diff options
| author | Jason Robinson <jasonr@matrix.org> | 2021-01-23 21:41:35 +0200 |
|---|---|---|
| committer | Jason Robinson <jasonr@matrix.org> | 2021-01-23 21:41:35 +0200 |
| commit | 8965b6cfec8a1de847efe3d1be4b9babf4622e2e (patch) | |
| tree | 4551f104ee2ce840689aa5ecffa939938482ffd5 /synapse/storage/databases | |
| parent | Add depth and received_ts to forward_extremities admin API response (diff) | |
| parent | Return a 404 if no valid thumbnail is found. (#9163) (diff) | |
| download | synapse-8965b6cfec8a1de847efe3d1be4b9babf4622e2e.tar.xz | |
Merge branch 'develop' into jaywink/admin-forward-extremities
Diffstat (limited to 'synapse/storage/databases')
35 files changed, 1871 insertions, 349 deletions
diff --git a/synapse/storage/databases/main/__init__.py b/synapse/storage/databases/main/__init__.py index b936f54f1e..5d0845588c 100644 --- a/synapse/storage/databases/main/__init__.py +++ b/synapse/storage/databases/main/__init__.py @@ -162,9 +162,13 @@ class DataStore( database, stream_name="caches", instance_name=hs.get_instance_name(), - table="cache_invalidation_stream_by_instance", - instance_column="instance_name", - id_column="stream_id", + tables=[ + ( + "cache_invalidation_stream_by_instance", + "instance_name", + "stream_id", + ) + ], sequence_name="cache_invalidation_stream_seq", writers=[], ) diff --git a/synapse/storage/databases/main/account_data.py b/synapse/storage/databases/main/account_data.py index bff51e92b9..a277a1ef13 100644 --- a/synapse/storage/databases/main/account_data.py +++ b/synapse/storage/databases/main/account_data.py @@ -14,14 +14,16 @@ # See the License for the specific language governing permissions and # limitations under the License. -import abc import logging from typing import Dict, List, Optional, Set, Tuple from synapse.api.constants import AccountDataTypes +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker +from synapse.replication.tcp.streams import AccountDataStream, TagAccountDataStream from synapse.storage._base import SQLBaseStore, db_to_json from synapse.storage.database import DatabasePool -from synapse.storage.util.id_generators import StreamIdGenerator +from synapse.storage.engines import PostgresEngine +from synapse.storage.util.id_generators import MultiWriterIdGenerator, StreamIdGenerator from synapse.types import JsonDict from synapse.util import json_encoder from synapse.util.caches.descriptors import cached @@ -30,14 +32,57 @@ from synapse.util.caches.stream_change_cache import StreamChangeCache logger = logging.getLogger(__name__) -# The ABCMeta metaclass ensures that it cannot be instantiated without -# the abstract methods being implemented. -class AccountDataWorkerStore(SQLBaseStore, metaclass=abc.ABCMeta): +class AccountDataWorkerStore(SQLBaseStore): """This is an abstract base class where subclasses must implement `get_max_account_data_stream_id` which can be called in the initializer. """ def __init__(self, database: DatabasePool, db_conn, hs): + self._instance_name = hs.get_instance_name() + + if isinstance(database.engine, PostgresEngine): + self._can_write_to_account_data = ( + self._instance_name in hs.config.worker.writers.account_data + ) + + self._account_data_id_gen = MultiWriterIdGenerator( + db_conn=db_conn, + db=database, + stream_name="account_data", + instance_name=self._instance_name, + tables=[ + ("room_account_data", "instance_name", "stream_id"), + ("room_tags_revisions", "instance_name", "stream_id"), + ("account_data", "instance_name", "stream_id"), + ], + sequence_name="account_data_sequence", + writers=hs.config.worker.writers.account_data, + ) + else: + self._can_write_to_account_data = True + + # We shouldn't be running in worker mode with SQLite, but its useful + # to support it for unit tests. + # + # If this process is the writer than we need to use + # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets + # updated over replication. (Multiple writers are not supported for + # SQLite). + if hs.get_instance_name() in hs.config.worker.writers.account_data: + self._account_data_id_gen = StreamIdGenerator( + db_conn, + "room_account_data", + "stream_id", + extra_tables=[("room_tags_revisions", "stream_id")], + ) + else: + self._account_data_id_gen = SlavedIdTracker( + db_conn, + "room_account_data", + "stream_id", + extra_tables=[("room_tags_revisions", "stream_id")], + ) + account_max = self.get_max_account_data_stream_id() self._account_data_stream_cache = StreamChangeCache( "AccountDataAndTagsChangeCache", account_max @@ -45,14 +90,13 @@ class AccountDataWorkerStore(SQLBaseStore, metaclass=abc.ABCMeta): super().__init__(database, db_conn, hs) - @abc.abstractmethod - def get_max_account_data_stream_id(self): + def get_max_account_data_stream_id(self) -> int: """Get the current max stream ID for account data stream Returns: int """ - raise NotImplementedError() + return self._account_data_id_gen.get_current_token() @cached() async def get_account_data_for_user( @@ -307,28 +351,26 @@ class AccountDataWorkerStore(SQLBaseStore, metaclass=abc.ABCMeta): ) ) - -class AccountDataStore(AccountDataWorkerStore): - def __init__(self, database: DatabasePool, db_conn, hs): - self._account_data_id_gen = StreamIdGenerator( - db_conn, - "account_data_max_stream_id", - "stream_id", - extra_tables=[ - ("room_account_data", "stream_id"), - ("room_tags_revisions", "stream_id"), - ], - ) - - super().__init__(database, db_conn, hs) - - def get_max_account_data_stream_id(self) -> int: - """Get the current max stream id for the private user data stream - - Returns: - The maximum stream ID. - """ - return self._account_data_id_gen.get_current_token() + def process_replication_rows(self, stream_name, instance_name, token, rows): + if stream_name == TagAccountDataStream.NAME: + self._account_data_id_gen.advance(instance_name, token) + for row in rows: + self.get_tags_for_user.invalidate((row.user_id,)) + self._account_data_stream_cache.entity_has_changed(row.user_id, token) + elif stream_name == AccountDataStream.NAME: + self._account_data_id_gen.advance(instance_name, token) + for row in rows: + if not row.room_id: + self.get_global_account_data_by_type_for_user.invalidate( + (row.data_type, row.user_id) + ) + self.get_account_data_for_user.invalidate((row.user_id,)) + self.get_account_data_for_room.invalidate((row.user_id, row.room_id)) + self.get_account_data_for_room_and_type.invalidate( + (row.user_id, row.room_id, row.data_type) + ) + self._account_data_stream_cache.entity_has_changed(row.user_id, token) + return super().process_replication_rows(stream_name, instance_name, token, rows) async def add_account_data_to_room( self, user_id: str, room_id: str, account_data_type: str, content: JsonDict @@ -344,6 +386,8 @@ class AccountDataStore(AccountDataWorkerStore): Returns: The maximum stream ID. """ + assert self._can_write_to_account_data + content_json = json_encoder.encode(content) async with self._account_data_id_gen.get_next() as next_id: @@ -362,14 +406,6 @@ class AccountDataStore(AccountDataWorkerStore): lock=False, ) - # it's theoretically possible for the above to succeed and the - # below to fail - in which case we might reuse a stream id on - # restart, and the above update might not get propagated. That - # doesn't sound any worse than the whole update getting lost, - # which is what would happen if we combined the two into one - # transaction. - await self._update_max_stream_id(next_id) - self._account_data_stream_cache.entity_has_changed(user_id, next_id) self.get_account_data_for_user.invalidate((user_id,)) self.get_account_data_for_room.invalidate((user_id, room_id)) @@ -392,6 +428,8 @@ class AccountDataStore(AccountDataWorkerStore): Returns: The maximum stream ID. """ + assert self._can_write_to_account_data + async with self._account_data_id_gen.get_next() as next_id: await self.db_pool.runInteraction( "add_user_account_data", @@ -402,18 +440,6 @@ class AccountDataStore(AccountDataWorkerStore): content, ) - # it's theoretically possible for the above to succeed and the - # below to fail - in which case we might reuse a stream id on - # restart, and the above update might not get propagated. That - # doesn't sound any worse than the whole update getting lost, - # which is what would happen if we combined the two into one - # transaction. - # - # Note: This is only here for backwards compat to allow admins to - # roll back to a previous Synapse version. Next time we update the - # database version we can remove this table. - await self._update_max_stream_id(next_id) - self._account_data_stream_cache.entity_has_changed(user_id, next_id) self.get_account_data_for_user.invalidate((user_id,)) self.get_global_account_data_by_type_for_user.invalidate( @@ -487,23 +513,6 @@ class AccountDataStore(AccountDataWorkerStore): for ignored_user_id in previously_ignored_users ^ currently_ignored_users: self._invalidate_cache_and_stream(txn, self.ignored_by, (ignored_user_id,)) - async def _update_max_stream_id(self, next_id: int) -> None: - """Update the max stream_id - - Args: - next_id: The the revision to advance to. - """ - # Note: This is only here for backwards compat to allow admins to - # roll back to a previous Synapse version. Next time we update the - # database version we can remove this table. - - def _update(txn): - update_max_id_sql = ( - "UPDATE account_data_max_stream_id" - " SET stream_id = ?" - " WHERE stream_id < ?" - ) - txn.execute(update_max_id_sql, (next_id, next_id)) - - await self.db_pool.runInteraction("update_account_data_max_stream_id", _update) +class AccountDataStore(AccountDataWorkerStore): + pass diff --git a/synapse/storage/databases/main/client_ips.py b/synapse/storage/databases/main/client_ips.py index c53c836337..ea1e8fb580 100644 --- a/synapse/storage/databases/main/client_ips.py +++ b/synapse/storage/databases/main/client_ips.py @@ -407,6 +407,34 @@ class ClientIpWorkerStore(ClientIpBackgroundUpdateStore): "_prune_old_user_ips", _prune_old_user_ips_txn ) + async def get_last_client_ip_by_device( + self, user_id: str, device_id: Optional[str] + ) -> Dict[Tuple[str, str], dict]: + """For each device_id listed, give the user_ip it was last seen on. + + The result might be slightly out of date as client IPs are inserted in batches. + + Args: + user_id: The user to fetch devices for. + device_id: If None fetches all devices for the user + + Returns: + A dictionary mapping a tuple of (user_id, device_id) to dicts, with + keys giving the column names from the devices table. + """ + + keyvalues = {"user_id": user_id} + if device_id is not None: + keyvalues["device_id"] = device_id + + res = await self.db_pool.simple_select_list( + table="devices", + keyvalues=keyvalues, + retcols=("user_id", "ip", "user_agent", "device_id", "last_seen"), + ) + + return {(d["user_id"], d["device_id"]): d for d in res} + class ClientIpStore(ClientIpWorkerStore): def __init__(self, database: DatabasePool, db_conn, hs): @@ -512,18 +540,9 @@ class ClientIpStore(ClientIpWorkerStore): A dictionary mapping a tuple of (user_id, device_id) to dicts, with keys giving the column names from the devices table. """ + ret = await super().get_last_client_ip_by_device(user_id, device_id) - keyvalues = {"user_id": user_id} - if device_id is not None: - keyvalues["device_id"] = device_id - - res = await self.db_pool.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} + # Update what is retrieved from the database with data which is pending insertion. for key in self._batch_row_update: uid, access_token, ip = key if uid == user_id: diff --git a/synapse/storage/databases/main/deviceinbox.py b/synapse/storage/databases/main/deviceinbox.py index 58d3f71e45..31f70ac5ef 100644 --- a/synapse/storage/databases/main/deviceinbox.py +++ b/synapse/storage/databases/main/deviceinbox.py @@ -54,9 +54,7 @@ class DeviceInboxWorkerStore(SQLBaseStore): db=database, stream_name="to_device", instance_name=self._instance_name, - table="device_inbox", - instance_column="instance_name", - id_column="stream_id", + tables=[("device_inbox", "instance_name", "stream_id")], sequence_name="device_inbox_sequence", writers=hs.config.worker.writers.to_device, ) diff --git a/synapse/storage/databases/main/devices.py b/synapse/storage/databases/main/devices.py index 9097677648..659d8f245f 100644 --- a/synapse/storage/databases/main/devices.py +++ b/synapse/storage/databases/main/devices.py @@ -897,7 +897,7 @@ class DeviceWorkerStore(SQLBaseStore): DELETE FROM device_lists_outbound_last_success WHERE destination = ? AND user_id = ? """ - txn.executemany(sql, ((row[0], row[1]) for row in rows)) + txn.execute_batch(sql, ((row[0], row[1]) for row in rows)) logger.info("Pruned %d device list outbound pokes", count) @@ -1343,7 +1343,7 @@ class DeviceStore(DeviceWorkerStore, DeviceBackgroundUpdateStore): # Delete older entries in the table, as we really only care about # when the latest change happened. - txn.executemany( + txn.execute_batch( """ DELETE FROM device_lists_stream WHERE user_id = ? AND device_id = ? AND stream_id < ? diff --git a/synapse/storage/databases/main/end_to_end_keys.py b/synapse/storage/databases/main/end_to_end_keys.py index 4d1b92d1aa..c128889bf9 100644 --- a/synapse/storage/databases/main/end_to_end_keys.py +++ b/synapse/storage/databases/main/end_to_end_keys.py @@ -25,6 +25,7 @@ from twisted.enterprise.adbapi import Connection from synapse.logging.opentracing import log_kv, set_tag, trace from synapse.storage._base import SQLBaseStore, db_to_json from synapse.storage.database import DatabasePool, make_in_list_sql_clause +from synapse.storage.engines import PostgresEngine from synapse.storage.types import Cursor from synapse.types import JsonDict from synapse.util import json_encoder @@ -513,21 +514,35 @@ class EndToEndKeyWorkerStore(EndToEndKeyBackgroundStore): for user_chunk in batch_iter(user_ids, 100): clause, params = make_in_list_sql_clause( - txn.database_engine, "k.user_id", user_chunk - ) - sql = ( - """ - SELECT k.user_id, k.keytype, k.keydata, k.stream_id - FROM e2e_cross_signing_keys k - INNER JOIN (SELECT user_id, keytype, MAX(stream_id) AS stream_id - FROM e2e_cross_signing_keys - GROUP BY user_id, keytype) s - USING (user_id, stream_id, keytype) - WHERE - """ - + clause + txn.database_engine, "user_id", user_chunk ) + # Fetch the latest key for each type per user. + if isinstance(self.database_engine, PostgresEngine): + # The `DISTINCT ON` clause will pick the *first* row it + # encounters, so ordering by stream ID desc will ensure we get + # the latest key. + sql = """ + SELECT DISTINCT ON (user_id, keytype) user_id, keytype, keydata, stream_id + FROM e2e_cross_signing_keys + WHERE %(clause)s + ORDER BY user_id, keytype, stream_id DESC + """ % { + "clause": clause + } + else: + # SQLite has special handling for bare columns when using + # MIN/MAX with a `GROUP BY` clause where it picks the value from + # a row that matches the MIN/MAX. + sql = """ + SELECT user_id, keytype, keydata, MAX(stream_id) + FROM e2e_cross_signing_keys + WHERE %(clause)s + GROUP BY user_id, keytype + """ % { + "clause": clause + } + txn.execute(sql, params) rows = self.db_pool.cursor_to_dict(txn) @@ -707,50 +722,6 @@ class EndToEndKeyWorkerStore(EndToEndKeyBackgroundStore): """Get the current stream id from the _device_list_id_gen""" ... - -class EndToEndKeyStore(EndToEndKeyWorkerStore, SQLBaseStore): - async def set_e2e_device_keys( - self, user_id: str, device_id: str, time_now: int, device_keys: JsonDict - ) -> bool: - """Stores device keys for a device. Returns whether there was a change - or the keys were already in the database. - """ - - def _set_e2e_device_keys_txn(txn): - set_tag("user_id", user_id) - set_tag("device_id", device_id) - set_tag("time_now", time_now) - set_tag("device_keys", device_keys) - - old_key_json = self.db_pool.simple_select_one_onecol_txn( - txn, - table="e2e_device_keys_json", - keyvalues={"user_id": user_id, "device_id": device_id}, - retcol="key_json", - allow_none=True, - ) - - # In py3 we need old_key_json to match new_key_json type. The DB - # returns unicode while encode_canonical_json returns bytes. - new_key_json = encode_canonical_json(device_keys).decode("utf-8") - - if old_key_json == new_key_json: - log_kv({"Message": "Device key already stored."}) - return False - - self.db_pool.simple_upsert_txn( - txn, - table="e2e_device_keys_json", - keyvalues={"user_id": user_id, "device_id": device_id}, - values={"ts_added_ms": time_now, "key_json": new_key_json}, - ) - log_kv({"message": "Device keys stored."}) - return True - - return await self.db_pool.runInteraction( - "set_e2e_device_keys", _set_e2e_device_keys_txn - ) - async def claim_e2e_one_time_keys( self, query_list: Iterable[Tuple[str, str, str]] ) -> Dict[str, Dict[str, Dict[str, bytes]]]: @@ -840,6 +811,50 @@ class EndToEndKeyStore(EndToEndKeyWorkerStore, SQLBaseStore): "claim_e2e_one_time_keys", _claim_e2e_one_time_keys ) + +class EndToEndKeyStore(EndToEndKeyWorkerStore, SQLBaseStore): + async def set_e2e_device_keys( + self, user_id: str, device_id: str, time_now: int, device_keys: JsonDict + ) -> bool: + """Stores device keys for a device. Returns whether there was a change + or the keys were already in the database. + """ + + def _set_e2e_device_keys_txn(txn): + set_tag("user_id", user_id) + set_tag("device_id", device_id) + set_tag("time_now", time_now) + set_tag("device_keys", device_keys) + + old_key_json = self.db_pool.simple_select_one_onecol_txn( + txn, + table="e2e_device_keys_json", + keyvalues={"user_id": user_id, "device_id": device_id}, + retcol="key_json", + allow_none=True, + ) + + # In py3 we need old_key_json to match new_key_json type. The DB + # returns unicode while encode_canonical_json returns bytes. + new_key_json = encode_canonical_json(device_keys).decode("utf-8") + + if old_key_json == new_key_json: + log_kv({"Message": "Device key already stored."}) + return False + + self.db_pool.simple_upsert_txn( + txn, + table="e2e_device_keys_json", + keyvalues={"user_id": user_id, "device_id": device_id}, + values={"ts_added_ms": time_now, "key_json": new_key_json}, + ) + log_kv({"message": "Device keys stored."}) + return True + + return await self.db_pool.runInteraction( + "set_e2e_device_keys", _set_e2e_device_keys_txn + ) + async def delete_e2e_keys_by_device(self, user_id: str, device_id: str) -> None: def delete_e2e_keys_by_device_txn(txn): log_kv( diff --git a/synapse/storage/databases/main/event_federation.py b/synapse/storage/databases/main/event_federation.py index ebffd89251..8326640d20 100644 --- a/synapse/storage/databases/main/event_federation.py +++ b/synapse/storage/databases/main/event_federation.py @@ -24,6 +24,8 @@ from synapse.storage._base import SQLBaseStore, make_in_list_sql_clause from synapse.storage.database import DatabasePool, LoggingTransaction from synapse.storage.databases.main.events_worker import EventsWorkerStore from synapse.storage.databases.main.signatures import SignatureWorkerStore +from synapse.storage.engines import PostgresEngine +from synapse.storage.types import Cursor from synapse.types import Collection from synapse.util.caches.descriptors import cached from synapse.util.caches.lrucache import LruCache @@ -32,6 +34,11 @@ from synapse.util.iterutils import batch_iter logger = logging.getLogger(__name__) +class _NoChainCoverIndex(Exception): + def __init__(self, room_id: str): + super().__init__("Unexpectedly no chain cover for events in %s" % (room_id,)) + + class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBaseStore): def __init__(self, database: DatabasePool, db_conn, hs): super().__init__(database, db_conn, hs) @@ -151,15 +158,193 @@ class EventFederationWorkerStore(EventsWorkerStore, SignatureWorkerStore, SQLBas The set of the difference in auth chains. """ + # Check if we have indexed the room so we can use the chain cover + # algorithm. + room = await self.get_room(room_id) + if room["has_auth_chain_index"]: + try: + return await self.db_pool.runInteraction( + "get_auth_chain_difference_chains", + self._get_auth_chain_difference_using_cover_index_txn, + room_id, + state_sets, + ) + except _NoChainCoverIndex: + # For whatever reason we don't actually have a chain cover index + # for the events in question, so we fall back to the old method. + pass + return await self.db_pool.runInteraction( "get_auth_chain_difference", self._get_auth_chain_difference_txn, state_sets, ) + def _get_auth_chain_difference_using_cover_index_txn( + self, txn: Cursor, room_id: str, state_sets: List[Set[str]] + ) -> Set[str]: + """Calculates the auth chain difference using the chain index. + + See docs/auth_chain_difference_algorithm.md for details + """ + + # First we look up the chain ID/sequence numbers for all the events, and + # work out the chain/sequence numbers reachable from each state set. + + initial_events = set(state_sets[0]).union(*state_sets[1:]) + + # Map from event_id -> (chain ID, seq no) + chain_info = {} # type: Dict[str, Tuple[int, int]] + + # Map from chain ID -> seq no -> event Id + chain_to_event = {} # type: Dict[int, Dict[int, str]] + + # All the chains that we've found that are reachable from the state + # sets. + seen_chains = set() # type: Set[int] + + sql = """ + SELECT event_id, chain_id, sequence_number + FROM event_auth_chains + WHERE %s + """ + for batch in batch_iter(initial_events, 1000): + clause, args = make_in_list_sql_clause( + txn.database_engine, "event_id", batch + ) + txn.execute(sql % (clause,), args) + + for event_id, chain_id, sequence_number in txn: + chain_info[event_id] = (chain_id, sequence_number) + seen_chains.add(chain_id) + chain_to_event.setdefault(chain_id, {})[sequence_number] = event_id + + # Check that we actually have a chain ID for all the events. + events_missing_chain_info = initial_events.difference(chain_info) + if events_missing_chain_info: + # This can happen due to e.g. downgrade/upgrade of the server. We + # raise an exception and fall back to the previous algorithm. + logger.info( + "Unexpectedly found that events don't have chain IDs in room %s: %s", + room_id, + events_missing_chain_info, + ) + raise _NoChainCoverIndex(room_id) + + # Corresponds to `state_sets`, except as a map from chain ID to max + # sequence number reachable from the state set. + set_to_chain = [] # type: List[Dict[int, int]] + for state_set in state_sets: + chains = {} # type: Dict[int, int] + set_to_chain.append(chains) + + for event_id in state_set: + chain_id, seq_no = chain_info[event_id] + + chains[chain_id] = max(seq_no, chains.get(chain_id, 0)) + + # Now we look up all links for the chains we have, adding chains to + # set_to_chain that are reachable from each set. + sql = """ + SELECT + origin_chain_id, origin_sequence_number, + target_chain_id, target_sequence_number + FROM event_auth_chain_links + WHERE %s + """ + + # (We need to take a copy of `seen_chains` as we want to mutate it in + # the loop) + for batch in batch_iter(set(seen_chains), 1000): + clause, args = make_in_list_sql_clause( + txn.database_engine, "origin_chain_id", batch + ) + txn.execute(sql % (clause,), args) + + for ( + origin_chain_id, + origin_sequence_number, + target_chain_id, + target_sequence_number, + ) in txn: + for chains in set_to_chain: + # chains are only reachable if the origin sequence number of + # the link is less than the max sequence number in the + # origin chain. + if origin_sequence_number <= chains.get(origin_chain_id, 0): + chains[target_chain_id] = max( + target_sequence_number, chains.get(target_chain_id, 0), + ) + + seen_chains.add(target_chain_id) + + # Now for each chain we figure out the maximum sequence number reachable + # from *any* state set and the minimum sequence number reachable from + # *all* state sets. Events in that range are in the auth chain + # difference. + result = set() + + # Mapping from chain ID to the range of sequence numbers that should be + # pulled from the database. + chain_to_gap = {} # type: Dict[int, Tuple[int, int]] + + for chain_id in seen_chains: + min_seq_no = min(chains.get(chain_id, 0) for chains in set_to_chain) + max_seq_no = max(chains.get(chain_id, 0) for chains in set_to_chain) + + if min_seq_no < max_seq_no: + # We have a non empty gap, try and fill it from the events that + # we have, otherwise add them to the list of gaps to pull out + # from the DB. + for seq_no in range(min_seq_no + 1, max_seq_no + 1): + event_id = chain_to_event.get(chain_id, {}).get(seq_no) + if event_id: + result.add(event_id) + else: + chain_to_gap[chain_id] = (min_seq_no, max_seq_no) + break + + if not chain_to_gap: + # If there are no gaps to fetch, we're done! + return result + + if isinstance(self.database_engine, PostgresEngine): + # We can use `execute_values` to efficiently fetch the gaps when + # using postgres. + sql = """ + SELECT event_id + FROM event_auth_chains AS c, (VALUES ?) AS l(chain_id, min_seq, max_seq) + WHERE + c.chain_id = l.chain_id + AND min_seq < sequence_number AND sequence_number <= max_seq + """ + + args = [ + (chain_id, min_no, max_no) + for chain_id, (min_no, max_no) in chain_to_gap.items() + ] + + rows = txn.execute_values(sql, args) + result.update(r for r, in rows) + else: + # For SQLite we just fall back to doing a noddy for loop. + sql = """ + SELECT event_id FROM event_auth_chains + WHERE chain_id = ? AND ? < sequence_number AND sequence_number <= ? + """ + for chain_id, (min_no, max_no) in chain_to_gap.items(): + txn.execute(sql, (chain_id, min_no, max_no)) + result.update(r for r, in txn) + + return result + def _get_auth_chain_difference_txn( self, txn, state_sets: List[Set[str]] ) -> Set[str]: + """Calculates the auth chain difference using a breadth first search. + + This is used when we don't have a cover index for the room. + """ # Algorithm Description # ~~~~~~~~~~~~~~~~~~~~~ diff --git a/synapse/storage/databases/main/event_push_actions.py b/synapse/storage/databases/main/event_push_actions.py index e5c03cc609..438383abe1 100644 --- a/synapse/storage/databases/main/event_push_actions.py +++ b/synapse/storage/databases/main/event_push_actions.py @@ -487,7 +487,7 @@ class EventPushActionsWorkerStore(SQLBaseStore): VALUES (?, ?, ?, ?, ?, ?) """ - txn.executemany( + txn.execute_batch( sql, ( _gen_entry(user_id, actions) @@ -803,7 +803,7 @@ class EventPushActionsWorkerStore(SQLBaseStore): ], ) - txn.executemany( + txn.execute_batch( """ UPDATE event_push_summary SET notif_count = ?, unread_count = ?, stream_ordering = ? @@ -835,6 +835,52 @@ class EventPushActionsWorkerStore(SQLBaseStore): (rotate_to_stream_ordering,), ) + def _remove_old_push_actions_before_txn( + self, txn, room_id, user_id, stream_ordering + ): + """ + Purges old push actions for a user and room before a given + stream_ordering. + + We however keep a months worth of highlighted notifications, so that + users can still get a list of recent highlights. + + Args: + txn: The transcation + room_id: Room ID to delete from + user_id: user ID to delete for + stream_ordering: The lowest stream ordering which will + not be deleted. + """ + txn.call_after( + self.get_unread_event_push_actions_by_room_for_user.invalidate_many, + (room_id, user_id), + ) + + # We need to join on the events table to get the received_ts for + # event_push_actions and sqlite won't let us use a join in a delete so + # we can't just delete where received_ts < x. Furthermore we can + # only identify event_push_actions by a tuple of room_id, event_id + # we we can't use a subquery. + # Instead, we look up the stream ordering for the last event in that + # room received before the threshold time and delete event_push_actions + # in the room with a stream_odering before that. + txn.execute( + "DELETE FROM event_push_actions " + " WHERE user_id = ? AND room_id = ? AND " + " stream_ordering <= ?" + " AND ((stream_ordering < ? AND highlight = 1) or highlight = 0)", + (user_id, room_id, stream_ordering, self.stream_ordering_month_ago), + ) + + txn.execute( + """ + DELETE FROM event_push_summary + WHERE room_id = ? AND user_id = ? AND stream_ordering <= ? + """, + (room_id, user_id, stream_ordering), + ) + class EventPushActionsStore(EventPushActionsWorkerStore): EPA_HIGHLIGHT_INDEX = "epa_highlight_index" @@ -894,52 +940,6 @@ class EventPushActionsStore(EventPushActionsWorkerStore): pa["actions"] = _deserialize_action(pa["actions"], pa["highlight"]) return push_actions - def _remove_old_push_actions_before_txn( - self, txn, room_id, user_id, stream_ordering - ): - """ - Purges old push actions for a user and room before a given - stream_ordering. - - We however keep a months worth of highlighted notifications, so that - users can still get a list of recent highlights. - - Args: - txn: The transcation - room_id: Room ID to delete from - user_id: user ID to delete for - stream_ordering: The lowest stream ordering which will - not be deleted. - """ - txn.call_after( - self.get_unread_event_push_actions_by_room_for_user.invalidate_many, - (room_id, user_id), - ) - - # We need to join on the events table to get the received_ts for - # event_push_actions and sqlite won't let us use a join in a delete so - # we can't just delete where received_ts < x. Furthermore we can - # only identify event_push_actions by a tuple of room_id, event_id - # we we can't use a subquery. - # Instead, we look up the stream ordering for the last event in that - # room received before the threshold time and delete event_push_actions - # in the room with a stream_odering before that. - txn.execute( - "DELETE FROM event_push_actions " - " WHERE user_id = ? AND room_id = ? AND " - " stream_ordering <= ?" - " AND ((stream_ordering < ? AND highlight = 1) or highlight = 0)", - (user_id, room_id, stream_ordering, self.stream_ordering_month_ago), - ) - - txn.execute( - """ - DELETE FROM event_push_summary - WHERE room_id = ? AND user_id = ? AND stream_ordering <= ? - """, - (room_id, user_id, stream_ordering), - ) - def _action_has_highlight(actions): for action in actions: diff --git a/synapse/storage/databases/main/events.py b/synapse/storage/databases/main/events.py index 90fb1a1f00..ccda9f1caa 100644 --- a/synapse/storage/databases/main/events.py +++ b/synapse/storage/databases/main/events.py @@ -17,7 +17,17 @@ import itertools import logging from collections import OrderedDict, namedtuple -from typing import TYPE_CHECKING, Any, Dict, Iterable, List, Optional, Set, Tuple +from typing import ( + TYPE_CHECKING, + Any, + Dict, + Generator, + Iterable, + List, + Optional, + Set, + Tuple, +) import attr from prometheus_client import Counter @@ -35,7 +45,7 @@ from synapse.storage.databases.main.search import SearchEntry from synapse.storage.util.id_generators import MultiWriterIdGenerator from synapse.types import StateMap, get_domain_from_id from synapse.util import json_encoder -from synapse.util.iterutils import batch_iter +from synapse.util.iterutils import batch_iter, sorted_topologically if TYPE_CHECKING: from synapse.server import HomeServer @@ -366,6 +376,36 @@ class PersistEventsStore: # Insert into event_to_state_groups. self._store_event_state_mappings_txn(txn, events_and_contexts) + self._persist_event_auth_chain_txn(txn, [e for e, _ in events_and_contexts]) + + # _store_rejected_events_txn filters out any events which were + # rejected, and returns the filtered list. + events_and_contexts = self._store_rejected_events_txn( + txn, events_and_contexts=events_and_contexts + ) + + # From this point onwards the events are only ones that weren't + # rejected. + + self._update_metadata_tables_txn( + txn, + events_and_contexts=events_and_contexts, + all_events_and_contexts=all_events_and_contexts, + backfilled=backfilled, + ) + + # We call this last as it assumes we've inserted the events into + # room_memberships, where applicable. + self._update_current_state_txn(txn, state_delta_for_room, min_stream_order) + + def _persist_event_auth_chain_txn( + self, txn: LoggingTransaction, events: List[EventBase], + ) -> None: + + # We only care about state events, so this if there are no state events. + if not any(e.is_state() for e in events): + return + # We want to store event_auth mappings for rejected events, as they're # used in state res v2. # This is only necessary if the rejected event appears in an accepted @@ -381,31 +421,467 @@ class PersistEventsStore: "room_id": event.room_id, "auth_id": auth_id, } - for event, _ in events_and_contexts + for event in events for auth_id in event.auth_event_ids() if event.is_state() ], ) - # _store_rejected_events_txn filters out any events which were - # rejected, and returns the filtered list. - events_and_contexts = self._store_rejected_events_txn( - txn, events_and_contexts=events_and_contexts + # We now calculate chain ID/sequence numbers for any state events we're + # persisting. We ignore out of band memberships as we're not in the room + # and won't have their auth chain (we'll fix it up later if we join the + # room). + # + # See: docs/auth_chain_difference_algorithm.md + + # We ignore legacy rooms that we aren't filling the chain cover index + # for. + rows = self.db_pool.simple_select_many_txn( + txn, + table="rooms", + column="room_id", + iterable={event.room_id for event in events if event.is_state()}, + keyvalues={}, + retcols=("room_id", "has_auth_chain_index"), ) + rooms_using_chain_index = { + row["room_id"] for row in rows if row["has_auth_chain_index"] + } - # From this point onwards the events are only ones that weren't - # rejected. + state_events = { + event.event_id: event + for event in events + if event.is_state() and event.room_id in rooms_using_chain_index + } - self._update_metadata_tables_txn( + if not state_events: + return + + # We need to know the type/state_key and auth events of the events we're + # calculating chain IDs for. We don't rely on having the full Event + # instances as we'll potentially be pulling more events from the DB and + # we don't need the overhead of fetching/parsing the full event JSON. + event_to_types = { + e.event_id: (e.type, e.state_key) for e in state_events.values() + } + event_to_auth_chain = { + e.event_id: e.auth_event_ids() for e in state_events.values() + } + event_to_room_id = {e.event_id: e.room_id for e in state_events.values()} + + self._add_chain_cover_index( + txn, self.db_pool, event_to_room_id, event_to_types, event_to_auth_chain, + ) + + @classmethod + def _add_chain_cover_index( + cls, + txn, + db_pool: DatabasePool, + event_to_room_id: Dict[str, str], + event_to_types: Dict[str, Tuple[str, str]], + event_to_auth_chain: Dict[str, List[str]], + ) -> None: + """Calculate the chain cover index for the given events. + + Args: + event_to_room_id: Event ID to the room ID of the event + event_to_types: Event ID to type and state_key of the event + event_to_auth_chain: Event ID to list of auth event IDs of the + event (events with no auth events can be excluded). + """ + + # Map from event ID to chain ID/sequence number. + chain_map = {} # type: Dict[str, Tuple[int, int]] + + # Set of event IDs to calculate chain ID/seq numbers for. + events_to_calc_chain_id_for = set(event_to_room_id) + + # We check if there are any events that need to be handled in the rooms + # we're looking at. These should just be out of band memberships, where + # we didn't have the auth chain when we first persisted. + rows = db_pool.simple_select_many_txn( txn, - events_and_contexts=events_and_contexts, - all_events_and_contexts=all_events_and_contexts, - backfilled=backfilled, + table="event_auth_chain_to_calculate", + keyvalues={}, + column="room_id", + iterable=set(event_to_room_id.values()), + retcols=("event_id", "type", "state_key"), ) + for row in rows: + event_id = row["event_id"] + event_type = row["type"] + state_key = row["state_key"] + + # (We could pull out the auth events for all rows at once using + # simple_select_many, but this case happens rarely and almost always + # with a single row.) + auth_events = db_pool.simple_select_onecol_txn( + txn, "event_auth", keyvalues={"event_id": event_id}, retcol="auth_id", + ) - # We call this last as it assumes we've inserted the events into - # room_memberships, where applicable. - self._update_current_state_txn(txn, state_delta_for_room, min_stream_order) + events_to_calc_chain_id_for.add(event_id) + event_to_types[event_id] = (event_type, state_key) + event_to_auth_chain[event_id] = auth_events + + # First we get the chain ID and sequence numbers for the events' + # auth events (that aren't also currently being persisted). + # + # Note that there there is an edge case here where we might not have + # calculated chains and sequence numbers for events that were "out + # of band". We handle this case by fetching the necessary info and + # adding it to the set of events to calculate chain IDs for. + + missing_auth_chains = { + a_id + for auth_events in event_to_auth_chain.values() + for a_id in auth_events + if a_id not in events_to_calc_chain_id_for + } + + # We loop here in case we find an out of band membership and need to + # fetch their auth event info. + while missing_auth_chains: + sql = """ + SELECT event_id, events.type, state_key, chain_id, sequence_number + FROM events + INNER JOIN state_events USING (event_id) + LEFT JOIN event_auth_chains USING (event_id) + WHERE + """ + clause, args = make_in_list_sql_clause( + txn.database_engine, "event_id", missing_auth_chains, + ) + txn.execute(sql + clause, args) + + missing_auth_chains.clear() + + for auth_id, event_type, state_key, chain_id, sequence_number in txn: + event_to_types[auth_id] = (event_type, state_key) + + if chain_id is None: + # No chain ID, so the event was persisted out of band. + # We add to list of events to calculate auth chains for. + + events_to_calc_chain_id_for.add(auth_id) + + event_to_auth_chain[auth_id] = db_pool.simple_select_onecol_txn( + txn, + "event_auth", + keyvalues={"event_id": auth_id}, + retcol="auth_id", + ) + + missing_auth_chains.update( + e + for e in event_to_auth_chain[auth_id] + if e not in event_to_types + ) + else: + chain_map[auth_id] = (chain_id, sequence_number) + + # Now we check if we have any events where we don't have auth chain, + # this should only be out of band memberships. + for event_id in sorted_topologically(event_to_auth_chain, event_to_auth_chain): + for auth_id in event_to_auth_chain[event_id]: + if ( + auth_id not in chain_map + and auth_id not in events_to_calc_chain_id_for + ): + events_to_calc_chain_id_for.discard(event_id) + + # If this is an event we're trying to persist we add it to + # the list of events to calculate chain IDs for next time + # around. (Otherwise we will have already added it to the + # table). + room_id = event_to_room_id.get(event_id) + if room_id: + e_type, state_key = event_to_types[event_id] + db_pool.simple_insert_txn( + txn, + table="event_auth_chain_to_calculate", + values={ + "event_id": event_id, + "room_id": room_id, + "type": e_type, + "state_key": state_key, + }, + ) + + # We stop checking the event's auth events since we've + # discarded it. + break + + if not events_to_calc_chain_id_for: + return + + # Allocate chain ID/sequence numbers to each new event. + new_chain_tuples = cls._allocate_chain_ids( + txn, + db_pool, + event_to_room_id, + event_to_types, + event_to_auth_chain, + events_to_calc_chain_id_for, + chain_map, + ) + chain_map.update(new_chain_tuples) + + db_pool.simple_insert_many_txn( + txn, + table="event_auth_chains", + values=[ + {"event_id": event_id, "chain_id": c_id, "sequence_number": seq} + for event_id, (c_id, seq) in new_chain_tuples.items() + ], + ) + + db_pool.simple_delete_many_txn( + txn, + table="event_auth_chain_to_calculate", + keyvalues={}, + column="event_id", + iterable=new_chain_tuples, + ) + + # Now we need to calculate any new links between chains caused by + # the new events. + # + # Links are pairs of chain ID/sequence numbers such that for any + # event A (CA, SA) and any event B (CB, SB), B is in A's auth chain + # if and only if there is at least one link (CA, S1) -> (CB, S2) + # where SA >= S1 and S2 >= SB. + # + # We try and avoid adding redundant links to the table, e.g. if we + # have two links between two chains which both start/end at the + # sequence number event (or cross) then one can be safely dropped. + # + # To calculate new links we look at every new event and: + # 1. Fetch the chain ID/sequence numbers of its auth events, + # discarding any that are reachable by other auth events, or + # that have the same chain ID as the event. + # 2. For each retained auth event we: + # a. Add a link from the event's to the auth event's chain + # ID/sequence number; and + # b. Add a link from the event to every chain reachable by the + # auth event. + + # Step 1, fetch all existing links from all the chains we've seen + # referenced. + chain_links = _LinkMap() + rows = db_pool.simple_select_many_txn( + txn, + table="event_auth_chain_links", + column="origin_chain_id", + iterable={chain_id for chain_id, _ in chain_map.values()}, + keyvalues={}, + retcols=( + "origin_chain_id", + "origin_sequence_number", + "target_chain_id", + "target_sequence_number", + ), + ) + for row in rows: + chain_links.add_link( + (row["origin_chain_id"], row["origin_sequence_number"]), + (row["target_chain_id"], row["target_sequence_number"]), + new=False, + ) + + # We do this in toplogical order to avoid adding redundant links. + for event_id in sorted_topologically( + events_to_calc_chain_id_for, event_to_auth_chain + ): + chain_id, sequence_number = chain_map[event_id] + + # Filter out auth events that are reachable by other auth + # events. We do this by looking at every permutation of pairs of + # auth events (A, B) to check if B is reachable from A. + reduction = { + a_id + for a_id in event_to_auth_chain.get(event_id, []) + if chain_map[a_id][0] != chain_id + } + for start_auth_id, end_auth_id in itertools.permutations( + event_to_auth_chain.get(event_id, []), r=2, + ): + if chain_links.exists_path_from( + chain_map[start_auth_id], chain_map[end_auth_id] + ): + reduction.discard(end_auth_id) + + # Step 2, figure out what the new links are from the reduced + # list of auth events. + for auth_id in reduction: + auth_chain_id, auth_sequence_number = chain_map[auth_id] + + # Step 2a, add link between the event and auth event + chain_links.add_link( + (chain_id, sequence_number), (auth_chain_id, auth_sequence_number) + ) + + # Step 2b, add a link to chains reachable from the auth + # event. + for target_id, target_seq in chain_links.get_links_from( + (auth_chain_id, auth_sequence_number) + ): + if target_id == chain_id: + continue + + chain_links.add_link( + (chain_id, sequence_number), (target_id, target_seq) + ) + + db_pool.simple_insert_many_txn( + txn, + table="event_auth_chain_links", + values=[ + { + "origin_chain_id": source_id, + "origin_sequence_number": source_seq, + "target_chain_id": target_id, + "target_sequence_number": target_seq, + } + for ( + source_id, + source_seq, + target_id, + target_seq, + ) in chain_links.get_additions() + ], + ) + + @staticmethod + def _allocate_chain_ids( + txn, + db_pool: DatabasePool, + event_to_room_id: Dict[str, str], + event_to_types: Dict[str, Tuple[str, str]], + event_to_auth_chain: Dict[str, List[str]], + events_to_calc_chain_id_for: Set[str], + chain_map: Dict[str, Tuple[int, int]], + ) -> Dict[str, Tuple[int, int]]: + """Allocates, but does not persist, chain ID/sequence numbers for the + events in `events_to_calc_chain_id_for`. (c.f. _add_chain_cover_index + for info on args) + """ + + # We now calculate the chain IDs/sequence numbers for the events. We do + # this by looking at the chain ID and sequence number of any auth event + # with the same type/state_key and incrementing the sequence number by + # one. If there was no match or the chain ID/sequence number is already + # taken we generate a new chain. + # + # We try to reduce the number of times that we hit the database by + # batching up calls, to make this more efficient when persisting large + # numbers of state events (e.g. during joins). + # + # We do this by: + # 1. Calculating for each event which auth event will be used to + # inherit the chain ID, i.e. converting the auth chain graph to a + # tree that we can allocate chains on. We also keep track of which + # existing chain IDs have been referenced. + # 2. Fetching the max allocated sequence number for each referenced + # existing chain ID, generating a map from chain ID to the max + # allocated sequence number. + # 3. Iterating over the tree and allocating a chain ID/seq no. to the + # new event, by incrementing the sequence number from the + # referenced event's chain ID/seq no. and checking that the + # incremented sequence number hasn't already been allocated (by + # looking in the map generated in the previous step). We generate a + # new chain if the sequence number has already been allocated. + # + + existing_chains = set() # type: Set[int] + tree = [] # type: List[Tuple[str, Optional[str]]] + + # We need to do this in a topologically sorted order as we want to + # generate chain IDs/sequence numbers of an event's auth events before + # the event itself. + for event_id in sorted_topologically( + events_to_calc_chain_id_for, event_to_auth_chain + ): + for auth_id in event_to_auth_chain.get(event_id, []): + if event_to_types.get(event_id) == event_to_types.get(auth_id): + existing_chain_id = chain_map.get(auth_id) + if existing_chain_id: + existing_chains.add(existing_chain_id[0]) + + tree.append((event_id, auth_id)) + break + else: + tree.append((event_id, None)) + + # Fetch the current max sequence number for each existing referenced chain. + sql = """ + SELECT chain_id, MAX(sequence_number) FROM event_auth_chains + WHERE %s + GROUP BY chain_id + """ + clause, args = make_in_list_sql_clause( + db_pool.engine, "chain_id", existing_chains + ) + txn.execute(sql % (clause,), args) + + chain_to_max_seq_no = {row[0]: row[1] for row in txn} # type: Dict[Any, int] + + # Allocate the new events chain ID/sequence numbers. + # + # To reduce the number of calls to the database we don't allocate a + # chain ID number in the loop, instead we use a temporary `object()` for + # each new chain ID. Once we've done the loop we generate the necessary + # number of new chain IDs in one call, replacing all temporary + # objects with real allocated chain IDs. + + unallocated_chain_ids = set() # type: Set[object] + new_chain_tuples = {} # type: Dict[str, Tuple[Any, int]] + for event_id, auth_event_id in tree: + # If we reference an auth_event_id we fetch the allocated chain ID, + # either from the existing `chain_map` or the newly generated + # `new_chain_tuples` map. + existing_chain_id = None + if auth_event_id: + existing_chain_id = new_chain_tuples.get(auth_event_id) + if not existing_chain_id: + existing_chain_id = chain_map[auth_event_id] + + new_chain_tuple = None # type: Optional[Tuple[Any, int]] + if existing_chain_id: + # We found a chain ID/sequence number candidate, check its + # not already taken. + proposed_new_id = existing_chain_id[0] + proposed_new_seq = existing_chain_id[1] + 1 + + if chain_to_max_seq_no[proposed_new_id] < proposed_new_seq: + new_chain_tuple = ( + proposed_new_id, + proposed_new_seq, + ) + + # If we need to start a new chain we allocate a temporary chain ID. + if not new_chain_tuple: + new_chain_tuple = (object(), 1) + unallocated_chain_ids.add(new_chain_tuple[0]) + + new_chain_tuples[event_id] = new_chain_tuple + chain_to_max_seq_no[new_chain_tuple[0]] = new_chain_tuple[1] + + # Generate new chain IDs for all unallocated chain IDs. + newly_allocated_chain_ids = db_pool.event_chain_id_gen.get_next_mult_txn( + txn, len(unallocated_chain_ids) + ) + + # Map from potentially temporary chain ID to real chain ID + chain_id_to_allocated_map = dict( + zip(unallocated_chain_ids, newly_allocated_chain_ids) + ) # type: Dict[Any, int] + chain_id_to_allocated_map.update((c, c) for c in existing_chains) + + return { + event_id: (chain_id_to_allocated_map[chain_id], seq) + for event_id, (chain_id, seq) in new_chain_tuples.items() + } def _persist_transaction_ids_txn( self, @@ -489,7 +965,7 @@ class PersistEventsStore: WHERE room_id = ? AND type = ? AND state_key = ? ) """ - txn.executemany( + txn.execute_batch( sql, ( ( @@ -508,7 +984,7 @@ class PersistEventsStore: ) # Now we actually update the current_state_events table - txn.executemany( + txn.execute_batch( "DELETE FROM current_state_events" " WHERE room_id = ? AND type = ? AND state_key = ?", ( @@ -520,7 +996,7 @@ class PersistEventsStore: # We include the membership in the current state table, hence we do # a lookup when we insert. This assumes that all events have already # been inserted into room_memberships. - txn.executemany( + txn.execute_batch( """INSERT INTO current_state_events (room_id, type, state_key, event_id, membership) VALUES (?, ?, ?, ?, (SELECT membership FROM room_memberships WHERE event_id = ?)) @@ -540,7 +1016,7 @@ class PersistEventsStore: # we have no record of the fact the user *was* a member of the # room but got, say, state reset out of it. if to_delete or to_insert: - txn.executemany( + txn.execute_batch( "DELETE FROM local_current_membership" " WHERE room_id = ? AND user_id = ?", ( @@ -551,7 +1027,7 @@ class PersistEventsStore: ) if to_insert: - txn.executemany( + txn.execute_batch( """INSERT INTO local_current_membership (room_id, user_id, event_id, membership) VALUES (?, ?, ?, (SELECT membership FROM room_memberships WHERE event_id = ?)) @@ -799,7 +1275,8 @@ class PersistEventsStore: return [ec for ec in events_and_contexts if ec[0] not in to_remove] def _store_event_txn(self, txn, events_and_contexts): - """Insert new events into the event and event_json tables + """Insert new events into the event, event_json, redaction and + state_events tables. Args: txn (twisted.enterprise.adbapi.Connection): db connection @@ -871,6 +1348,29 @@ class PersistEventsStore: updatevalues={"have_censored": False}, ) + state_events_and_contexts = [ + ec for ec in events_and_contexts if ec[0].is_state() + ] + + state_values = [] + for event, context in state_events_and_contexts: + 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 + + state_values.append(vals) + + self.db_pool.simple_insert_many_txn( + txn, table="state_events", values=state_values + ) + def _store_rejected_events_txn(self, txn, events_and_contexts): """Add rows to the 'rejections' table for received events which were rejected @@ -987,29 +1487,6 @@ class PersistEventsStore: txn, [event for event, _ in events_and_contexts] ) - state_events_and_contexts = [ - ec for ec in events_and_contexts if ec[0].is_state() - ] - - state_values = [] - for event, context in state_events_and_contexts: - 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 - - state_values.append(vals) - - self.db_pool.simple_insert_many_txn( - txn, table="state_events", values=state_values - ) - # Prefill the event cache self._add_to_cache(txn, events_and_contexts) @@ -1350,7 +1827,7 @@ class PersistEventsStore: """ if events_and_contexts: - txn.executemany( + txn.execute_batch( sql, ( ( @@ -1379,7 +1856,7 @@ class PersistEventsStore: # Now we delete the staging area for *all* events that were being # persisted. - txn.executemany( + txn.execute_batch( "DELETE FROM event_push_actions_staging WHERE event_id = ?", ((event.event_id,) for event, _ in all_events_and_contexts), ) @@ -1498,7 +1975,7 @@ class PersistEventsStore: " )" ) - txn.executemany( + txn.execute_batch( query, [ (e_id, ev.room_id, e_id, ev.room_id, e_id, ev.room_id, False) @@ -1512,7 +1989,7 @@ class PersistEventsStore: "DELETE FROM event_backward_extremities" " WHERE event_id = ? AND room_id = ?" ) - txn.executemany( + txn.execute_batch( query, [ (ev.event_id, ev.room_id) @@ -1520,3 +1997,131 @@ class PersistEventsStore: if not ev.internal_metadata.is_outlier() ], ) + + +@attr.s(slots=True) +class _LinkMap: + """A helper type for tracking links between chains. + """ + + # Stores the set of links as nested maps: source chain ID -> target chain ID + # -> source sequence number -> target sequence number. + maps = attr.ib(type=Dict[int, Dict[int, Dict[int, int]]], factory=dict) + + # Stores the links that have been added (with new set to true), as tuples of + # `(source chain ID, source sequence no, target chain ID, target sequence no.)` + additions = attr.ib(type=Set[Tuple[int, int, int, int]], factory=set) + + def add_link( + self, + src_tuple: Tuple[int, int], + target_tuple: Tuple[int, int], + new: bool = True, + ) -> bool: + """Add a new link between two chains, ensuring no redundant links are added. + + New links should be added in topological order. + + Args: + src_tuple: The chain ID/sequence number of the source of the link. + target_tuple: The chain ID/sequence number of the target of the link. + new: Whether this is a "new" link, i.e. should it be returned + by `get_additions`. + + Returns: + True if a link was added, false if the given link was dropped as redundant + """ + src_chain, src_seq = src_tuple + target_chain, target_seq = target_tuple + + current_links = self.maps.setdefault(src_chain, {}).setdefault(target_chain, {}) + + assert src_chain != target_chain + + if new: + # Check if the new link is redundant + for current_seq_src, current_seq_target in current_links.items(): + # If a link "crosses" another link then its redundant. For example + # in the following link 1 (L1) is redundant, as any event reachable + # via L1 is *also* reachable via L2. + # + # Chain A Chain B + # | | + # L1 |------ | + # | | | + # L2 |---- | -->| + # | | | + # | |--->| + # | | + # | | + # + # So we only need to keep links which *do not* cross, i.e. links + # that both start and end above or below an existing link. + # + # Note, since we add links in topological ordering we should never + # see `src_seq` less than `current_seq_src`. + + if current_seq_src <= src_seq and target_seq <= current_seq_target: + # This new link is redundant, nothing to do. + return False + + self.additions.add((src_chain, src_seq, target_chain, target_seq)) + + current_links[src_seq] = target_seq + return True + + def get_links_from( + self, src_tuple: Tuple[int, int] + ) -> Generator[Tuple[int, int], None, None]: + """Gets the chains reachable from the given chain/sequence number. + + Yields: + The chain ID and sequence number the link points to. + """ + src_chain, src_seq = src_tuple + for target_id, sequence_numbers in self.maps.get(src_chain, {}).items(): + for link_src_seq, target_seq in sequence_numbers.items(): + if link_src_seq <= src_seq: + yield target_id, target_seq + + def get_links_between( + self, source_chain: int, target_chain: int + ) -> Generator[Tuple[int, int], None, None]: + """Gets the links between two chains. + + Yields: + The source and target sequence numbers. + """ + + yield from self.maps.get(source_chain, {}).get(target_chain, {}).items() + + def get_additions(self) -> Generator[Tuple[int, int, int, int], None, None]: + """Gets any newly added links. + + Yields: + The source chain ID/sequence number and target chain ID/sequence number + """ + + for src_chain, src_seq, target_chain, _ in self.additions: + target_seq = self.maps.get(src_chain, {}).get(target_chain, {}).get(src_seq) + if target_seq is not None: + yield (src_chain, src_seq, target_chain, target_seq) + + def exists_path_from( + self, src_tuple: Tuple[int, int], target_tuple: Tuple[int, int], + ) -> bool: + """Checks if there is a path between the source chain ID/sequence and + target chain ID/sequence. + """ + src_chain, src_seq = src_tuple + target_chain, target_seq = target_tuple + + if src_chain == target_chain: + return target_seq <= src_seq + + links = self.get_links_between(src_chain, target_chain) + for link_start_seq, link_end_seq in links: + if link_start_seq <= src_seq and target_seq <= link_end_seq: + return True + + return False diff --git a/synapse/storage/databases/main/events_bg_updates.py b/synapse/storage/databases/main/events_bg_updates.py index 97b6754846..5ca4fa6817 100644 --- a/synapse/storage/databases/main/events_bg_updates.py +++ b/synapse/storage/databases/main/events_bg_updates.py @@ -14,14 +14,41 @@ # limitations under the License. import logging +from typing import Dict, List, Optional, Tuple + +import attr from synapse.api.constants import EventContentFields +from synapse.api.room_versions import KNOWN_ROOM_VERSIONS +from synapse.events import make_event_from_dict from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause -from synapse.storage.database import DatabasePool +from synapse.storage.database import DatabasePool, make_tuple_comparison_clause +from synapse.storage.databases.main.events import PersistEventsStore +from synapse.storage.types import Cursor +from synapse.types import JsonDict logger = logging.getLogger(__name__) +@attr.s(slots=True, frozen=True) +class _CalculateChainCover: + """Return value for _calculate_chain_cover_txn. + """ + + # The last room_id/depth/stream processed. + room_id = attr.ib(type=str) + depth = attr.ib(type=int) + stream = attr.ib(type=int) + + # Number of rows processed + processed_count = attr.ib(type=int) + + # Map from room_id to last depth/stream processed for each room that we have + # processed all events for (i.e. the rooms we can flip the + # `has_auth_chain_index` for) + finished_room_map = attr.ib(type=Dict[str, Tuple[int, int]]) + + class EventsBackgroundUpdatesStore(SQLBaseStore): EVENT_ORIGIN_SERVER_TS_NAME = "event_origin_server_ts" @@ -99,13 +126,19 @@ class EventsBackgroundUpdatesStore(SQLBaseStore): columns=["user_id", "created_ts"], ) + self.db_pool.updates.register_background_update_handler( + "rejected_events_metadata", self._rejected_events_metadata, + ) + + self.db_pool.updates.register_background_update_handler( + "chain_cover", self._chain_cover_index, + ) + async def _background_reindex_fields_sender(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_txn(txn): sql = ( "SELECT stream_ordering, event_id, json FROM events" @@ -143,9 +176,7 @@ class EventsBackgroundUpdatesStore(SQLBaseStore): sql = "UPDATE events SET sender = ?, contains_url = ? WHERE event_id = ?" - for index in range(0, len(update_rows), INSERT_CLUMP_SIZE): - clump = update_rows[index : index + INSERT_CLUMP_SIZE] - txn.executemany(sql, clump) + txn.execute_batch(sql, update_rows) progress = { "target_min_stream_id_inclusive": target_min_stream_id, @@ -175,8 +206,6 @@ class EventsBackgroundUpdatesStore(SQLBaseStore): 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" @@ -221,9 +250,7 @@ class EventsBackgroundUpdatesStore(SQLBaseStore): sql = "UPDATE events SET origin_server_ts = ? WHERE event_id = ?" - for index in range(0, len(rows_to_update), INSERT_CLUMP_SIZE): - clump = rows_to_update[index : index + INSERT_CLUMP_SIZE] - txn.executemany(sql, clump) + txn.execute_batch(sql, rows_to_update) progress = { "target_min_stream_id_inclusive": target_min_stream_id, @@ -582,3 +609,314 @@ class EventsBackgroundUpdatesStore(SQLBaseStore): await self.db_pool.updates._end_background_update("event_store_labels") return num_rows + + async def _rejected_events_metadata(self, progress: dict, batch_size: int) -> int: + """Adds rejected events to the `state_events` and `event_auth` metadata + tables. + """ + + last_event_id = progress.get("last_event_id", "") + + def get_rejected_events( + txn: Cursor, + ) -> List[Tuple[str, str, JsonDict, bool, bool]]: + # Fetch rejected event json, their room version and whether we have + # inserted them into the state_events or auth_events tables. + # + # Note we can assume that events that don't have a corresponding + # room version are V1 rooms. + sql = """ + SELECT DISTINCT + event_id, + COALESCE(room_version, '1'), + json, + state_events.event_id IS NOT NULL, + event_auth.event_id IS NOT NULL + FROM rejections + INNER JOIN event_json USING (event_id) + LEFT JOIN rooms USING (room_id) + LEFT JOIN state_events USING (event_id) + LEFT JOIN event_auth USING (event_id) + WHERE event_id > ? + ORDER BY event_id + LIMIT ? + """ + + txn.execute(sql, (last_event_id, batch_size,)) + + return [(row[0], row[1], db_to_json(row[2]), row[3], row[4]) for row in txn] # type: ignore + + results = await self.db_pool.runInteraction( + desc="_rejected_events_metadata_get", func=get_rejected_events + ) + + if not results: + await self.db_pool.updates._end_background_update( + "rejected_events_metadata" + ) + return 0 + + state_events = [] + auth_events = [] + for event_id, room_version, event_json, has_state, has_event_auth in results: + last_event_id = event_id + + if has_state and has_event_auth: + continue + + room_version_obj = KNOWN_ROOM_VERSIONS.get(room_version) + if not room_version_obj: + # We no longer support this room version, so we just ignore the + # events entirely. + logger.info( + "Ignoring event with unknown room version %r: %r", + room_version, + event_id, + ) + continue + + event = make_event_from_dict(event_json, room_version_obj) + + if not event.is_state(): + continue + + if not has_state: + state_events.append( + { + "event_id": event.event_id, + "room_id": event.room_id, + "type": event.type, + "state_key": event.state_key, + } + ) + + if not has_event_auth: + for auth_id in event.auth_event_ids(): + auth_events.append( + { + "room_id": event.room_id, + "event_id": event.event_id, + "auth_id": auth_id, + } + ) + + if state_events: + await self.db_pool.simple_insert_many( + table="state_events", + values=state_events, + desc="_rejected_events_metadata_state_events", + ) + + if auth_events: + await self.db_pool.simple_insert_many( + table="event_auth", + values=auth_events, + desc="_rejected_events_metadata_event_auth", + ) + + await self.db_pool.updates._background_update_progress( + "rejected_events_metadata", {"last_event_id": last_event_id} + ) + + if len(results) < batch_size: + await self.db_pool.updates._end_background_update( + "rejected_events_metadata" + ) + + return len(results) + + async def _chain_cover_index(self, progress: dict, batch_size: int) -> int: + """A background updates that iterates over all rooms and generates the + chain cover index for them. + """ + + current_room_id = progress.get("current_room_id", "") + + # Where we've processed up to in the room, defaults to the start of the + # room. + last_depth = progress.get("last_depth", -1) + last_stream = progress.get("last_stream", -1) + + result = await self.db_pool.runInteraction( + "_chain_cover_index", + self._calculate_chain_cover_txn, + current_room_id, + last_depth, + last_stream, + batch_size, + single_room=False, + ) + + finished = result.processed_count == 0 + + total_rows_processed = result.processed_count + current_room_id = result.room_id + last_depth = result.depth + last_stream = result.stream + + for room_id, (depth, stream) in result.finished_room_map.items(): + # If we've done all the events in the room we flip the + # `has_auth_chain_index` in the DB. Note that its possible for + # further events to be persisted between the above and setting the + # flag without having the chain cover calculated for them. This is + # fine as a) the code gracefully handles these cases and b) we'll + # calculate them below. + + await self.db_pool.simple_update( + table="rooms", + keyvalues={"room_id": room_id}, + updatevalues={"has_auth_chain_index": True}, + desc="_chain_cover_index", + ) + + # Handle any events that might have raced with us flipping the + # bit above. + result = await self.db_pool.runInteraction( + "_chain_cover_index", + self._calculate_chain_cover_txn, + room_id, + depth, + stream, + batch_size=None, + single_room=True, + ) + + total_rows_processed += result.processed_count + + if finished: + await self.db_pool.updates._end_background_update("chain_cover") + return total_rows_processed + + await self.db_pool.updates._background_update_progress( + "chain_cover", + { + "current_room_id": current_room_id, + "last_depth": last_depth, + "last_stream": last_stream, + }, + ) + + return total_rows_processed + + def _calculate_chain_cover_txn( + self, + txn: Cursor, + last_room_id: str, + last_depth: int, + last_stream: int, + batch_size: Optional[int], + single_room: bool, + ) -> _CalculateChainCover: + """Calculate the chain cover for `batch_size` events, ordered by + `(room_id, depth, stream)`. + + Args: + txn, + last_room_id, last_depth, last_stream: The `(room_id, depth, stream)` + tuple to fetch results after. + batch_size: The maximum number of events to process. If None then + no limit. + single_room: Whether to calculate the index for just the given + room. + """ + + # Get the next set of events in the room (that we haven't already + # computed chain cover for). We do this in topological order. + + # We want to do a `(topological_ordering, stream_ordering) > (?,?)` + # comparison, but that is not supported on older SQLite versions + tuple_clause, tuple_args = make_tuple_comparison_clause( + self.database_engine, + [ + ("events.room_id", last_room_id), + ("topological_ordering", last_depth), + ("stream_ordering", last_stream), + ], + ) + + extra_clause = "" + if single_room: + extra_clause = "AND events.room_id = ?" + tuple_args.append(last_room_id) + + sql = """ + SELECT + event_id, state_events.type, state_events.state_key, + topological_ordering, stream_ordering, + events.room_id + FROM events + INNER JOIN state_events USING (event_id) + LEFT JOIN event_auth_chains USING (event_id) + LEFT JOIN event_auth_chain_to_calculate USING (event_id) + WHERE event_auth_chains.event_id IS NULL + AND event_auth_chain_to_calculate.event_id IS NULL + AND %(tuple_cmp)s + %(extra)s + ORDER BY events.room_id, topological_ordering, stream_ordering + %(limit)s + """ % { + "tuple_cmp": tuple_clause, + "limit": "LIMIT ?" if batch_size is not None else "", + "extra": extra_clause, + } + + if batch_size is not None: + tuple_args.append(batch_size) + + txn.execute(sql, tuple_args) + rows = txn.fetchall() + + # Put the results in the necessary format for + # `_add_chain_cover_index` + event_to_room_id = {row[0]: row[5] for row in rows} + event_to_types = {row[0]: (row[1], row[2]) for row in rows} + + # Calculate the new last position we've processed up to. + new_last_depth = rows[-1][3] if rows else last_depth # type: int + new_last_stream = rows[-1][4] if rows else last_stream # type: int + new_last_room_id = rows[-1][5] if rows else "" # type: str + + # Map from room_id to last depth/stream_ordering processed for the room, + # excluding the last room (which we're likely still processing). We also + # need to include the room passed in if it's not included in the result + # set (as we then know we've processed all events in said room). + # + # This is the set of rooms that we can now safely flip the + # `has_auth_chain_index` bit for. + finished_rooms = { + row[5]: (row[3], row[4]) for row in rows if row[5] != new_last_room_id + } + if last_room_id not in finished_rooms and last_room_id != new_last_room_id: + finished_rooms[last_room_id] = (last_depth, last_stream) + + count = len(rows) + + # We also need to fetch the auth events for them. + auth_events = self.db_pool.simple_select_many_txn( + txn, + table="event_auth", + column="event_id", + iterable=event_to_room_id, + keyvalues={}, + retcols=("event_id", "auth_id"), + ) + + event_to_auth_chain = {} # type: Dict[str, List[str]] + for row in auth_events: + event_to_auth_chain.setdefault(row["event_id"], []).append(row["auth_id"]) + + # Calculate and persist the chain cover index for this set of events. + # + # Annoyingly we need to gut wrench into the persit event store so that + # we can reuse the function to calculate the chain cover for rooms. + PersistEventsStore._add_chain_cover_index( + txn, self.db_pool, event_to_room_id, event_to_types, event_to_auth_chain, + ) + + return _CalculateChainCover( + room_id=new_last_room_id, + depth=new_last_depth, + stream=new_last_stream, + processed_count=count, + finished_room_map=finished_rooms, + ) diff --git a/synapse/storage/databases/main/events_worker.py b/synapse/storage/databases/main/events_worker.py index 4732685f6e..71d823be72 100644 --- a/synapse/storage/databases/main/events_worker.py +++ b/synapse/storage/databases/main/events_worker.py @@ -96,9 +96,7 @@ class EventsWorkerStore(SQLBaseStore): db=database, stream_name="events", instance_name=hs.get_instance_name(), - table="events", - instance_column="instance_name", - id_column="stream_ordering", + tables=[("events", "instance_name", "stream_ordering")], sequence_name="events_stream_seq", writers=hs.config.worker.writers.events, ) @@ -107,9 +105,7 @@ class EventsWorkerStore(SQLBaseStore): db=database, stream_name="backfill", instance_name=hs.get_instance_name(), - table="events", - instance_column="instance_name", - id_column="stream_ordering", + tables=[("events", "instance_name", "stream_ordering")], sequence_name="events_backfill_stream_seq", positive=False, writers=hs.config.worker.writers.events, diff --git a/synapse/storage/databases/main/media_repository.py b/synapse/storage/databases/main/media_repository.py index 4b2f224718..e017177655 100644 --- a/synapse/storage/databases/main/media_repository.py +++ b/synapse/storage/databases/main/media_repository.py @@ -1,5 +1,6 @@ # -*- coding: utf-8 -*- # Copyright 2014-2016 OpenMarket Ltd +# Copyright 2020-2021 The Matrix.org Foundation C.I.C. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. @@ -169,7 +170,7 @@ class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore): async def get_local_media_before( self, before_ts: int, size_gt: int, keep_profiles: bool, - ) -> Optional[List[str]]: + ) -> List[str]: # to find files that have never been accessed (last_access_ts IS NULL) # compare with `created_ts` @@ -416,7 +417,7 @@ class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore): " WHERE media_origin = ? AND media_id = ?" ) - txn.executemany( + txn.execute_batch( sql, ( (time_ms, media_origin, media_id) @@ -429,7 +430,7 @@ class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore): " WHERE media_id = ?" ) - txn.executemany(sql, ((time_ms, media_id) for media_id in local_media)) + txn.execute_batch(sql, ((time_ms, media_id) for media_id in local_media)) return await self.db_pool.runInteraction( "update_cached_last_access_time", update_cache_txn @@ -556,7 +557,7 @@ class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore): sql = "DELETE FROM local_media_repository_url_cache WHERE media_id = ?" def _delete_url_cache_txn(txn): - txn.executemany(sql, [(media_id,) for media_id in media_ids]) + txn.execute_batch(sql, [(media_id,) for media_id in media_ids]) return await self.db_pool.runInteraction( "delete_url_cache", _delete_url_cache_txn @@ -585,11 +586,11 @@ class MediaRepositoryStore(MediaRepositoryBackgroundUpdateStore): def _delete_url_cache_media_txn(txn): sql = "DELETE FROM local_media_repository WHERE media_id = ?" - txn.executemany(sql, [(media_id,) for media_id in media_ids]) + txn.execute_batch(sql, [(media_id,) for media_id in media_ids]) sql = "DELETE FROM local_media_repository_thumbnails WHERE media_id = ?" - txn.executemany(sql, [(media_id,) for media_id in media_ids]) + txn.execute_batch(sql, [(media_id,) for media_id in media_ids]) return await self.db_pool.runInteraction( "delete_url_cache_media", _delete_url_cache_media_txn diff --git a/synapse/storage/databases/main/profile.py b/synapse/storage/databases/main/profile.py index 0e25ca3d7a..54ef0f1f54 100644 --- a/synapse/storage/databases/main/profile.py +++ b/synapse/storage/databases/main/profile.py @@ -82,7 +82,7 @@ class ProfileWorkerStore(SQLBaseStore): ) async def set_profile_avatar_url( - self, user_localpart: str, new_avatar_url: str + self, user_localpart: str, new_avatar_url: Optional[str] ) -> None: await self.db_pool.simple_update_one( table="profiles", diff --git a/synapse/storage/databases/main/purge_events.py b/synapse/storage/databases/main/purge_events.py index 5d668aadb2..ecfc9f20b1 100644 --- a/synapse/storage/databases/main/purge_events.py +++ b/synapse/storage/databases/main/purge_events.py @@ -172,7 +172,7 @@ class PurgeEventsStore(StateGroupWorkerStore, SQLBaseStore): ) # Update backward extremeties - txn.executemany( + txn.execute_batch( "INSERT INTO event_backward_extremities (room_id, event_id)" " VALUES (?, ?)", [(room_id, event_id) for event_id, in new_backwards_extrems], diff --git a/synapse/storage/databases/main/pusher.py b/synapse/storage/databases/main/pusher.py index 77ba9d819e..bc7621b8d6 100644 --- a/synapse/storage/databases/main/pusher.py +++ b/synapse/storage/databases/main/pusher.py @@ -17,14 +17,13 @@ import logging from typing import TYPE_CHECKING, Any, Dict, Iterable, Iterator, List, Optional, Tuple -from canonicaljson import encode_canonical_json - from synapse.push import PusherConfig, ThrottleParams from synapse.storage._base import SQLBaseStore, db_to_json from synapse.storage.database import DatabasePool from synapse.storage.types import Connection from synapse.storage.util.id_generators import StreamIdGenerator from synapse.types import JsonDict +from synapse.util import json_encoder from synapse.util.caches.descriptors import cached, cachedList if TYPE_CHECKING: @@ -315,7 +314,7 @@ class PusherStore(PusherWorkerStore): "device_display_name": device_display_name, "ts": pushkey_ts, "lang": lang, - "data": bytearray(encode_canonical_json(data)), + "data": json_encoder.encode(data), "last_stream_ordering": last_stream_ordering, "profile_tag": profile_tag, "id": stream_id, diff --git a/synapse/storage/databases/main/receipts.py b/synapse/storage/databases/main/receipts.py index 1e7949a323..e4843a202c 100644 --- a/synapse/storage/databases/main/receipts.py +++ b/synapse/storage/databases/main/receipts.py @@ -14,15 +14,17 @@ # See the License for the specific language governing permissions and # limitations under the License. -import abc import logging from typing import Any, Dict, List, Optional, Tuple from twisted.internet import defer +from synapse.replication.slave.storage._slaved_id_tracker import SlavedIdTracker +from synapse.replication.tcp.streams import ReceiptsStream from synapse.storage._base import SQLBaseStore, db_to_json, make_in_list_sql_clause from synapse.storage.database import DatabasePool -from synapse.storage.util.id_generators import StreamIdGenerator +from synapse.storage.engines import PostgresEngine +from synapse.storage.util.id_generators import MultiWriterIdGenerator, StreamIdGenerator from synapse.types import JsonDict from synapse.util import json_encoder from synapse.util.caches.descriptors import cached, cachedList @@ -31,28 +33,56 @@ from synapse.util.caches.stream_change_cache import StreamChangeCache logger = logging.getLogger(__name__) -# The ABCMeta metaclass ensures that it cannot be instantiated without -# the abstract methods being implemented. -class ReceiptsWorkerStore(SQLBaseStore, metaclass=abc.ABCMeta): - """This is an abstract base class where subclasses must implement - `get_max_receipt_stream_id` which can be called in the initializer. - """ - +class ReceiptsWorkerStore(SQLBaseStore): def __init__(self, database: DatabasePool, db_conn, hs): + self._instance_name = hs.get_instance_name() + + if isinstance(database.engine, PostgresEngine): + self._can_write_to_receipts = ( + self._instance_name in hs.config.worker.writers.receipts + ) + + self._receipts_id_gen = MultiWriterIdGenerator( + db_conn=db_conn, + db=database, + stream_name="receipts", + instance_name=self._instance_name, + tables=[("receipts_linearized", "instance_name", "stream_id")], + sequence_name="receipts_sequence", + writers=hs.config.worker.writers.receipts, + ) + else: + self._can_write_to_receipts = True + + # We shouldn't be running in worker mode with SQLite, but its useful + # to support it for unit tests. + # + # If this process is the writer than we need to use + # `StreamIdGenerator`, otherwise we use `SlavedIdTracker` which gets + # updated over replication. (Multiple writers are not supported for + # SQLite). + if hs.get_instance_name() in hs.config.worker.writers.receipts: + self._receipts_id_gen = StreamIdGenerator( + db_conn, "receipts_linearized", "stream_id" + ) + else: + self._receipts_id_gen = SlavedIdTracker( + db_conn, "receipts_linearized", "stream_id" + ) + super().__init__(database, db_conn, hs) self._receipts_stream_cache = StreamChangeCache( "ReceiptsRoomChangeCache", self.get_max_receipt_stream_id() ) - @abc.abstractmethod def get_max_receipt_stream_id(self): """Get the current max stream ID for receipts stream Returns: int """ - raise NotImplementedError() + return self._receipts_id_gen.get_current_token() @cached() async def get_users_with_read_receipts_in_room(self, room_id): @@ -428,19 +458,25 @@ class ReceiptsWorkerStore(SQLBaseStore, metaclass=abc.ABCMeta): self.get_users_with_read_receipts_in_room.invalidate((room_id,)) - -class ReceiptsStore(ReceiptsWorkerStore): - def __init__(self, database: DatabasePool, db_conn, hs): - # We instantiate this first as the ReceiptsWorkerStore constructor - # needs to be able to call get_max_receipt_stream_id - self._receipts_id_gen = StreamIdGenerator( - db_conn, "receipts_linearized", "stream_id" + def invalidate_caches_for_receipt(self, room_id, receipt_type, user_id): + self.get_receipts_for_user.invalidate((user_id, receipt_type)) + self._get_linearized_receipts_for_room.invalidate_many((room_id,)) + self.get_last_receipt_event_id_for_user.invalidate( + (user_id, room_id, receipt_type) ) + self._invalidate_get_users_with_receipts_in_room(room_id, receipt_type, user_id) + self.get_receipts_for_room.invalidate((room_id, receipt_type)) + + def process_replication_rows(self, stream_name, instance_name, token, rows): + if stream_name == ReceiptsStream.NAME: + self._receipts_id_gen.advance(instance_name, token) + for row in rows: + self.invalidate_caches_for_receipt( + row.room_id, row.receipt_type, row.user_id + ) + self._receipts_stream_cache.entity_has_changed(row.room_id, token) - super().__init__(database, db_conn, hs) - - def get_max_receipt_stream_id(self): - return self._receipts_id_gen.get_current_token() + return super().process_replication_rows(stream_name, instance_name, token, rows) def insert_linearized_receipt_txn( self, txn, room_id, receipt_type, user_id, event_id, data, stream_id @@ -452,6 +488,8 @@ class ReceiptsStore(ReceiptsWorkerStore): otherwise, the rx timestamp of the event that the RR corresponds to (or 0 if the event is unknown) """ + assert self._can_write_to_receipts + res = self.db_pool.simple_select_one_txn( txn, table="events", @@ -483,28 +521,14 @@ class ReceiptsStore(ReceiptsWorkerStore): ) return None - txn.call_after(self.get_receipts_for_room.invalidate, (room_id, receipt_type)) - txn.call_after( - self._invalidate_get_users_with_receipts_in_room, - room_id, - receipt_type, - user_id, - ) - txn.call_after(self.get_receipts_for_user.invalidate, (user_id, receipt_type)) - # FIXME: This shouldn't invalidate the whole cache txn.call_after( - self._get_linearized_receipts_for_room.invalidate_many, (room_id,) + self.invalidate_caches_for_receipt, room_id, receipt_type, user_id ) txn.call_after( self._receipts_stream_cache.entity_has_changed, room_id, stream_id ) - txn.call_after( - self.get_last_receipt_event_id_for_user.invalidate, - (user_id, room_id, receipt_type), - ) - self.db_pool.simple_upsert_txn( txn, table="receipts_linearized", @@ -543,6 +567,8 @@ class ReceiptsStore(ReceiptsWorkerStore): Automatically does conversion between linearized and graph representations. """ + assert self._can_write_to_receipts + if not event_ids: return None @@ -607,6 +633,8 @@ class ReceiptsStore(ReceiptsWorkerStore): async def insert_graph_receipt( self, room_id, receipt_type, user_id, event_ids, data ): + assert self._can_write_to_receipts + return await self.db_pool.runInteraction( "insert_graph_receipt", self.insert_graph_receipt_txn, @@ -620,6 +648,8 @@ class ReceiptsStore(ReceiptsWorkerStore): def insert_graph_receipt_txn( self, txn, room_id, receipt_type, user_id, event_ids, data ): + assert self._can_write_to_receipts + txn.call_after(self.get_receipts_for_room.invalidate, (room_id, receipt_type)) txn.call_after( self._invalidate_get_users_with_receipts_in_room, @@ -653,3 +683,7 @@ class ReceiptsStore(ReceiptsWorkerStore): "data": json_encoder.encode(data), }, ) + + +class ReceiptsStore(ReceiptsWorkerStore): + pass diff --git a/synapse/storage/databases/main/registration.py b/synapse/storage/databases/main/registration.py index 8d05288ed4..585b4049d6 100644 --- a/synapse/storage/databases/main/registration.py +++ b/synapse/storage/databases/main/registration.py @@ -1104,7 +1104,7 @@ class RegistrationBackgroundUpdateStore(RegistrationWorkerStore): FROM user_threepids """ - txn.executemany(sql, [(id_server,) for id_server in id_servers]) + txn.execute_batch(sql, [(id_server,) for id_server in id_servers]) if id_servers: await self.db_pool.runInteraction( diff --git a/synapse/storage/databases/main/room.py b/synapse/storage/databases/main/room.py index 4650d0689b..a9fcb5f59c 100644 --- a/synapse/storage/databases/main/room.py +++ b/synapse/storage/databases/main/room.py @@ -16,7 +16,6 @@ import collections import logging -import re from abc import abstractmethod from enum import Enum from typing import Any, Dict, List, Optional, Tuple @@ -30,6 +29,7 @@ from synapse.storage.databases.main.search import SearchStore from synapse.types import JsonDict, ThirdPartyInstanceID from synapse.util import json_encoder from synapse.util.caches.descriptors import cached +from synapse.util.stringutils import MXC_REGEX logger = logging.getLogger(__name__) @@ -84,7 +84,7 @@ class RoomWorkerStore(SQLBaseStore): return await self.db_pool.simple_select_one( table="rooms", keyvalues={"room_id": room_id}, - retcols=("room_id", "is_public", "creator"), + retcols=("room_id", "is_public", "creator", "has_auth_chain_index"), desc="get_room", allow_none=True, ) @@ -660,8 +660,6 @@ class RoomWorkerStore(SQLBaseStore): The local and remote media as a lists of tuples where the key is the hostname and the value is the media ID. """ - mxc_re = re.compile("^mxc://([^/]+)/([^/#?]+)") - sql = """ SELECT stream_ordering, json FROM events JOIN event_json USING (room_id, event_id) @@ -688,7 +686,7 @@ class RoomWorkerStore(SQLBaseStore): for url in (content_url, thumbnail_url): if not url: continue - matches = mxc_re.match(url) + matches = MXC_REGEX.match(url) if matches: hostname = matches.group(1) media_id = matches.group(2) @@ -1166,6 +1164,37 @@ class RoomBackgroundUpdateStore(SQLBaseStore): # It's overridden by RoomStore for the synapse master. raise NotImplementedError() + async def has_auth_chain_index(self, room_id: str) -> bool: + """Check if the room has (or can have) a chain cover index. + + Defaults to True if we don't have an entry in `rooms` table nor any + events for the room. + """ + + has_auth_chain_index = await self.db_pool.simple_select_one_onecol( + table="rooms", + keyvalues={"room_id": room_id}, + retcol="has_auth_chain_index", + desc="has_auth_chain_index", + allow_none=True, + ) + + if has_auth_chain_index: + return True + + # It's possible that we already have events for the room in our DB + # without a corresponding room entry. If we do then we don't want to + # mark the room as having an auth chain cover index. + max_ordering = await self.db_pool.simple_select_one_onecol( + table="events", + keyvalues={"room_id": room_id}, + retcol="MAX(stream_ordering)", + allow_none=True, + desc="upsert_room_on_join", + ) + + return max_ordering is None + class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore, SearchStore): def __init__(self, database: DatabasePool, db_conn, hs): @@ -1179,12 +1208,21 @@ class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore, SearchStore): Called when we join a room over federation, and overwrites any room version currently in the table. """ + # It's possible that we already have events for the room in our DB + # without a corresponding room entry. If we do then we don't want to + # mark the room as having an auth chain cover index. + has_auth_chain_index = await self.has_auth_chain_index(room_id) + await self.db_pool.simple_upsert( desc="upsert_room_on_join", table="rooms", keyvalues={"room_id": room_id}, values={"room_version": room_version.identifier}, - insertion_values={"is_public": False, "creator": ""}, + insertion_values={ + "is_public": False, + "creator": "", + "has_auth_chain_index": has_auth_chain_index, + }, # rooms has a unique constraint on room_id, so no need to lock when doing an # emulated upsert. lock=False, @@ -1219,6 +1257,7 @@ class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore, SearchStore): "creator": room_creator_user_id, "is_public": is_public, "room_version": room_version.identifier, + "has_auth_chain_index": True, }, ) if is_public: @@ -1247,6 +1286,11 @@ class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore, SearchStore): When we receive an invite or any other event over federation that may relate to a room we are not in, store the version of the room if we don't already know the room version. """ + # It's possible that we already have events for the room in our DB + # without a corresponding room entry. If we do then we don't want to + # mark the room as having an auth chain cover index. + has_auth_chain_index = await self.has_auth_chain_index(room_id) + await self.db_pool.simple_upsert( desc="maybe_store_room_on_outlier_membership", table="rooms", @@ -1256,6 +1300,7 @@ class RoomStore(RoomBackgroundUpdateStore, RoomWorkerStore, SearchStore): "room_version": room_version.identifier, "is_public": False, "creator": "", + "has_auth_chain_index": has_auth_chain_index, }, # rooms has a unique constraint on room_id, so no need to lock when doing an # emulated upsert. diff --git a/synapse/storage/databases/main/roommember.py b/synapse/storage/databases/main/roommember.py index dcdaf09682..92382bed28 100644 --- a/synapse/storage/databases/main/roommember.py +++ b/synapse/storage/databases/main/roommember.py @@ -873,8 +873,6 @@ class RoomMemberBackgroundUpdateStore(SQLBaseStore): "max_stream_id_exclusive", self._stream_order_on_start + 1 ) - INSERT_CLUMP_SIZE = 1000 - def add_membership_profile_txn(txn): sql = """ SELECT stream_ordering, event_id, events.room_id, event_json.json @@ -915,9 +913,7 @@ class RoomMemberBackgroundUpdateStore(SQLBaseStore): UPDATE room_memberships SET display_name = ?, avatar_url = ? WHERE event_id = ? AND room_id = ? """ - for index in range(0, len(to_update), INSERT_CLUMP_SIZE): - clump = to_update[index : index + INSERT_CLUMP_SIZE] - txn.executemany(to_update_sql, clump) + txn.execute_batch(to_update_sql, to_update) progress = { "target_min_stream_id_inclusive": target_min_stream_id, diff --git a/synapse/storage/databases/main/schema/delta/58/28drop_last_used_column.sql.postgres b/synapse/storage/databases/main/schema/delta/58/28drop_last_used_column.sql.postgres new file mode 100644 index 0000000000..de57645019 --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/58/28drop_last_used_column.sql.postgres @@ -0,0 +1,16 @@ +/* Copyright 2020 The Matrix.org Foundation C.I.C + * + * 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. + */ + +ALTER TABLE access_tokens DROP COLUMN last_used; \ No newline at end of file diff --git a/synapse/storage/databases/main/schema/delta/58/28drop_last_used_column.sql.sqlite b/synapse/storage/databases/main/schema/delta/58/28drop_last_used_column.sql.sqlite new file mode 100644 index 0000000000..ee0e3521bf --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/58/28drop_last_used_column.sql.sqlite @@ -0,0 +1,62 @@ +/* + * Copyright 2020 The Matrix.org Foundation C.I.C. + * + * 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. + */ + + -- Dropping last_used column from access_tokens table. + +CREATE TABLE access_tokens2 ( + id BIGINT PRIMARY KEY, + user_id TEXT NOT NULL, + device_id TEXT, + token TEXT NOT NULL, + valid_until_ms BIGINT, + puppets_user_id TEXT, + last_validated BIGINT, + UNIQUE(token) +); + +INSERT INTO access_tokens2(id, user_id, device_id, token) + SELECT id, user_id, device_id, token FROM access_tokens; + +DROP TABLE access_tokens; +ALTER TABLE access_tokens2 RENAME TO access_tokens; + +CREATE INDEX access_tokens_device_id ON access_tokens (user_id, device_id); + + +-- Re-adding foreign key reference in event_txn_id table + +CREATE TABLE event_txn_id2 ( + event_id TEXT NOT NULL, + room_id TEXT NOT NULL, + user_id TEXT NOT NULL, + token_id BIGINT NOT NULL, + txn_id TEXT NOT NULL, + inserted_ts BIGINT NOT NULL, + FOREIGN KEY (event_id) + REFERENCES events (event_id) ON DELETE CASCADE, + FOREIGN KEY (token_id) + REFERENCES access_tokens (id) ON DELETE CASCADE +); + +INSERT INTO event_txn_id2(event_id, room_id, user_id, token_id, txn_id, inserted_ts) + SELECT event_id, room_id, user_id, token_id, txn_id, inserted_ts FROM event_txn_id; + +DROP TABLE event_txn_id; +ALTER TABLE event_txn_id2 RENAME TO event_txn_id; + +CREATE UNIQUE INDEX IF NOT EXISTS event_txn_id_event_id ON event_txn_id(event_id); +CREATE UNIQUE INDEX IF NOT EXISTS event_txn_id_txn_id ON event_txn_id(room_id, user_id, token_id, txn_id); +CREATE INDEX IF NOT EXISTS event_txn_id_ts ON event_txn_id(inserted_ts); \ No newline at end of file diff --git a/synapse/storage/databases/main/schema/delta/58/28rejected_events_metadata.sql b/synapse/storage/databases/main/schema/delta/58/28rejected_events_metadata.sql new file mode 100644 index 0000000000..9c95646281 --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/58/28rejected_events_metadata.sql @@ -0,0 +1,17 @@ +/* Copyright 2020 The Matrix.org Foundation C.I.C + * + * 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. + */ + +INSERT INTO background_updates (ordering, update_name, progress_json) VALUES + (5828, 'rejected_events_metadata', '{}'); diff --git a/synapse/storage/databases/main/schema/delta/59/01ignored_user.py b/synapse/storage/databases/main/schema/delta/59/01ignored_user.py index f35c70b699..9e8f35c1d2 100644 --- a/synapse/storage/databases/main/schema/delta/59/01ignored_user.py +++ b/synapse/storage/databases/main/schema/delta/59/01ignored_user.py @@ -55,7 +55,7 @@ def run_create(cur: Cursor, database_engine: BaseDatabaseEngine, *args, **kwargs # { "ignored_users": "@someone:example.org": {} } ignored_users = content.get("ignored_users", {}) if isinstance(ignored_users, dict) and ignored_users: - cur.executemany(insert_sql, [(user_id, u) for u in ignored_users]) + cur.execute_batch(insert_sql, [(user_id, u) for u in ignored_users]) # Add indexes after inserting data for efficiency. logger.info("Adding constraints to ignored_users table") diff --git a/synapse/storage/databases/main/schema/delta/59/04_event_auth_chains.sql b/synapse/storage/databases/main/schema/delta/59/04_event_auth_chains.sql new file mode 100644 index 0000000000..729196cfd5 --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/04_event_auth_chains.sql @@ -0,0 +1,52 @@ +/* Copyright 2020 The Matrix.org Foundation C.I.C + * + * 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. + */ + +-- See docs/auth_chain_difference_algorithm.md + +CREATE TABLE event_auth_chains ( + event_id TEXT PRIMARY KEY, + chain_id BIGINT NOT NULL, + sequence_number BIGINT NOT NULL +); + +CREATE UNIQUE INDEX event_auth_chains_c_seq_index ON event_auth_chains (chain_id, sequence_number); + + +CREATE TABLE event_auth_chain_links ( + origin_chain_id BIGINT NOT NULL, + origin_sequence_number BIGINT NOT NULL, + + target_chain_id BIGINT NOT NULL, + target_sequence_number BIGINT NOT NULL +); + + +CREATE INDEX event_auth_chain_links_idx ON event_auth_chain_links (origin_chain_id, target_chain_id); + + +-- Events that we have persisted but not calculated auth chains for, +-- e.g. out of band memberships (where we don't have the auth chain) +CREATE TABLE event_auth_chain_to_calculate ( + event_id TEXT PRIMARY KEY, + room_id TEXT NOT NULL, + type TEXT NOT NULL, + state_key TEXT NOT NULL +); + +CREATE INDEX event_auth_chain_to_calculate_rm_id ON event_auth_chain_to_calculate(room_id); + + +-- Whether we've calculated the above index for a room. +ALTER TABLE rooms ADD COLUMN has_auth_chain_index BOOLEAN; diff --git a/synapse/storage/databases/main/schema/delta/59/04_event_auth_chains.sql.postgres b/synapse/storage/databases/main/schema/delta/59/04_event_auth_chains.sql.postgres new file mode 100644 index 0000000000..e8a035bbeb --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/04_event_auth_chains.sql.postgres @@ -0,0 +1,16 @@ +/* Copyright 2020 The Matrix.org Foundation C.I.C + * + * 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. + */ + +CREATE SEQUENCE IF NOT EXISTS event_auth_chain_id; diff --git a/synapse/storage/databases/main/schema/delta/59/04drop_account_data.sql b/synapse/storage/databases/main/schema/delta/59/04drop_account_data.sql new file mode 100644 index 0000000000..64ab696cfe --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/04drop_account_data.sql @@ -0,0 +1,17 @@ +/* Copyright 2021 The Matrix.org Foundation C.I.C + * + * 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. + */ + +-- This is no longer used and was only kept until we bumped the schema version. +DROP TABLE IF EXISTS account_data_max_stream_id; diff --git a/synapse/storage/databases/main/schema/delta/59/05cache_invalidation.sql b/synapse/storage/databases/main/schema/delta/59/05cache_invalidation.sql new file mode 100644 index 0000000000..fb71b360a0 --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/05cache_invalidation.sql @@ -0,0 +1,17 @@ +/* Copyright 2021 The Matrix.org Foundation C.I.C + * + * 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. + */ + +-- This is no longer used and was only kept until we bumped the schema version. +DROP TABLE IF EXISTS cache_invalidation_stream; diff --git a/synapse/storage/databases/main/schema/delta/59/06chain_cover_index.sql b/synapse/storage/databases/main/schema/delta/59/06chain_cover_index.sql new file mode 100644 index 0000000000..fe3dca71dd --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/06chain_cover_index.sql @@ -0,0 +1,17 @@ +/* Copyright 2020 The Matrix.org Foundation C.I.C + * + * 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. + */ + +INSERT INTO background_updates (ordering, update_name, progress_json, depends_on) VALUES + (5906, 'chain_cover', '{}', 'rejected_events_metadata'); diff --git a/synapse/storage/databases/main/schema/delta/59/06shard_account_data.sql b/synapse/storage/databases/main/schema/delta/59/06shard_account_data.sql new file mode 100644 index 0000000000..46abf8d562 --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/06shard_account_data.sql @@ -0,0 +1,20 @@ +/* Copyright 2021 The Matrix.org Foundation C.I.C + * + * 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. + */ + +ALTER TABLE room_account_data ADD COLUMN instance_name TEXT; +ALTER TABLE room_tags_revisions ADD COLUMN instance_name TEXT; +ALTER TABLE account_data ADD COLUMN instance_name TEXT; + +ALTER TABLE receipts_linearized ADD COLUMN instance_name TEXT; diff --git a/synapse/storage/databases/main/schema/delta/59/06shard_account_data.sql.postgres b/synapse/storage/databases/main/schema/delta/59/06shard_account_data.sql.postgres new file mode 100644 index 0000000000..4a6e6c74f5 --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/06shard_account_data.sql.postgres @@ -0,0 +1,32 @@ +/* Copyright 2021 The Matrix.org Foundation C.I.C + * + * 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. + */ + +CREATE SEQUENCE IF NOT EXISTS account_data_sequence; + +-- We need to take the max across all the account_data tables as they share the +-- ID generator +SELECT setval('account_data_sequence', ( + SELECT GREATEST( + (SELECT COALESCE(MAX(stream_id), 1) FROM room_account_data), + (SELECT COALESCE(MAX(stream_id), 1) FROM room_tags_revisions), + (SELECT COALESCE(MAX(stream_id), 1) FROM account_data) + ) +)); + +CREATE SEQUENCE IF NOT EXISTS receipts_sequence; + +SELECT setval('receipts_sequence', ( + SELECT COALESCE(MAX(stream_id), 1) FROM receipts_linearized +)); diff --git a/synapse/storage/databases/main/schema/delta/59/07shard_account_data_fix.sql b/synapse/storage/databases/main/schema/delta/59/07shard_account_data_fix.sql new file mode 100644 index 0000000000..9f2b5ebc5a --- /dev/null +++ b/synapse/storage/databases/main/schema/delta/59/07shard_account_data_fix.sql @@ -0,0 +1,18 @@ +/* Copyright 2021 The Matrix.org Foundation C.I.C + * + * 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. + */ + +-- We incorrectly populated these, so we delete them and let the +-- MultiWriterIdGenerator repopulate it. +DELETE FROM stream_positions WHERE stream_name = 'receipts' OR stream_name = 'account_data'; diff --git a/synapse/storage/databases/main/search.py b/synapse/storage/databases/main/search.py index e34fce6281..871af64b11 100644 --- a/synapse/storage/databases/main/search.py +++ b/synapse/storage/databases/main/search.py @@ -63,7 +63,7 @@ class SearchWorkerStore(SQLBaseStore): for entry in entries ) - txn.executemany(sql, args) + txn.execute_batch(sql, args) elif isinstance(self.database_engine, Sqlite3Engine): sql = ( @@ -75,7 +75,7 @@ class SearchWorkerStore(SQLBaseStore): for entry in entries ) - txn.executemany(sql, args) + txn.execute_batch(sql, args) else: # This should be unreachable. raise Exception("Unrecognized database engine") diff --git a/synapse/storage/databases/main/tags.py b/synapse/storage/databases/main/tags.py index 9f120d3cb6..50067eabfc 100644 --- a/synapse/storage/databases/main/tags.py +++ b/synapse/storage/databases/main/tags.py @@ -183,8 +183,6 @@ class TagsWorkerStore(AccountDataWorkerStore): ) return {row["tag"]: db_to_json(row["content"]) for row in rows} - -class TagsStore(TagsWorkerStore): async def add_tag_to_room( self, user_id: str, room_id: str, tag: str, content: JsonDict ) -> int: @@ -199,6 +197,8 @@ class TagsStore(TagsWorkerStore): Returns: The next account data ID. """ + assert self._can_write_to_account_data + content_json = json_encoder.encode(content) def add_tag_txn(txn, next_id): @@ -223,6 +223,7 @@ class TagsStore(TagsWorkerStore): Returns: The next account data ID. """ + assert self._can_write_to_account_data def remove_tag_txn(txn, next_id): sql = ( @@ -250,21 +251,12 @@ class TagsStore(TagsWorkerStore): room_id: The ID of the room. next_id: The the revision to advance to. """ + assert self._can_write_to_account_data txn.call_after( self._account_data_stream_cache.entity_has_changed, user_id, next_id ) - # Note: This is only here for backwards compat to allow admins to - # roll back to a previous Synapse version. Next time we update the - # database version we can remove this table. - update_max_id_sql = ( - "UPDATE account_data_max_stream_id" - " SET stream_id = ?" - " WHERE stream_id < ?" - ) - txn.execute(update_max_id_sql, (next_id, next_id)) - update_sql = ( "UPDATE room_tags_revisions" " SET stream_id = ?" @@ -288,3 +280,7 @@ class TagsStore(TagsWorkerStore): # which stream_id ends up in the table, as long as it is higher # than the id that the client has. pass + + +class TagsStore(TagsWorkerStore): + pass diff --git a/synapse/storage/databases/main/transactions.py b/synapse/storage/databases/main/transactions.py index 59207cadd4..cea595ff19 100644 --- a/synapse/storage/databases/main/transactions.py +++ b/synapse/storage/databases/main/transactions.py @@ -464,19 +464,17 @@ class TransactionStore(TransactionWorkerStore): txn: LoggingTransaction, now_time_ms: int, after_destination: Optional[str] ) -> List[str]: q = """ - SELECT destination FROM destinations - WHERE destination IN ( - SELECT destination FROM destination_rooms - WHERE destination_rooms.stream_ordering > - destinations.last_successful_stream_ordering - ) - AND destination > ? - AND ( - retry_last_ts IS NULL OR - retry_last_ts + retry_interval < ? - ) - ORDER BY destination - LIMIT 25 + SELECT DISTINCT destination FROM destinations + INNER JOIN destination_rooms USING (destination) + WHERE + stream_ordering > last_successful_stream_ordering + AND destination > ? + AND ( + retry_last_ts IS NULL OR + retry_last_ts + retry_interval < ? + ) + ORDER BY destination + LIMIT 25 """ txn.execute( q, diff --git a/synapse/storage/databases/state/store.py b/synapse/storage/databases/state/store.py index 0e31cc811a..89cdc84a9c 100644 --- a/synapse/storage/databases/state/store.py +++ b/synapse/storage/databases/state/store.py @@ -565,11 +565,11 @@ class StateGroupDataStore(StateBackgroundUpdateStore, SQLBaseStore): ) logger.info("[purge] removing redundant state groups") - txn.executemany( + txn.execute_batch( "DELETE FROM state_groups_state WHERE state_group = ?", ((sg,) for sg in state_groups_to_delete), ) - txn.executemany( + txn.execute_batch( "DELETE FROM state_groups WHERE id = ?", ((sg,) for sg in state_groups_to_delete), ) |