From e3debf9682ed59b2972f236fe2982b6af0a9bb9a Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Fri, 2 Oct 2020 15:20:45 +0100 Subject: Add logging on startup/shutdown (#8448) This is so we can tell what is going on when things are taking a while to start up. The main change here is to ensure that transactions that are created during startup get correctly logged like normal transactions. --- synapse/storage/database.py | 89 +++++++++++++++++++++++++++++++++++++-------- 1 file changed, 74 insertions(+), 15 deletions(-) (limited to 'synapse/storage/database.py') diff --git a/synapse/storage/database.py b/synapse/storage/database.py index 79ec8f119d..0d9d9b7cc0 100644 --- a/synapse/storage/database.py +++ b/synapse/storage/database.py @@ -32,6 +32,7 @@ from typing import ( overload, ) +import attr from prometheus_client import Histogram from typing_extensions import Literal @@ -90,13 +91,17 @@ def make_pool( return adbapi.ConnectionPool( db_config.config["name"], cp_reactor=reactor, - cp_openfun=engine.on_new_connection, + cp_openfun=lambda conn: engine.on_new_connection( + LoggingDatabaseConnection(conn, engine, "on_new_connection") + ), **db_config.config.get("args", {}) ) def make_conn( - db_config: DatabaseConnectionConfig, engine: BaseDatabaseEngine + db_config: DatabaseConnectionConfig, + engine: BaseDatabaseEngine, + default_txn_name: str, ) -> Connection: """Make a new connection to the database and return it. @@ -109,11 +114,60 @@ def make_conn( for k, v in db_config.config.get("args", {}).items() if not k.startswith("cp_") } - db_conn = engine.module.connect(**db_params) + native_db_conn = engine.module.connect(**db_params) + db_conn = LoggingDatabaseConnection(native_db_conn, engine, default_txn_name) + engine.on_new_connection(db_conn) return db_conn +@attr.s(slots=True) +class LoggingDatabaseConnection: + """A wrapper around a database connection that returns `LoggingTransaction` + as its cursor class. + + This is mainly used on startup to ensure that queries get logged correctly + """ + + conn = attr.ib(type=Connection) + engine = attr.ib(type=BaseDatabaseEngine) + default_txn_name = attr.ib(type=str) + + def cursor( + self, *, txn_name=None, after_callbacks=None, exception_callbacks=None + ) -> "LoggingTransaction": + if not txn_name: + txn_name = self.default_txn_name + + return LoggingTransaction( + self.conn.cursor(), + name=txn_name, + database_engine=self.engine, + after_callbacks=after_callbacks, + exception_callbacks=exception_callbacks, + ) + + def close(self) -> None: + self.conn.close() + + def commit(self) -> None: + self.conn.commit() + + def rollback(self, *args, **kwargs) -> None: + self.conn.rollback(*args, **kwargs) + + def __enter__(self) -> "Connection": + self.conn.__enter__() + return self + + def __exit__(self, exc_type, exc_value, traceback) -> bool: + return self.conn.__exit__(exc_type, exc_value, traceback) + + # Proxy through any unknown lookups to the DB conn class. + def __getattr__(self, name): + return getattr(self.conn, name) + + # The type of entry which goes on our after_callbacks and exception_callbacks lists. # # Python 3.5.2 doesn't support Callable with an ellipsis, so we wrap it in quotes so @@ -247,6 +301,12 @@ class LoggingTransaction: def close(self) -> None: self.txn.close() + def __enter__(self) -> "LoggingTransaction": + return self + + def __exit__(self, exc_type, exc_value, traceback): + self.close() + class PerformanceCounters: def __init__(self): @@ -395,7 +455,7 @@ class DatabasePool: def new_transaction( self, - conn: Connection, + conn: LoggingDatabaseConnection, desc: str, after_callbacks: List[_CallbackListEntry], exception_callbacks: List[_CallbackListEntry], @@ -418,12 +478,10 @@ class DatabasePool: i = 0 N = 5 while True: - cursor = LoggingTransaction( - conn.cursor(), - name, - self.engine, - after_callbacks, - exception_callbacks, + cursor = conn.cursor( + txn_name=name, + after_callbacks=after_callbacks, + exception_callbacks=exception_callbacks, ) try: r = func(cursor, *args, **kwargs) @@ -584,7 +642,10 @@ class DatabasePool: logger.debug("Reconnecting closed database connection") conn.reconnect() - return func(conn, *args, **kwargs) + db_conn = LoggingDatabaseConnection( + conn, self.engine, "runWithConnection" + ) + return func(db_conn, *args, **kwargs) return await make_deferred_yieldable( self._db_pool.runWithConnection(inner_func, *args, **kwargs) @@ -1621,7 +1682,7 @@ class DatabasePool: def get_cache_dict( self, - db_conn: Connection, + db_conn: LoggingDatabaseConnection, table: str, entity_column: str, stream_column: str, @@ -1642,9 +1703,7 @@ class DatabasePool: "limit": limit, } - sql = self.engine.convert_param_style(sql) - - txn = db_conn.cursor() + txn = db_conn.cursor(txn_name="get_cache_dict") txn.execute(sql, (int(max_value),)) cache = {row[0]: int(row[1]) for row in txn} -- cgit 1.5.1 From ae5b2a72c09d67311c9830f5a6fae1decce03e1f Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Wed, 7 Oct 2020 15:15:57 +0100 Subject: Reduce serialization errors in MultiWriterIdGen (#8456) We call `_update_stream_positions_table_txn` a lot, which is an UPSERT that can conflict in `REPEATABLE READ` isolation level. Instead of doing a transaction consisting of a single query we may as well run it outside of a transaction. --- changelog.d/8456.misc | 1 + synapse/storage/database.py | 69 ++++++++++++++++++++++++++++++++--- synapse/storage/engines/_base.py | 17 +++++++++ synapse/storage/engines/postgres.py | 10 ++++- synapse/storage/engines/sqlite.py | 10 +++++ synapse/storage/util/id_generators.py | 12 +++++- tests/storage/test_base.py | 1 + 7 files changed, 112 insertions(+), 8 deletions(-) create mode 100644 changelog.d/8456.misc (limited to 'synapse/storage/database.py') diff --git a/changelog.d/8456.misc b/changelog.d/8456.misc new file mode 100644 index 0000000000..ccd260069b --- /dev/null +++ b/changelog.d/8456.misc @@ -0,0 +1 @@ +Reduce number of serialization errors of `MultiWriterIdGenerator._update_table`. diff --git a/synapse/storage/database.py b/synapse/storage/database.py index 0d9d9b7cc0..0ba3a025cf 100644 --- a/synapse/storage/database.py +++ b/synapse/storage/database.py @@ -463,6 +463,24 @@ class DatabasePool: *args: Any, **kwargs: Any ) -> R: + """Start a new database transaction with the given connection. + + Note: The given func may be called multiple times under certain + failure modes. This is normally fine when in a standard transaction, + but care must be taken if the connection is in `autocommit` mode that + the function will correctly handle being aborted and retried half way + through its execution. + + Args: + conn + desc + after_callbacks + exception_callbacks + func + *args + **kwargs + """ + start = monotonic_time() txn_id = self._TXN_ID @@ -566,7 +584,12 @@ class DatabasePool: sql_txn_timer.labels(desc).observe(duration) async def runInteraction( - self, desc: str, func: "Callable[..., R]", *args: Any, **kwargs: Any + self, + desc: str, + func: "Callable[..., R]", + *args: Any, + db_autocommit: bool = False, + **kwargs: Any ) -> R: """Starts a transaction on the database and runs a given function @@ -576,6 +599,18 @@ class DatabasePool: database transaction (twisted.enterprise.adbapi.Transaction) as its first argument, followed by `args` and `kwargs`. + db_autocommit: Whether to run the function in "autocommit" mode, + i.e. outside of a transaction. This is useful for transactions + that are only a single query. + + Currently, this is only implemented for Postgres. SQLite will still + run the function inside a transaction. + + WARNING: This means that if func fails half way through then + the changes will *not* be rolled back. `func` may also get + called multiple times if the transaction is retried, so must + correctly handle that case. + args: positional args to pass to `func` kwargs: named args to pass to `func` @@ -596,6 +631,7 @@ class DatabasePool: exception_callbacks, func, *args, + db_autocommit=db_autocommit, **kwargs ) @@ -609,7 +645,11 @@ class DatabasePool: return cast(R, result) async def runWithConnection( - self, func: "Callable[..., R]", *args: Any, **kwargs: Any + self, + func: "Callable[..., R]", + *args: Any, + db_autocommit: bool = False, + **kwargs: Any ) -> R: """Wraps the .runWithConnection() method on the underlying db_pool. @@ -618,6 +658,9 @@ class DatabasePool: database connection (twisted.enterprise.adbapi.Connection) as its first argument, followed by `args` and `kwargs`. args: positional args to pass to `func` + db_autocommit: Whether to run the function in "autocommit" mode, + i.e. outside of a transaction. This is useful for transaction + that are only a single query. Currently only affects postgres. kwargs: named args to pass to `func` Returns: @@ -633,6 +676,13 @@ class DatabasePool: start_time = monotonic_time() def inner_func(conn, *args, **kwargs): + # We shouldn't be in a transaction. If we are then something + # somewhere hasn't committed after doing work. (This is likely only + # possible during startup, as `run*` will ensure changes are + # committed/rolled back before putting the connection back in the + # pool). + assert not self.engine.in_transaction(conn) + with LoggingContext("runWithConnection", parent_context) as context: sched_duration_sec = monotonic_time() - start_time sql_scheduling_timer.observe(sched_duration_sec) @@ -642,10 +692,17 @@ class DatabasePool: logger.debug("Reconnecting closed database connection") conn.reconnect() - db_conn = LoggingDatabaseConnection( - conn, self.engine, "runWithConnection" - ) - return func(db_conn, *args, **kwargs) + try: + if db_autocommit: + self.engine.attempt_to_set_autocommit(conn, True) + + db_conn = LoggingDatabaseConnection( + conn, self.engine, "runWithConnection" + ) + return func(db_conn, *args, **kwargs) + finally: + if db_autocommit: + self.engine.attempt_to_set_autocommit(conn, False) return await make_deferred_yieldable( self._db_pool.runWithConnection(inner_func, *args, **kwargs) diff --git a/synapse/storage/engines/_base.py b/synapse/storage/engines/_base.py index 908cbc79e3..d6d632dc10 100644 --- a/synapse/storage/engines/_base.py +++ b/synapse/storage/engines/_base.py @@ -97,3 +97,20 @@ class BaseDatabaseEngine(Generic[ConnectionType], metaclass=abc.ABCMeta): """Gets a string giving the server version. For example: '3.22.0' """ ... + + @abc.abstractmethod + def in_transaction(self, conn: Connection) -> bool: + """Whether the connection is currently in a transaction. + """ + ... + + @abc.abstractmethod + def attempt_to_set_autocommit(self, conn: Connection, autocommit: bool): + """Attempt to set the connections autocommit mode. + + When True queries are run outside of transactions. + + Note: This has no effect on SQLite3, so callers still need to + commit/rollback the connections. + """ + ... diff --git a/synapse/storage/engines/postgres.py b/synapse/storage/engines/postgres.py index ff39281f85..7719ac32f7 100644 --- a/synapse/storage/engines/postgres.py +++ b/synapse/storage/engines/postgres.py @@ -15,7 +15,8 @@ import logging -from ._base import BaseDatabaseEngine, IncorrectDatabaseSetup +from synapse.storage.engines._base import BaseDatabaseEngine, IncorrectDatabaseSetup +from synapse.storage.types import Connection logger = logging.getLogger(__name__) @@ -119,6 +120,7 @@ class PostgresEngine(BaseDatabaseEngine): cursor.execute("SET synchronous_commit TO OFF") cursor.close() + db_conn.commit() @property def can_native_upsert(self): @@ -171,3 +173,9 @@ class PostgresEngine(BaseDatabaseEngine): return "%i.%i" % (numver / 10000, numver % 10000) else: return "%i.%i.%i" % (numver / 10000, (numver % 10000) / 100, numver % 100) + + def in_transaction(self, conn: Connection) -> bool: + return conn.status != self.module.extensions.STATUS_READY # type: ignore + + def attempt_to_set_autocommit(self, conn: Connection, autocommit: bool): + return conn.set_session(autocommit=autocommit) # type: ignore diff --git a/synapse/storage/engines/sqlite.py b/synapse/storage/engines/sqlite.py index 8a0f8c89d1..5db0f0b520 100644 --- a/synapse/storage/engines/sqlite.py +++ b/synapse/storage/engines/sqlite.py @@ -17,6 +17,7 @@ import threading import typing from synapse.storage.engines import BaseDatabaseEngine +from synapse.storage.types import Connection if typing.TYPE_CHECKING: import sqlite3 # noqa: F401 @@ -86,6 +87,7 @@ class Sqlite3Engine(BaseDatabaseEngine["sqlite3.Connection"]): db_conn.create_function("rank", 1, _rank) db_conn.execute("PRAGMA foreign_keys = ON;") + db_conn.commit() def is_deadlock(self, error): return False @@ -105,6 +107,14 @@ class Sqlite3Engine(BaseDatabaseEngine["sqlite3.Connection"]): """ return "%i.%i.%i" % self.module.sqlite_version_info + def in_transaction(self, conn: Connection) -> bool: + return conn.in_transaction # type: ignore + + def attempt_to_set_autocommit(self, conn: Connection, autocommit: bool): + # Twisted doesn't let us set attributes on the connections, so we can't + # set the connection to autocommit mode. + pass + # Following functions taken from: https://github.com/coleifer/peewee diff --git a/synapse/storage/util/id_generators.py b/synapse/storage/util/id_generators.py index 51f680d05d..d7e40aaa8b 100644 --- a/synapse/storage/util/id_generators.py +++ b/synapse/storage/util/id_generators.py @@ -24,6 +24,7 @@ from typing_extensions import Deque from synapse.metrics.background_process_metrics import run_as_background_process from synapse.storage.database import DatabasePool, LoggingTransaction +from synapse.storage.types import Cursor from synapse.storage.util.sequence import PostgresSequenceGenerator logger = logging.getLogger(__name__) @@ -548,7 +549,7 @@ class MultiWriterIdGenerator: # do. break - def _update_stream_positions_table_txn(self, txn): + def _update_stream_positions_table_txn(self, txn: Cursor): """Update the `stream_positions` table with newly persisted position. """ @@ -598,10 +599,13 @@ class _MultiWriterCtxManager: stream_ids = attr.ib(type=List[int], factory=list) async def __aenter__(self) -> Union[int, List[int]]: + # It's safe to run this in autocommit mode as fetching values from a + # sequence ignores transaction semantics anyway. self.stream_ids = await self.id_gen._db.runInteraction( "_load_next_mult_id", self.id_gen._load_next_mult_id_txn, self.multiple_ids or 1, + db_autocommit=True, ) # Assert the fetched ID is actually greater than any ID we've already @@ -632,10 +636,16 @@ class _MultiWriterCtxManager: # # We only do this on the success path so that the persisted current # position points to a persisted row with the correct instance name. + # + # We do this in autocommit mode as a) the upsert works correctly outside + # transactions and b) reduces the amount of time the rows are locked + # for. If we don't do this then we'll often hit serialization errors due + # to the fact we default to REPEATABLE READ isolation levels. if self.id_gen._writers: await self.id_gen._db.runInteraction( "MultiWriterIdGenerator._update_table", self.id_gen._update_stream_positions_table_txn, + db_autocommit=True, ) return False diff --git a/tests/storage/test_base.py b/tests/storage/test_base.py index 40ba652248..eac7e4dcd2 100644 --- a/tests/storage/test_base.py +++ b/tests/storage/test_base.py @@ -56,6 +56,7 @@ class SQLBaseStoreTestCase(unittest.TestCase): engine = create_engine(sqlite_config) fake_engine = Mock(wraps=engine) fake_engine.can_native_upsert = False + fake_engine.in_transaction.return_value = False db = DatabasePool(Mock(), Mock(config=sqlite_config), fake_engine) db._db_pool = self.db_pool -- cgit 1.5.1 From 19b15d63e80dd1d925ff78158f7a191427d6403f Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Wed, 14 Oct 2020 15:50:59 +0100 Subject: Use autocommit mode for single statement DB functions. (#8542) Autocommit means that we don't wrap the functions in transactions, and instead get executed directly. Introduced in #8456. This will help: 1. reduce the number of `could not serialize access due to concurrent delete` errors that we see (though there are a few functions that often cause serialization errors that we don't fix here); 2. improve the DB performance, as it no longer needs to deal with the overhead of `REPEATABLE READ` isolation levels; and 3. improve wall clock speed of these functions, as we no longer need to send `BEGIN` and `COMMIT` to the DB. Some notes about the differences between autocommit mode and our default `REPEATABLE READ` transactions: 1. Currently `autocommit` only applies when using PostgreSQL, and is ignored when using SQLite (due to silliness with [Twisted DB classes](https://twistedmatrix.com/trac/ticket/9998)). 2. Autocommit functions may get retried on error, which means they can get applied *twice* (or more) to the DB (since they are not in a transaction the previous call would not get rolled back). This means that the functions need to be idempotent (or otherwise not care about being called multiple times). Read queries, simple deletes, and updates/upserts that replace rows (rather than generating new values from existing rows) are all idempotent. 3. Autocommit functions no longer get executed in [`REPEATABLE READ`](https://www.postgresql.org/docs/current/transaction-iso.html) isolation level, and so data can change queries, which is fine for single statement queries. --- changelog.d/8542.misc | 1 + synapse/storage/database.py | 99 ++++++++++++++++++++++-- synapse/storage/databases/main/keys.py | 5 +- synapse/storage/databases/main/transactions.py | 76 ++++++++++-------- synapse/storage/databases/main/user_directory.py | 45 ++++------- 5 files changed, 156 insertions(+), 70 deletions(-) create mode 100644 changelog.d/8542.misc (limited to 'synapse/storage/database.py') diff --git a/changelog.d/8542.misc b/changelog.d/8542.misc new file mode 100644 index 0000000000..63149fd9b9 --- /dev/null +++ b/changelog.d/8542.misc @@ -0,0 +1 @@ +Improve database performance by executing more queries without starting transactions. diff --git a/synapse/storage/database.py b/synapse/storage/database.py index 0ba3a025cf..763722d6bc 100644 --- a/synapse/storage/database.py +++ b/synapse/storage/database.py @@ -893,6 +893,12 @@ class DatabasePool: attempts = 0 while True: try: + # We can autocommit if we are going to use native upserts + autocommit = ( + self.engine.can_native_upsert + and table not in self._unsafe_to_upsert_tables + ) + return await self.runInteraction( desc, self.simple_upsert_txn, @@ -901,6 +907,7 @@ class DatabasePool: values, insertion_values, lock=lock, + db_autocommit=autocommit, ) except self.engine.module.IntegrityError as e: attempts += 1 @@ -1063,6 +1070,43 @@ class DatabasePool: ) txn.execute(sql, list(allvalues.values())) + async def simple_upsert_many( + self, + table: str, + key_names: Collection[str], + key_values: Collection[Iterable[Any]], + value_names: Collection[str], + value_values: Iterable[Iterable[Any]], + desc: str, + ) -> None: + """ + Upsert, many times. + + Args: + table: The table to upsert into + key_names: The key column names. + key_values: A list of each row's key column values. + value_names: The value column names + value_values: A list of each row's value column values. + Ignored if value_names is empty. + """ + + # We can autocommit if we are going to use native upserts + autocommit = ( + self.engine.can_native_upsert and table not in self._unsafe_to_upsert_tables + ) + + return await self.runInteraction( + desc, + self.simple_upsert_many_txn, + table, + key_names, + key_values, + value_names, + value_values, + db_autocommit=autocommit, + ) + def simple_upsert_many_txn( self, txn: LoggingTransaction, @@ -1214,7 +1258,13 @@ class DatabasePool: desc: description of the transaction, for logging and metrics """ return await self.runInteraction( - desc, self.simple_select_one_txn, table, keyvalues, retcols, allow_none + desc, + self.simple_select_one_txn, + table, + keyvalues, + retcols, + allow_none, + db_autocommit=True, ) @overload @@ -1265,6 +1315,7 @@ class DatabasePool: keyvalues, retcol, allow_none=allow_none, + db_autocommit=True, ) @overload @@ -1346,7 +1397,12 @@ class DatabasePool: Results in a list """ return await self.runInteraction( - desc, self.simple_select_onecol_txn, table, keyvalues, retcol + desc, + self.simple_select_onecol_txn, + table, + keyvalues, + retcol, + db_autocommit=True, ) async def simple_select_list( @@ -1371,7 +1427,12 @@ class DatabasePool: A list of dictionaries. """ return await self.runInteraction( - desc, self.simple_select_list_txn, table, keyvalues, retcols + desc, + self.simple_select_list_txn, + table, + keyvalues, + retcols, + db_autocommit=True, ) @classmethod @@ -1450,6 +1511,7 @@ class DatabasePool: chunk, keyvalues, retcols, + db_autocommit=True, ) results.extend(rows) @@ -1548,7 +1610,12 @@ class DatabasePool: desc: description of the transaction, for logging and metrics """ await self.runInteraction( - desc, self.simple_update_one_txn, table, keyvalues, updatevalues + desc, + self.simple_update_one_txn, + table, + keyvalues, + updatevalues, + db_autocommit=True, ) @classmethod @@ -1607,7 +1674,9 @@ class DatabasePool: keyvalues: dict of column names and values to select the row with desc: description of the transaction, for logging and metrics """ - await self.runInteraction(desc, self.simple_delete_one_txn, table, keyvalues) + await self.runInteraction( + desc, self.simple_delete_one_txn, table, keyvalues, db_autocommit=True, + ) @staticmethod def simple_delete_one_txn( @@ -1646,7 +1715,9 @@ class DatabasePool: Returns: The number of deleted rows. """ - return await self.runInteraction(desc, self.simple_delete_txn, table, keyvalues) + return await self.runInteraction( + desc, self.simple_delete_txn, table, keyvalues, db_autocommit=True + ) @staticmethod def simple_delete_txn( @@ -1694,7 +1765,13 @@ class DatabasePool: Number rows deleted """ return await self.runInteraction( - desc, self.simple_delete_many_txn, table, column, iterable, keyvalues + desc, + self.simple_delete_many_txn, + table, + column, + iterable, + keyvalues, + db_autocommit=True, ) @staticmethod @@ -1860,7 +1937,13 @@ class DatabasePool: """ return await self.runInteraction( - desc, self.simple_search_list_txn, table, term, col, retcols + desc, + self.simple_search_list_txn, + table, + term, + col, + retcols, + db_autocommit=True, ) @classmethod diff --git a/synapse/storage/databases/main/keys.py b/synapse/storage/databases/main/keys.py index ad43bb05ab..f8f4bb9b3f 100644 --- a/synapse/storage/databases/main/keys.py +++ b/synapse/storage/databases/main/keys.py @@ -122,9 +122,7 @@ class KeyStore(SQLBaseStore): # param, which is itself the 2-tuple (server_name, key_id). invalidations.append((server_name, key_id)) - await self.db_pool.runInteraction( - "store_server_verify_keys", - self.db_pool.simple_upsert_many_txn, + await self.db_pool.simple_upsert_many( table="server_signature_keys", key_names=("server_name", "key_id"), key_values=key_values, @@ -135,6 +133,7 @@ class KeyStore(SQLBaseStore): "verify_key", ), value_values=value_values, + desc="store_server_verify_keys", ) invalidate = self._get_server_verify_key.invalidate diff --git a/synapse/storage/databases/main/transactions.py b/synapse/storage/databases/main/transactions.py index 7d46090267..59207cadd4 100644 --- a/synapse/storage/databases/main/transactions.py +++ b/synapse/storage/databases/main/transactions.py @@ -208,42 +208,56 @@ class TransactionStore(TransactionWorkerStore): """ self._destination_retry_cache.pop(destination, None) - return await self.db_pool.runInteraction( - "set_destination_retry_timings", - self._set_destination_retry_timings, - destination, - failure_ts, - retry_last_ts, - retry_interval, - ) + if self.database_engine.can_native_upsert: + return await self.db_pool.runInteraction( + "set_destination_retry_timings", + self._set_destination_retry_timings_native, + destination, + failure_ts, + retry_last_ts, + retry_interval, + db_autocommit=True, # Safe as its a single upsert + ) + else: + return await self.db_pool.runInteraction( + "set_destination_retry_timings", + self._set_destination_retry_timings_emulated, + destination, + failure_ts, + retry_last_ts, + retry_interval, + ) - def _set_destination_retry_timings( + def _set_destination_retry_timings_native( self, txn, destination, failure_ts, retry_last_ts, retry_interval ): + assert self.database_engine.can_native_upsert + + # Upsert retry time interval if retry_interval is zero (i.e. we're + # resetting it) or greater than the existing retry interval. + # + # WARNING: This is executed in autocommit, so we shouldn't add any more + # SQL calls in here (without being very careful). + sql = """ + INSERT INTO destinations ( + destination, failure_ts, retry_last_ts, retry_interval + ) + VALUES (?, ?, ?, ?) + ON CONFLICT (destination) DO UPDATE SET + failure_ts = EXCLUDED.failure_ts, + retry_last_ts = EXCLUDED.retry_last_ts, + retry_interval = EXCLUDED.retry_interval + WHERE + EXCLUDED.retry_interval = 0 + OR destinations.retry_interval IS NULL + OR destinations.retry_interval < EXCLUDED.retry_interval + """ - if self.database_engine.can_native_upsert: - # Upsert retry time interval if retry_interval is zero (i.e. we're - # resetting it) or greater than the existing retry interval. - - sql = """ - INSERT INTO destinations ( - destination, failure_ts, retry_last_ts, retry_interval - ) - VALUES (?, ?, ?, ?) - ON CONFLICT (destination) DO UPDATE SET - failure_ts = EXCLUDED.failure_ts, - retry_last_ts = EXCLUDED.retry_last_ts, - retry_interval = EXCLUDED.retry_interval - WHERE - EXCLUDED.retry_interval = 0 - OR destinations.retry_interval IS NULL - OR destinations.retry_interval < EXCLUDED.retry_interval - """ - - txn.execute(sql, (destination, failure_ts, retry_last_ts, retry_interval)) - - return + txn.execute(sql, (destination, failure_ts, retry_last_ts, retry_interval)) + def _set_destination_retry_timings_emulated( + self, txn, destination, failure_ts, retry_last_ts, retry_interval + ): self.database_engine.lock_table(txn, "destinations") # We need to be careful here as the data may have changed from under us diff --git a/synapse/storage/databases/main/user_directory.py b/synapse/storage/databases/main/user_directory.py index 5a390ff2f6..d87ceec6da 100644 --- a/synapse/storage/databases/main/user_directory.py +++ b/synapse/storage/databases/main/user_directory.py @@ -480,21 +480,16 @@ class UserDirectoryBackgroundUpdateStore(StateDeltasStore): user_id_tuples: iterable of 2-tuple of user IDs. """ - def _add_users_who_share_room_txn(txn): - self.db_pool.simple_upsert_many_txn( - txn, - table="users_who_share_private_rooms", - key_names=["user_id", "other_user_id", "room_id"], - key_values=[ - (user_id, other_user_id, room_id) - for user_id, other_user_id in user_id_tuples - ], - value_names=(), - value_values=None, - ) - - await self.db_pool.runInteraction( - "add_users_who_share_room", _add_users_who_share_room_txn + await self.db_pool.simple_upsert_many( + table="users_who_share_private_rooms", + key_names=["user_id", "other_user_id", "room_id"], + key_values=[ + (user_id, other_user_id, room_id) + for user_id, other_user_id in user_id_tuples + ], + value_names=(), + value_values=None, + desc="add_users_who_share_room", ) async def add_users_in_public_rooms( @@ -508,19 +503,13 @@ class UserDirectoryBackgroundUpdateStore(StateDeltasStore): user_ids """ - def _add_users_in_public_rooms_txn(txn): - - self.db_pool.simple_upsert_many_txn( - txn, - table="users_in_public_rooms", - key_names=["user_id", "room_id"], - key_values=[(user_id, room_id) for user_id in user_ids], - value_names=(), - value_values=None, - ) - - await self.db_pool.runInteraction( - "add_users_in_public_rooms", _add_users_in_public_rooms_txn + await self.db_pool.simple_upsert_many( + table="users_in_public_rooms", + key_names=["user_id", "room_id"], + key_values=[(user_id, room_id) for user_id in user_ids], + value_names=(), + value_values=None, + desc="add_users_in_public_rooms", ) async def delete_all_from_user_dir(self) -> None: -- cgit 1.5.1 From 79c1f973cee73f2d24c1b9140500e1999b25a479 Mon Sep 17 00:00:00 2001 From: Jonathan de Jong Date: Sat, 17 Oct 2020 10:51:38 +0200 Subject: Pre-emptively fix synapse.storage.types.Connection for future mypy release (#8577) Fix the Connection protocol according to typeshed's assertions about sqlite3.Connection --- changelog.d/8577.misc | 1 + synapse/storage/database.py | 2 +- synapse/storage/types.py | 4 ++-- 3 files changed, 4 insertions(+), 3 deletions(-) create mode 100644 changelog.d/8577.misc (limited to 'synapse/storage/database.py') diff --git a/changelog.d/8577.misc b/changelog.d/8577.misc new file mode 100644 index 0000000000..75fe563a02 --- /dev/null +++ b/changelog.d/8577.misc @@ -0,0 +1 @@ +Adjust a protocol-type definition to fit `sqlite3` assertions. \ No newline at end of file diff --git a/synapse/storage/database.py b/synapse/storage/database.py index 763722d6bc..0217e63108 100644 --- a/synapse/storage/database.py +++ b/synapse/storage/database.py @@ -160,7 +160,7 @@ class LoggingDatabaseConnection: self.conn.__enter__() return self - def __exit__(self, exc_type, exc_value, traceback) -> bool: + def __exit__(self, exc_type, exc_value, traceback) -> Optional[bool]: return self.conn.__exit__(exc_type, exc_value, traceback) # Proxy through any unknown lookups to the DB conn class. diff --git a/synapse/storage/types.py b/synapse/storage/types.py index 970bb1b9da..9cadcba18f 100644 --- a/synapse/storage/types.py +++ b/synapse/storage/types.py @@ -12,7 +12,7 @@ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. -from typing import Any, Iterable, Iterator, List, Tuple +from typing import Any, Iterable, Iterator, List, Optional, Tuple from typing_extensions import Protocol @@ -65,5 +65,5 @@ class Connection(Protocol): def __enter__(self) -> "Connection": ... - def __exit__(self, exc_type, exc_value, traceback) -> bool: + def __exit__(self, exc_type, exc_value, traceback) -> Optional[bool]: ... -- cgit 1.5.1 From aff1eb7c671b0a3813407321d2702ec46c71fa56 Mon Sep 17 00:00:00 2001 From: Dan Callahan Date: Tue, 27 Oct 2020 23:26:36 +0000 Subject: Tell Black to format code for Python 3.5 (#8664) This allows trailing commas in multi-line arg lists. Minor, but we might as well keep our formatting current with regard to our minimum supported Python version. Signed-off-by: Dan Callahan --- changelog.d/8664.misc | 1 + pyproject.toml | 2 +- synapse/http/client.py | 2 +- synapse/storage/database.py | 4 ++-- synapse/util/retryutils.py | 2 +- tests/replication/_base.py | 2 +- tests/replication/tcp/streams/test_events.py | 2 +- tests/server.py | 4 ++-- tests/storage/test_client_ips.py | 2 +- tests/test_utils/event_injection.py | 2 +- 10 files changed, 12 insertions(+), 11 deletions(-) create mode 100644 changelog.d/8664.misc (limited to 'synapse/storage/database.py') diff --git a/changelog.d/8664.misc b/changelog.d/8664.misc new file mode 100644 index 0000000000..278cf53adc --- /dev/null +++ b/changelog.d/8664.misc @@ -0,0 +1 @@ +Tell Black to format code for Python 3.5. diff --git a/pyproject.toml b/pyproject.toml index db4a2e41e4..cd880d4e39 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -35,7 +35,7 @@ showcontent = true [tool.black] -target-version = ['py34'] +target-version = ['py35'] exclude = ''' ( diff --git a/synapse/http/client.py b/synapse/http/client.py index 8324632cb6..f409368802 100644 --- a/synapse/http/client.py +++ b/synapse/http/client.py @@ -359,7 +359,7 @@ class SimpleHttpClient: agent=self.agent, data=body_producer, headers=headers, - **self._extra_treq_args + **self._extra_treq_args, ) # type: defer.Deferred # we use our own timeout mechanism rather than treq's as a workaround diff --git a/synapse/storage/database.py b/synapse/storage/database.py index 0217e63108..a0572b2952 100644 --- a/synapse/storage/database.py +++ b/synapse/storage/database.py @@ -94,7 +94,7 @@ def make_pool( cp_openfun=lambda conn: engine.on_new_connection( LoggingDatabaseConnection(conn, engine, "on_new_connection") ), - **db_config.config.get("args", {}) + **db_config.config.get("args", {}), ) @@ -632,7 +632,7 @@ class DatabasePool: func, *args, db_autocommit=db_autocommit, - **kwargs + **kwargs, ) for after_callback, after_args, after_kwargs in after_callbacks: diff --git a/synapse/util/retryutils.py b/synapse/util/retryutils.py index a5cc9d0551..4ab379e429 100644 --- a/synapse/util/retryutils.py +++ b/synapse/util/retryutils.py @@ -110,7 +110,7 @@ async def get_retry_limiter(destination, clock, store, ignore_backoff=False, **k failure_ts, retry_interval, backoff_on_failure=backoff_on_failure, - **kwargs + **kwargs, ) diff --git a/tests/replication/_base.py b/tests/replication/_base.py index 093e2faac7..f1e53f33cd 100644 --- a/tests/replication/_base.py +++ b/tests/replication/_base.py @@ -269,7 +269,7 @@ class BaseMultiWorkerStreamTestCase(unittest.HomeserverTestCase): homeserver_to_use=GenericWorkerServer, config=config, reactor=self.reactor, - **kwargs + **kwargs, ) # If the instance is in the `instance_map` config then workers may try diff --git a/tests/replication/tcp/streams/test_events.py b/tests/replication/tcp/streams/test_events.py index c9998e88e6..bad0df08cf 100644 --- a/tests/replication/tcp/streams/test_events.py +++ b/tests/replication/tcp/streams/test_events.py @@ -449,7 +449,7 @@ class EventsStreamTestCase(BaseStreamTestCase): sender=sender, type="test_event", content={"body": body}, - **kwargs + **kwargs, ) ) diff --git a/tests/server.py b/tests/server.py index 4d33b84097..ea9c22bc51 100644 --- a/tests/server.py +++ b/tests/server.py @@ -380,7 +380,7 @@ def setup_test_homeserver(cleanup_func, *args, **kwargs): pool._runWithConnection, func, *args, - **kwargs + **kwargs, ) def runInteraction(interaction, *args, **kwargs): @@ -390,7 +390,7 @@ def setup_test_homeserver(cleanup_func, *args, **kwargs): pool._runInteraction, interaction, *args, - **kwargs + **kwargs, ) pool.runWithConnection = runWithConnection diff --git a/tests/storage/test_client_ips.py b/tests/storage/test_client_ips.py index 755c70db31..e96ca1c8ca 100644 --- a/tests/storage/test_client_ips.py +++ b/tests/storage/test_client_ips.py @@ -412,7 +412,7 @@ class ClientIpAuthTestCase(unittest.HomeserverTestCase): "GET", "/_matrix/client/r0/admin/users/" + self.user_id, access_token=access_token, - **make_request_args + **make_request_args, ) request.requestHeaders.addRawHeader(b"User-Agent", b"Mozzila pizza") diff --git a/tests/test_utils/event_injection.py b/tests/test_utils/event_injection.py index e93aa84405..c3c4a93e1f 100644 --- a/tests/test_utils/event_injection.py +++ b/tests/test_utils/event_injection.py @@ -50,7 +50,7 @@ async def inject_member_event( sender=sender, state_key=target, content=content, - **kwargs + **kwargs, ) -- cgit 1.5.1 From c2d4467cd435851484b29685959124c541d47842 Mon Sep 17 00:00:00 2001 From: Erik Johnston Date: Thu, 12 Nov 2020 14:26:24 +0000 Subject: Enable reconnection in DB pool (#8726) `adbapi.ConnectionPool` let's you turn on auto reconnect of DB connections. This is off by default. As far as I can tell if its not enabled dead connections never get removed from the pool. Maybe helps #8574 --- changelog.d/8726.bugfix | 1 + synapse/storage/database.py | 7 ++++++- 2 files changed, 7 insertions(+), 1 deletion(-) create mode 100644 changelog.d/8726.bugfix (limited to 'synapse/storage/database.py') diff --git a/changelog.d/8726.bugfix b/changelog.d/8726.bugfix new file mode 100644 index 0000000000..831f773a25 --- /dev/null +++ b/changelog.d/8726.bugfix @@ -0,0 +1 @@ +Fix bug where Synapse would not recover after losing connection to the database. diff --git a/synapse/storage/database.py b/synapse/storage/database.py index a0572b2952..d1b5760c2c 100644 --- a/synapse/storage/database.py +++ b/synapse/storage/database.py @@ -88,13 +88,18 @@ def make_pool( """Get the connection pool for the database. """ + # By default enable `cp_reconnect`. We need to fiddle with db_args in case + # someone has explicitly set `cp_reconnect`. + db_args = dict(db_config.config.get("args", {})) + db_args.setdefault("cp_reconnect", True) + return adbapi.ConnectionPool( db_config.config["name"], cp_reactor=reactor, cp_openfun=lambda conn: engine.on_new_connection( LoggingDatabaseConnection(conn, engine, "on_new_connection") ), - **db_config.config.get("args", {}), + **db_args, ) -- cgit 1.5.1