to event row. Note that it may well contain additional events that were not part of this request. """ with Measure(self._clock, "_fetch_event_list"): try: events_to_fetch = set(event_id for events, _ in event_list for event_id in events) row_dict = self.db.new_transaction(conn, "do_fetch", [], [], self._fetch_event_rows, events_to_fetch) # We only want to resolve deferreds from the main thread def fire(): for _, d in event_list: d.callback(row_dict) with PreserveLoggingContext(): self.hs.get_reactor().callFromThread(fire) except Exception as e: logger.exception("do_fetch") # We only want to resolve deferreds from the main thread def fire(evs, exc): for _, d in evs: if not d.called: with PreserveLoggingContext(): d.errback(exc) with PreserveLoggingContext(): self.hs.get_reactor().callFromThread(fire, event_list, e) @defer.inlineCallbacks def _get_events_from_db(self, event_ids, allow_rejected=False): """Fetch a bunch of events from the database. Returned events will be added to the cache for future lookups. Unknown events are omitted from the response. Args: event_ids (Iterable[str]): The event_ids of the events to fetch allow_rejected (bool): Whether to include rejected events. If False, rejected events are omitted from the response. Returns: Deferred[Dict[str, _EventCacheEntry]]: map from event id to result. May return extra events which weren't asked for. """ fetched_events = {} events_to_fetch = event_ids while events_to_fetch: row_map = yield self._enqueue_events(events_to_fetch) # we need to recursively fetch any redactions of those events redaction_ids = set() for event_id in events_to_fetch: row = row_map.get(event_id) fetched_events[event_id] = row if row: redaction_ids.update(row["redactions"]) events_to_fetch = redaction_ids.difference(fetched_events.keys()) if events_to_fetch: logger.debug("Also fetching redaction events %s", events_to_fetch) # build a map from event_id to EventBase event_map = {} for event_id, row in fetched_events.items(): if not row: continue assert row["event_id"] == event_id rejected_reason = row["rejected_reason"] if not allow_rejected and rejected_reason: continue d = json.loads(row["json"]) internal_metadata = json.loads(row["internal_metadata"]) format_version = row["format_version"] if format_version is None: # This means that we stored the event before we had the concept # of a event format version, so it must be a V1 event. format_version = EventFormatVersions.V1 original_ev = event_type_from_format_version(format_version)( event_dict=d, internal_metadata_dict=internal_metadata, rejected_reason=rejected_reason, ) event_map[event_id] = original_ev # finally, we can decide whether each one nededs redacting, and build # the cache entries. result_map = {} for event_id, original_ev in event_map.items(): redactions = fetched_events[event_id]["redactions"] redacted_event = self._maybe_redact_event_row( original_ev, redactions, event_map) cache_entry = _EventCacheEntry(event=original_ev, redacted_event=redacted_event) self._get_event_cache.prefill((event_id, ), cache_entry) result_map[event_id] = cache_entry return result_map @defer.inlineCallbacks def _enqueue_events(self, events): """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. Args: events (Iterable[str]): events to be fetched. Returns: Deferred[Dict[str, Dict]]: map from event id to row data from the database. May contain events that weren't requested. """ events_d = defer.Deferred() with self._event_fetch_lock: self._event_fetch_list.append((events, events_d)) self._event_fetch_lock.notify() if self._event_fetch_ongoing < EVENT_QUEUE_THREADS: self._event_fetch_ongoing += 1 should_start = True else: should_start = False if should_start: run_as_background_process("fetch_events", self.db.runWithConnection, self._do_fetch) logger.debug("Loading %d events: %s", len(events), events) with PreserveLoggingContext(): row_map = yield events_d logger.debug("Loaded %d events (%d rows)", len(events), len(row_map)) return row_map def _fetch_event_rows(self, txn, event_ids): """Fetch event rows from the database Events which are not found are omitted from the result. The returned per-event dicts contain the following keys: * event_id (str) * json (str): json-encoded event structure * internal_metadata (str): json-encoded internal metadata dict * format_version (int|None): The format of the event. Hopefully one of EventFormatVersions. 'None' means the event predates EventFormatVersions (so the event is format V1). * rejected_reason (str|None): if the event was rejected, the reason why. * redactions (List[str]): a list of event-ids which (claim to) redact this event. Args: txn (twisted.enterprise.adbapi.Connection): event_ids (Iterable[str]): event IDs to fetch Returns: Dict[str, Dict]: a map from event id to event info. """ event_dict = {} for evs in batch_iter(event_ids, 200): sql = ("SELECT " " e.event_id, " " e.internal_metadata," " e.json," " e.format_version, " " rej.reason " " FROM event_json as e" " LEFT JOIN rejections as rej USING (event_id)" " WHERE ") clause, args = make_in_list_sql_clause(txn.database_engine, "e.event_id", evs) txn.execute(sql + clause, args) for row in txn: event_id = row[0] event_dict[event_id] = { "event_id": event_id, "internal_metadata": row[1], "json": row[2], "format_version": row[3], "rejected_reason": row[4], "redactions": [], } # check for redactions redactions_sql = "SELECT event_id, redacts FROM redactions WHERE " clause, args = make_in_list_sql_clause(txn.database_engine, "redacts", evs) txn.execute(redactions_sql + clause, args) for (redacter, redacted) in txn: d = event_dict.get(redacted) if d: d["redactions"].append(redacter) return event_dict def _maybe_redact_event_row(self, original_ev, redactions, event_map): """Given an event object and a list of possible redacting event ids, determine whether to honour any of those redactions and if so return a redacted event. Args: original_ev (EventBase): redactions (iterable[str]): list of event ids of potential redaction events event_map (dict[str, EventBase]): other events which have been fetched, in which we can look up the redaaction events. Map from event id to event. Returns: Deferred[EventBase|None]: if the event should be redacted, a pruned event object. Otherwise, None. """ if original_ev.type == "m.room.create": # we choose to ignore redactions of m.room.create events. return None for redaction_id in redactions: redaction_event = event_map.get(redaction_id) if not redaction_event or redaction_event.rejected_reason: # we don't have the redaction event, or the redaction event was not # authorized. logger.debug( "%s was redacted by %s but redaction not found/authed", original_ev.event_id, redaction_id, ) continue if redaction_event.room_id != original_ev.room_id: logger.debug( "%s was redacted by %s but redaction was in a different room!", original_ev.event_id, redaction_id, ) continue # Starting in room version v3, some redactions need to be # rechecked if we didn't have the redacted event at the # time, so we recheck on read instead. if redaction_event.internal_metadata.need_to_check_redaction(): expected_domain = get_domain_from_id(original_ev.sender) if get_domain_from_id( redaction_event.sender) == expected_domain: # This redaction event is allowed. Mark as not needing a recheck. redaction_event.internal_metadata.recheck_redaction = False else: # Senders don't match, so the event isn't actually redacted logger.debug( "%s was redacted by %s but the senders don't match", original_ev.event_id, redaction_id, ) continue logger.debug("Redacting %s due to %s", original_ev.event_id, redaction_id) # we found a good redaction event. Redact! redacted_event = prune_event(original_ev) redacted_event.unsigned["redacted_by"] = redaction_id # It's fine to add the event directly, since get_pdu_json # will serialise this field correctly redacted_event.unsigned["redacted_because"] = redaction_event return redacted_event # no valid redaction found for this event return None @defer.inlineCallbacks def have_events_in_timeline(self, event_ids): """Given a list of event ids, check if we have already processed and stored them as non outliers. """ rows = yield self.db.simple_select_many_batch( table="events", retcols=("event_id", ), column="event_id", iterable=list(event_ids), keyvalues={"outlier": False}, desc="have_events_in_timeline", ) return set(r["event_id"] for r in rows) @defer.inlineCallbacks def have_seen_events(self, event_ids): """Given a list of event ids, check if we have already processed them. Args: event_ids (iterable[str]): Returns: Deferred[set[str]]: The events we have already seen. """ results = set() def have_seen_events_txn(txn, chunk): sql = "SELECT event_id FROM events as e WHERE " clause, args = make_in_list_sql_clause(txn.database_engine, "e.event_id", chunk) txn.execute(sql + clause, args) for (event_id, ) in txn: results.add(event_id) # break the input up into chunks of 100 input_iterator = iter(event_ids) for chunk in iter(lambda: list(itertools.islice(input_iterator, 100)), []): yield self.db.runInteraction("have_seen_events", have_seen_events_txn, chunk) return results def _get_total_state_event_counts_txn(self, txn, room_id): """ See get_total_state_event_counts. """ # We join against the events table as that has an index on room_id sql = """ SELECT COUNT(*) FROM state_events INNER JOIN events USING (room_id, event_id) WHERE room_id=? """ txn.execute(sql, (room_id, )) row = txn.fetchone() return row[0] if row else 0 def get_total_state_event_counts(self, room_id): """ Gets the total number of state events in a room. Args: room_id (str) Returns: Deferred[int] """ return self.db.runInteraction( "get_total_state_event_counts", self._get_total_state_event_counts_txn, room_id, ) def _get_current_state_event_counts_txn(self, txn, room_id): """ See get_current_state_event_counts. """ sql = "SELECT COUNT(*) FROM current_state_events WHERE room_id=?" txn.execute(sql, (room_id, )) row = txn.fetchone() return row[0] if row else 0 def get_current_state_event_counts(self, room_id): """ Gets the current number of state events in a room. Args: room_id (str) Returns: Deferred[int] """ return self.db.runInteraction( "get_current_state_event_counts", self._get_current_state_event_counts_txn, room_id, ) @defer.inlineCallbacks def get_room_complexity(self, room_id): """ Get a rough approximation of the complexity of the room. This is used by remote servers to decide whether they wish to join the room or not. Higher complexity value indicates that being in the room will consume more resources. Args: room_id (str) Returns: Deferred[dict[str:int]] of complexity version to complexity. """ state_events = yield self.get_current_state_event_counts(room_id) # Call this one "v1", so we can introduce new ones as we want to develop # it. complexity_v1 = round(state_events / 500, 2) return {"v1": complexity_v1}
class CacheTestCase(unittest.TestCase): def setUp(self): self.cache = Cache("test") def test_empty(self): failed = False try: self.cache.get("foo") except KeyError: failed = True self.assertTrue(failed) def test_hit(self): self.cache.prefill("foo", 123) self.assertEquals(self.cache.get("foo"), 123) def test_invalidate(self): self.cache.prefill(("foo",), 123) self.cache.invalidate(("foo",)) failed = False try: self.cache.get(("foo",)) except KeyError: failed = True self.assertTrue(failed) def test_eviction(self): cache = Cache("test", max_entries=2) cache.prefill(1, "one") cache.prefill(2, "two") cache.prefill(3, "three") # 1 will be evicted failed = False try: cache.get(1) except KeyError: failed = True self.assertTrue(failed) cache.get(2) cache.get(3) def test_eviction_lru(self): cache = Cache("test", max_entries=2, lru=True) cache.prefill(1, "one") cache.prefill(2, "two") # Now access 1 again, thus causing 2 to be least-recently used cache.get(1) cache.prefill(3, "three") failed = False try: cache.get(2) except KeyError: failed = True self.assertTrue(failed) cache.get(1) cache.get(3)
class EventsWorkerStore(SQLBaseStore): def __init__(self, database: DatabasePool, db_conn, hs): super(EventsWorkerStore, self).__init__(database, db_conn, hs) if hs.config.worker.writers.events == hs.get_instance_name(): # We are the process in charge of generating stream ids for events, # so instantiate ID generators based on the database self._stream_id_gen = StreamIdGenerator( db_conn, "events", "stream_ordering", ) self._backfill_id_gen = StreamIdGenerator( db_conn, "events", "stream_ordering", step=-1, extra_tables=[("ex_outlier_stream", "event_stream_ordering")], ) else: # Another process is in charge of persisting events and generating # stream IDs: rely on the replication streams to let us know which # IDs we can process. self._stream_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering") self._backfill_id_gen = SlavedIdTracker(db_conn, "events", "stream_ordering", step=-1) self._get_event_cache = Cache( "*getEvent*", keylen=3, max_entries=hs.config.caches.event_cache_size, apply_cache_factor_from_config=False, ) self._event_fetch_lock = threading.Condition() self._event_fetch_list = [] self._event_fetch_ongoing = 0 def process_replication_rows(self, stream_name, instance_name, token, rows): if stream_name == EventsStream.NAME: self._stream_id_gen.advance(instance_name, token) elif stream_name == BackfillStream.NAME: self._backfill_id_gen.advance(instance_name, -token) super().process_replication_rows(stream_name, instance_name, token, rows) async def get_received_ts(self, event_id: str) -> Optional[int]: """Get received_ts (when it was persisted) for the event. Raises an exception for unknown events. Args: event_id: The event ID to query. Returns: Timestamp in milliseconds, or None for events that were persisted before received_ts was implemented. """ return await self.db_pool.simple_select_one_onecol( table="events", keyvalues={"event_id": event_id}, retcol="received_ts", desc="get_received_ts", ) # Inform mypy that if allow_none is False (the default) then get_event # always returns an EventBase. @overload async def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: Literal[False] = False, check_room_id: Optional[str] = None, ) -> EventBase: ... @overload async def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: Literal[True] = False, check_room_id: Optional[str] = None, ) -> Optional[EventBase]: ... async def get_event( self, event_id: str, redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, allow_none: bool = False, check_room_id: Optional[str] = None, ) -> Optional[EventBase]: """Get an event from the database by event_id. Args: event_id: The event_id of the event to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (behave as per allow_none if the event is redacted) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, behave as per allow_none. allow_none: If True, return None if no event found, if False throw a NotFoundError check_room_id: if not None, check the room of the found event. If there is a mismatch, behave as per allow_none. Returns: The event, or None if the event was not found. """ if not isinstance(event_id, str): raise TypeError("Invalid event event_id %r" % (event_id, )) events = await self.get_events_as_list( [event_id], redact_behaviour=redact_behaviour, get_prev_content=get_prev_content, allow_rejected=allow_rejected, ) event = events[0] if events else None if event is not None and check_room_id is not None: if event.room_id != check_room_id: event = None if event is None and not allow_none: raise NotFoundError("Could not find event %s" % (event_id, )) return event async def get_events( self, event_ids: Iterable[str], redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, ) -> Dict[str, EventBase]: """Get events from the database Args: event_ids: The event_ids of the events to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (omit them from the response) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, omits rejeted events from the response. Returns: A mapping from event_id to event. """ events = await self.get_events_as_list( event_ids, redact_behaviour=redact_behaviour, get_prev_content=get_prev_content, allow_rejected=allow_rejected, ) return {e.event_id: e for e in events} async def get_events_as_list( self, event_ids: Collection[str], redact_behaviour: EventRedactBehaviour = EventRedactBehaviour.REDACT, get_prev_content: bool = False, allow_rejected: bool = False, ) -> List[EventBase]: """Get events from the database and return in a list in the same order as given by `event_ids` arg. Unknown events will be omitted from the response. Args: event_ids: The event_ids of the events to fetch redact_behaviour: Determine what to do with a redacted event. Possible values: * AS_IS - Return the full event body with no redacted content * REDACT - Return the event but with a redacted body * DISALLOW - Do not return redacted events (omit them from the response) get_prev_content: If True and event is a state event, include the previous states content in the unsigned field. allow_rejected: If True, return rejected events. Otherwise, omits rejected events from the response. Returns: List of events fetched from the database. The events are in the same order as `event_ids` arg. Note that the returned list may be smaller than the list of event IDs if not all events could be fetched. """ if not event_ids: return [] # there may be duplicates so we cast the list to a set event_entry_map = await self._get_events_from_cache_or_db( set(event_ids), allow_rejected=allow_rejected) events = [] for event_id in event_ids: entry = event_entry_map.get(event_id, None) if not entry: continue if not allow_rejected: assert not entry.event.rejected_reason, ( "rejected event returned from _get_events_from_cache_or_db despite " "allow_rejected=False") # We may not have had the original event when we received a redaction, so # we have to recheck auth now. if not allow_rejected and entry.event.type == EventTypes.Redaction: if entry.event.redacts is None: # A redacted redaction doesn't have a `redacts` key, in # which case lets just withhold the event. # # Note: Most of the time if the redactions has been # redacted we still have the un-redacted event in the DB # and so we'll still see the `redacts` key. However, this # isn't always true e.g. if we have censored the event. logger.debug( "Withholding redaction event %s as we don't have redacts key", event_id, ) continue redacted_event_id = entry.event.redacts event_map = await self._get_events_from_cache_or_db( [redacted_event_id]) original_event_entry = event_map.get(redacted_event_id) if not original_event_entry: # we don't have the redacted event (or it was rejected). # # We assume that the redaction isn't authorized for now; if the # redacted event later turns up, the redaction will be re-checked, # and if it is found valid, the original will get redacted before it # is served to the client. logger.debug( "Withholding redaction event %s since we don't (yet) have the " "original %s", event_id, redacted_event_id, ) continue original_event = original_event_entry.event if original_event.type == EventTypes.Create: # we never serve redactions of Creates to clients. logger.info( "Withholding redaction %s of create event %s", event_id, redacted_event_id, ) continue if original_event.room_id != entry.event.room_id: logger.info( "Withholding redaction %s of event %s from a different room", event_id, redacted_event_id, ) continue if entry.event.internal_metadata.need_to_check_redaction(): original_domain = get_domain_from_id(original_event.sender) redaction_domain = get_domain_from_id(entry.event.sender) if original_domain != redaction_domain: # the senders don't match, so this is forbidden logger.info( "Withholding redaction %s whose sender domain %s doesn't " "match that of redacted event %s %s", event_id, redaction_domain, redacted_event_id, original_domain, ) continue # Update the cache to save doing the checks again. entry.event.internal_metadata.recheck_redaction = False event = entry.event if entry.redacted_event: if redact_behaviour == EventRedactBehaviour.BLOCK: # Skip this event continue elif redact_behaviour == EventRedactBehaviour.REDACT: event = entry.redacted_event events.append(event) if get_prev_content: if "replaces_state" in event.unsigned: prev = await self.get_event( event.unsigned["replaces_state"], get_prev_content=False, allow_none=True, ) if prev: event.unsigned = dict(event.unsigned) event.unsigned["prev_content"] = prev.content event.unsigned["prev_sender"] = prev.sender return events async def _get_events_from_cache_or_db(self, event_ids, allow_rejected=False): """Fetch a bunch of events from the cache or the database. If events are pulled from the database, they will be cached for future lookups. Unknown events are omitted from the response. Args: event_ids (Iterable[str]): The event_ids of the events to fetch allow_rejected (bool): Whether to include rejected events. If False, rejected events are omitted from the response. Returns: Dict[str, _EventCacheEntry]: map from event id to result """ event_entry_map = self._get_events_from_cache( event_ids, allow_rejected=allow_rejected) missing_events_ids = [e for e in event_ids if e not in event_entry_map] if missing_events_ids: log_ctx = current_context() log_ctx.record_event_fetch(len(missing_events_ids)) # Note that _get_events_from_db is also responsible for turning db rows # into FrozenEvents (via _get_event_from_row), which involves seeing if # the events have been redacted, and if so pulling the redaction event out # of the database to check it. # missing_events = await self._get_events_from_db( missing_events_ids, allow_rejected=allow_rejected) event_entry_map.update(missing_events) return event_entry_map def _invalidate_get_event_cache(self, event_id): self._get_event_cache.invalidate((event_id, )) def _get_events_from_cache(self, events, allow_rejected, update_metrics=True): """Fetch events from the caches Args: events (Iterable[str]): list of event_ids to fetch allow_rejected (bool): Whether to return events that were rejected update_metrics (bool): Whether to update the cache hit ratio metrics Returns: dict of event_id -> _EventCacheEntry for each event_id in cache. If allow_rejected is `False` then there will still be an entry but it will be `None` """ event_map = {} for event_id in events: ret = self._get_event_cache.get((event_id, ), None, update_metrics=update_metrics) if not ret: continue if allow_rejected or not ret.event.rejected_reason: event_map[event_id] = ret else: event_map[event_id] = None return event_map def _do_fetch(self, conn): """Takes a database connection and waits for requests for events from the _event_fetch_list queue. """ i = 0 while True: with self._event_fetch_lock: event_list = self._event_fetch_list self._event_fetch_list = [] if not event_list: single_threaded = self.database_engine.single_threaded if single_threaded or i > EVENT_QUEUE_ITERATIONS: self._event_fetch_ongoing -= 1 return else: self._event_fetch_lock.wait(EVENT_QUEUE_TIMEOUT_S) i += 1 continue i = 0 self._fetch_event_list(conn, event_list) def _fetch_event_list(self, conn, event_list): """Handle a load of requests from the _event_fetch_list queue Args: conn (twisted.enterprise.adbapi.Connection): database connection event_list (list[Tuple[list[str], Deferred]]): The fetch requests. Each entry consists of a list of event ids to be fetched, and a deferred to be completed once the events have been fetched. The deferreds are callbacked with a dictionary mapping from event id to event row. Note that it may well contain additional events that were not part of this request. """ with Measure(self._clock, "_fetch_event_list"): try: events_to_fetch = { event_id for events, _ in event_list for event_id in events } row_dict = self.db_pool.new_transaction( conn, "do_fetch", [], [], self._fetch_event_rows, events_to_fetch) # We only want to resolve deferreds from the main thread def fire(): for _, d in event_list: d.callback(row_dict) with PreserveLoggingContext(): self.hs.get_reactor().callFromThread(fire) except Exception as e: logger.exception("do_fetch") # We only want to resolve deferreds from the main thread def fire(evs, exc): for _, d in evs: if not d.called: with PreserveLoggingContext(): d.errback(exc) with PreserveLoggingContext(): self.hs.get_reactor().callFromThread(fire, event_list, e) async def _get_events_from_db(self, event_ids, allow_rejected=False): """Fetch a bunch of events from the database. Returned events will be added to the cache for future lookups. Unknown events are omitted from the response. Args: event_ids (Iterable[str]): The event_ids of the events to fetch allow_rejected (bool): Whether to include rejected events. If False, rejected events are omitted from the response. Returns: Dict[str, _EventCacheEntry]: map from event id to result. May return extra events which weren't asked for. """ fetched_events = {} events_to_fetch = event_ids while events_to_fetch: row_map = await self._enqueue_events(events_to_fetch) # we need to recursively fetch any redactions of those events redaction_ids = set() for event_id in events_to_fetch: row = row_map.get(event_id) fetched_events[event_id] = row if row: redaction_ids.update(row["redactions"]) events_to_fetch = redaction_ids.difference(fetched_events.keys()) if events_to_fetch: logger.debug("Also fetching redaction events %s", events_to_fetch) # build a map from event_id to EventBase event_map = {} for event_id, row in fetched_events.items(): if not row: continue assert row["event_id"] == event_id rejected_reason = row["rejected_reason"] if not allow_rejected and rejected_reason: continue # If the event or metadata cannot be parsed, log the error and act # as if the event is unknown. try: d = db_to_json(row["json"]) except ValueError: logger.error("Unable to parse json from event: %s", event_id) continue try: internal_metadata = db_to_json(row["internal_metadata"]) except ValueError: logger.error( "Unable to parse internal_metadata from event: %s", event_id) continue format_version = row["format_version"] if format_version is None: # This means that we stored the event before we had the concept # of a event format version, so it must be a V1 event. format_version = EventFormatVersions.V1 room_version_id = row["room_version_id"] if not room_version_id: # this should only happen for out-of-band membership events which # arrived before #6983 landed. For all other events, we should have # an entry in the 'rooms' table. # # However, the 'out_of_band_membership' flag is unreliable for older # invites, so just accept it for all membership events. # if d["type"] != EventTypes.Member: raise Exception("Room %s for event %s is unknown" % (d["room_id"], event_id)) # so, assuming this is an out-of-band-invite that arrived before #6983 # landed, we know that the room version must be v5 or earlier (because # v6 hadn't been invented at that point, so invites from such rooms # would have been rejected.) # # The main reason we need to know the room version here (other than # choosing the right python Event class) is in case the event later has # to be redacted - and all the room versions up to v5 used the same # redaction algorithm. # # So, the following approximations should be adequate. if format_version == EventFormatVersions.V1: # if it's event format v1 then it must be room v1 or v2 room_version = RoomVersions.V1 elif format_version == EventFormatVersions.V2: # if it's event format v2 then it must be room v3 room_version = RoomVersions.V3 else: # if it's event format v3 then it must be room v4 or v5 room_version = RoomVersions.V5 else: room_version = KNOWN_ROOM_VERSIONS.get(room_version_id) if not room_version: logger.warning( "Event %s in room %s has unknown room version %s", event_id, d["room_id"], room_version_id, ) continue if room_version.event_format != format_version: logger.error( "Event %s in room %s with version %s has wrong format: " "expected %s, was %s", event_id, d["room_id"], room_version_id, room_version.event_format, format_version, ) continue original_ev = make_event_from_dict( event_dict=d, room_version=room_version, internal_metadata_dict=internal_metadata, rejected_reason=rejected_reason, ) event_map[event_id] = original_ev # finally, we can decide whether each one needs redacting, and build # the cache entries. result_map = {} for event_id, original_ev in event_map.items(): redactions = fetched_events[event_id]["redactions"] redacted_event = self._maybe_redact_event_row( original_ev, redactions, event_map) cache_entry = _EventCacheEntry(event=original_ev, redacted_event=redacted_event) self._get_event_cache.prefill((event_id, ), cache_entry) result_map[event_id] = cache_entry return result_map async def _enqueue_events(self, events): """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. Args: events (Iterable[str]): events to be fetched. Returns: Dict[str, Dict]: map from event id to row data from the database. May contain events that weren't requested. """ events_d = defer.Deferred() with self._event_fetch_lock: self._event_fetch_list.append((events, events_d)) self._event_fetch_lock.notify() if self._event_fetch_ongoing < EVENT_QUEUE_THREADS: self._event_fetch_ongoing += 1 should_start = True else: should_start = False if should_start: run_as_background_process("fetch_events", self.db_pool.runWithConnection, self._do_fetch) logger.debug("Loading %d events: %s", len(events), events) with PreserveLoggingContext(): row_map = await events_d logger.debug("Loaded %d events (%d rows)", len(events), len(row_map)) return row_map def _fetch_event_rows(self, txn, event_ids): """Fetch event rows from the database Events which are not found are omitted from the result. The returned per-event dicts contain the following keys: * event_id (str) * json (str): json-encoded event structure * internal_metadata (str): json-encoded internal metadata dict * format_version (int|None): The format of the event. Hopefully one of EventFormatVersions. 'None' means the event predates EventFormatVersions (so the event is format V1). * room_version_id (str|None): The version of the room which contains the event. Hopefully one of RoomVersions. Due to historical reasons, there may be a few events in the database which do not have an associated room; in this case None will be returned here. * rejected_reason (str|None): if the event was rejected, the reason why. * redactions (List[str]): a list of event-ids which (claim to) redact this event. Args: txn (twisted.enterprise.adbapi.Connection): event_ids (Iterable[str]): event IDs to fetch Returns: Dict[str, Dict]: a map from event id to event info. """ event_dict = {} for evs in batch_iter(event_ids, 200): sql = """\ SELECT e.event_id, e.internal_metadata, e.json, e.format_version, r.room_version, rej.reason FROM event_json as e LEFT JOIN rooms r USING (room_id) LEFT JOIN rejections as rej USING (event_id) WHERE """ clause, args = make_in_list_sql_clause(txn.database_engine, "e.event_id", evs) txn.execute(sql + clause, args) for row in txn: event_id = row[0] event_dict[event_id] = { "event_id": event_id, "internal_metadata": row[1], "json": row[2], "format_version": row[3], "room_version_id": row[4], "rejected_reason": row[5], "redactions": [], } # check for redactions redactions_sql = "SELECT event_id, redacts FROM redactions WHERE " clause, args = make_in_list_sql_clause(txn.database_engine, "redacts", evs) txn.execute(redactions_sql + clause, args) for (redacter, redacted) in txn: d = event_dict.get(redacted) if d: d["redactions"].append(redacter) return event_dict def _maybe_redact_event_row( self, original_ev: EventBase, redactions: Iterable[str], event_map: Dict[str, EventBase], ) -> Optional[EventBase]: """Given an event object and a list of possible redacting event ids, determine whether to honour any of those redactions and if so return a redacted event. Args: original_ev: The original event. redactions: list of event ids of potential redaction events event_map: other events which have been fetched, in which we can look up the redaaction events. Map from event id to event. Returns: If the event should be redacted, a pruned event object. Otherwise, None. """ if original_ev.type == "m.room.create": # we choose to ignore redactions of m.room.create events. return None for redaction_id in redactions: redaction_event = event_map.get(redaction_id) if not redaction_event or redaction_event.rejected_reason: # we don't have the redaction event, or the redaction event was not # authorized. logger.debug( "%s was redacted by %s but redaction not found/authed", original_ev.event_id, redaction_id, ) continue if redaction_event.room_id != original_ev.room_id: logger.debug( "%s was redacted by %s but redaction was in a different room!", original_ev.event_id, redaction_id, ) continue # Starting in room version v3, some redactions need to be # rechecked if we didn't have the redacted event at the # time, so we recheck on read instead. if redaction_event.internal_metadata.need_to_check_redaction(): expected_domain = get_domain_from_id(original_ev.sender) if get_domain_from_id( redaction_event.sender) == expected_domain: # This redaction event is allowed. Mark as not needing a recheck. redaction_event.internal_metadata.recheck_redaction = False else: # Senders don't match, so the event isn't actually redacted logger.debug( "%s was redacted by %s but the senders don't match", original_ev.event_id, redaction_id, ) continue logger.debug("Redacting %s due to %s", original_ev.event_id, redaction_id) # we found a good redaction event. Redact! redacted_event = prune_event(original_ev) redacted_event.unsigned["redacted_by"] = redaction_id # It's fine to add the event directly, since get_pdu_json # will serialise this field correctly redacted_event.unsigned["redacted_because"] = redaction_event return redacted_event # no valid redaction found for this event return None async def have_events_in_timeline(self, event_ids): """Given a list of event ids, check if we have already processed and stored them as non outliers. """ rows = await self.db_pool.simple_select_many_batch( table="events", retcols=("event_id", ), column="event_id", iterable=list(event_ids), keyvalues={"outlier": False}, desc="have_events_in_timeline", ) return {r["event_id"] for r in rows} async def have_seen_events(self, event_ids): """Given a list of event ids, check if we have already processed them. Args: event_ids (iterable[str]): Returns: set[str]: The events we have already seen. """ results = set() def have_seen_events_txn(txn, chunk): sql = "SELECT event_id FROM events as e WHERE " clause, args = make_in_list_sql_clause(txn.database_engine, "e.event_id", chunk) txn.execute(sql + clause, args) for (event_id, ) in txn: results.add(event_id) # break the input up into chunks of 100 input_iterator = iter(event_ids) for chunk in iter(lambda: list(itertools.islice(input_iterator, 100)), []): await self.db_pool.runInteraction("have_seen_events", have_seen_events_txn, chunk) return results def _get_current_state_event_counts_txn(self, txn, room_id): """ See get_current_state_event_counts. """ sql = "SELECT COUNT(*) FROM current_state_events WHERE room_id=?" txn.execute(sql, (room_id, )) row = txn.fetchone() return row[0] if row else 0 async def get_current_state_event_counts(self, room_id: str) -> int: """ Gets the current number of state events in a room. Args: room_id: The room ID to query. Returns: The current number of state events. """ return await self.db_pool.runInteraction( "get_current_state_event_counts", self._get_current_state_event_counts_txn, room_id, ) async def get_room_complexity(self, room_id): """ Get a rough approximation of the complexity of the room. This is used by remote servers to decide whether they wish to join the room or not. Higher complexity value indicates that being in the room will consume more resources. Args: room_id (str) Returns: dict[str:int] of complexity version to complexity. """ state_events = await self.get_current_state_event_counts(room_id) # Call this one "v1", so we can introduce new ones as we want to develop # it. complexity_v1 = round(state_events / 500, 2) return {"v1": complexity_v1} 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() async def get_all_new_forward_event_rows(self, last_id: int, current_id: int, limit: int) -> List[Tuple]: """Returns new events, for the Events replication stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to limit: the maximum number of rows to return Returns: a list of events stream rows. Each tuple consists of a stream id as the first element, followed by fields suitable for casting into an EventsStreamRow. """ def get_all_new_forward_event_rows(txn): sql = ("SELECT e.stream_ordering, e.event_id, e.room_id, e.type," " state_key, redacts, relates_to_id" " FROM events AS e" " LEFT JOIN redactions USING (event_id)" " LEFT JOIN state_events USING (event_id)" " 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)) return txn.fetchall() return await self.db_pool.runInteraction( "get_all_new_forward_event_rows", get_all_new_forward_event_rows) async def get_ex_outlier_stream_rows(self, last_id: int, current_id: int) -> List[Tuple]: """Returns de-outliered events, for the Events replication stream Args: last_id: the last stream_id from the previous batch. current_id: the maximum stream_id to return up to Returns: a list of events stream rows. Each tuple consists of a stream id as the first element, followed by fields suitable for casting into an EventsStreamRow. """ def get_ex_outlier_stream_rows_txn(txn): sql = ( "SELECT event_stream_ordering, e.event_id, e.room_id, e.type," " 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)" " LEFT JOIN event_relations USING (event_id)" " WHERE ? < event_stream_ordering" " AND event_stream_ordering <= ?" " ORDER BY event_stream_ordering ASC") txn.execute(sql, (last_id, current_id)) return txn.fetchall() return await self.db_pool.runInteraction( "get_ex_outlier_stream_rows", get_ex_outlier_stream_rows_txn) async def get_all_new_backfill_event_rows( self, instance_name: str, last_id: int, current_id: int, limit: int) -> Tuple[List[Tuple[int, list]], int, bool]: """Get updates for backfill replication stream, including all new backfilled events and events that have gone from being outliers to not. Args: instance_name: The writer we want to fetch updates from. Unused here since there is only ever one writer. last_id: The token to fetch updates from. Exclusive. current_id: The token to fetch updates up to. Inclusive. limit: The requested limit for the number of rows to return. The function may return more or fewer rows. Returns: A tuple consisting of: the updates, a token to use to fetch subsequent updates, and whether we returned fewer rows than exists between the requested tokens due to the limit. The token returned can be used in a subsequent call to this function to get further updatees. The updates are a list of 2-tuples of stream ID and the row data """ if last_id == current_id: return [], current_id, False def get_all_new_backfill_event_rows(txn): sql = ("SELECT -e.stream_ordering, e.event_id, e.room_id, e.type," " state_key, redacts, relates_to_id" " FROM events AS e" " LEFT JOIN redactions USING (event_id)" " LEFT JOIN state_events USING (event_id)" " 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 = [(row[0], row[1:]) for row in txn] limited = False if len(new_event_updates) == limit: upper_bound = new_event_updates[-1][0] limited = True else: upper_bound = current_id sql = ( "SELECT -event_stream_ordering, e.event_id, e.room_id, e.type," " 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)" " 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((row[0], row[1:]) for row in txn) if len(new_event_updates) >= limit: upper_bound = new_event_updates[-1][0] limited = True return new_event_updates, upper_bound, limited return await self.db_pool.runInteraction( "get_all_new_backfill_event_rows", get_all_new_backfill_event_rows) async def get_all_updated_current_state_deltas( self, from_token: int, to_token: int, target_row_count: int) -> Tuple[List[Tuple], int, bool]: """Fetch updates from current_state_delta_stream Args: from_token: The previous stream token. Updates from this stream id will be excluded. to_token: The current stream token (ie the upper limit). Updates up to this stream id will be included (modulo the 'limit' param) target_row_count: The number of rows to try to return. If more rows are available, we will set 'limited' in the result. In the event of a large batch, we may return more rows than this. Returns: A triplet `(updates, new_last_token, limited)`, where: * `updates` is a list of database tuples. * `new_last_token` is the new position in stream. * `limited` is whether there are more updates to fetch. """ 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, target_row_count)) return txn.fetchall() def get_deltas_for_stream_id_txn(txn, stream_id): sql = """ SELECT stream_id, room_id, type, state_key, event_id FROM current_state_delta_stream WHERE stream_id = ? """ txn.execute(sql, [stream_id]) return txn.fetchall() # we need to make sure that, for every stream id in the results, we get *all* # the rows with that stream id. rows = await self.db_pool.runInteraction( "get_all_updated_current_state_deltas", get_all_updated_current_state_deltas_txn, ) # type: List[Tuple] # if we've got fewer rows than the limit, we're good if len(rows) < target_row_count: return rows, to_token, False # we hit the limit, so reduce the upper limit so that we exclude the stream id # of the last row in the result. assert rows[-1][0] <= to_token to_token = rows[-1][0] - 1 # search backwards through the list for the point to truncate for idx in range(len(rows) - 1, 0, -1): if rows[idx - 1][0] <= to_token: return rows[:idx], to_token, True # bother. We didn't get a full set of changes for even a single # stream id. let's run the query again, without a row limit, but for # just one stream id. to_token += 1 rows = await self.db_pool.runInteraction("get_deltas_for_stream_id", get_deltas_for_stream_id_txn, to_token) return rows, to_token, True async def is_event_after(self, event_id1, event_id2): """Returns True if event_id1 is after event_id2 in the stream """ to_1, so_1 = await self.get_event_ordering(event_id1) to_2, so_2 = await self.get_event_ordering(event_id2) return (to_1, so_1) > (to_2, so_2) @cached(max_entries=5000) async def get_event_ordering(self, event_id): res = await self.db_pool.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"])) async def get_next_event_to_expire(self) -> Optional[Tuple[str, int]]: """Retrieve the entry with the lowest expiry timestamp in the event_expiry table, or None if there's no more event to expire. Returns: A tuple containing the event ID as its first element and an expiry timestamp as its second one, if there's at least one row in the event_expiry table. None otherwise. """ def get_next_event_to_expire_txn(txn): txn.execute(""" SELECT event_id, expiry_ts FROM event_expiry ORDER BY expiry_ts ASC LIMIT 1 """) return txn.fetchone() return await self.db_pool.runInteraction( desc="get_next_event_to_expire", func=get_next_event_to_expire_txn)