# Copyright 2011-present MongoDB, Inc. # # 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 # # https://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or # implied. See the License for the specific language governing # permissions and limitations under the License. from __future__ import annotations import asyncio import collections import os import socket import time import weakref from collections.abc import Mapping, MutableMapping, Sequence from typing import ( TYPE_CHECKING, Any, NoReturn, Optional, Union, ) from bson import DEFAULT_CODEC_OPTIONS from pymongo import _csot, helpers_shared from pymongo._telemetry import _CmapTelemetry from pymongo.client_session_shared import _validate_session_write_concern from pymongo.common import ( MAX_BSON_SIZE, MAX_MESSAGE_SIZE, MAX_WIRE_VERSION, MAX_WRITE_BATCH_SIZE, ORDERED_TYPES, ) from pymongo.errors import ( # type:ignore[attr-defined] AutoReconnect, ConfigurationError, DocumentTooLarge, ExecutionTimeout, InvalidOperation, NetworkTimeout, NotPrimaryError, OperationFailure, WaitQueueTimeoutError, _CertificateError, ) from pymongo.hello import Hello, HelloCompat from pymongo.helpers_shared import _get_timeout_details, format_timeout_details from pymongo.lock import ( _cond_wait, _create_condition, _create_lock, ) from pymongo.logger import _CONNECTION_LOGGER, _is_debug_enabled from pymongo.monitoring import ( ConnectionCheckOutFailedReason, ConnectionClosedReason, ) from pymongo.network_layer import NetworkingInterface, receive_message, sendall from pymongo.pool_options import PoolOptions from pymongo.pool_shared import ( PoolState, SSLErrors, _CancellationContext, _configured_socket_interface, _ConnectionTelemetryInfo, _PoolClosedError, _PoolGeneration, _raise_connection_failure, ) from pymongo.read_preferences import ReadPreference from pymongo.server_api import _add_to_command from pymongo.server_type import SERVER_TYPE from pymongo.socket_checker import SocketChecker from pymongo.ssl_support import SSL_EOF_ERRORS from pymongo.synchronous.command_runner import run_command from pymongo.synchronous.helpers import _handle_reauth if TYPE_CHECKING: from types import TracebackType from bson import CodecOptions from bson.objectid import ObjectId from pymongo.compression_support import ( SnappyContext, ZlibContext, ZstdContext, ) from pymongo.message import _OpMsg from pymongo.read_concern import ReadConcern from pymongo.read_preferences import _ServerMode from pymongo.synchronous.auth import _AuthContext from pymongo.synchronous.client_session import ClientSession from pymongo.synchronous.mongo_client import MongoClient, _ClientCheckout from pymongo.typings import _Address, _CollationIn from pymongo.write_concern import WriteConcern _IS_SYNC = True class Connection(_ConnectionTelemetryInfo): """Store a connection with some metadata. :param conn: a raw connection object :param pool: a Pool instance :param address: the server's (host, port) :param id: the id of this socket in it's pool :param is_sdam: SDAM connections do not call hello on creation """ def __init__( self, conn: NetworkingInterface, pool: Pool, address: tuple[str, int], id: int, is_sdam: bool, ): self.pool_ref = weakref.ref(pool) self.conn = conn self.address = address self.id = id self.is_sdam = is_sdam self.closed = False self.creation_time = time.monotonic() self.last_checkin_time = time.monotonic() self.performed_handshake = False self.is_writable: bool = False self.max_wire_version = MAX_WIRE_VERSION self.max_bson_size = MAX_BSON_SIZE self.max_message_size = MAX_MESSAGE_SIZE self.max_write_batch_size = MAX_WRITE_BATCH_SIZE self.supports_sessions = False self.hello_ok: bool = False self.is_mongos = False self.listeners = pool.opts._event_listeners self._telemetry = pool._telemetry self.compression_settings = pool.opts._compression_settings self.compression_context: Union[SnappyContext, ZlibContext, ZstdContext, None] = None self.socket_checker: SocketChecker = SocketChecker() self.oidc_token_gen_id: Optional[int] = None # Support for mechanism negotiation on the initial handshake. self.negotiated_mechs: Optional[list[str]] = None self.auth_ctx: Optional[_AuthContext] = None # The pool's generation changes with each reset() so we can close # sockets created before the last reset. self.pool_gen = pool.gen self.generation = self.pool_gen.get_overall() self.ready = False self.cancel_context: _CancellationContext = _CancellationContext() self.opts = pool.opts self.more_to_come: bool = False # For load balancer support. self.service_id: Optional[ObjectId] = None self.server_connection_id: Optional[int] = None # When executing a transaction in load balancing mode, this flag is # set to true to indicate that the session now owns the connection. self.pinned_txn = False self.pinned_cursor = False self.active = False self.last_timeout = self.opts.socket_timeout self.connect_rtt = 0.0 # For gossiping $clusterTime from the connection handshake to the client. self._cluster_time = None def set_conn_timeout(self, timeout: Optional[float]) -> None: """Cache last timeout to avoid duplicate calls to conn.settimeout.""" if timeout == self.last_timeout: return self.last_timeout = timeout self.conn.get_conn.settimeout(timeout) def apply_timeout( self, client: MongoClient[Any], cmd: Optional[MutableMapping[str, Any]] ) -> Optional[float]: # CSOT: use remaining timeout when set. timeout = _csot.remaining() if timeout is None: # Reset the socket timeout unless we're performing a streaming monitor check. if not self.more_to_come: self.set_conn_timeout(self.opts.socket_timeout) return None # RTT validation. rtt = _csot.get_rtt() if rtt is None: rtt = self.connect_rtt max_time_ms = timeout - rtt if max_time_ms < 0: timeout_details = _get_timeout_details(self.opts) formatted = format_timeout_details(timeout_details) # CSOT: raise an error without running the command since we know it will time out. errmsg = f"operation would exceed time limit, remaining timeout:{timeout:.5f} <= network round trip time:{rtt:.5f} {formatted}" raise ExecutionTimeout( errmsg, 50, {"ok": 0, "errmsg": errmsg, "code": 50}, self.max_wire_version, ) if cmd is not None: cmd["maxTimeMS"] = int(max_time_ms * 1000) self.set_conn_timeout(timeout) return timeout def pin_txn(self) -> None: self.pinned_txn = True assert not self.pinned_cursor def pin_cursor(self) -> None: self.pinned_cursor = True assert not self.pinned_txn def unpin(self) -> None: pool = self.pool_ref() if pool: pool.checkin(self) else: self.close_conn(ConnectionClosedReason.STALE) def hello_cmd(self) -> dict[str, Any]: # As of PYTHON-5713, always use OP_MSG for the handshake since all # supported servers (MongoDB 4.4+, wire version >= 9) support it. if self.opts.server_api or self.hello_ok or self.opts.load_balanced: return {HelloCompat.CMD: 1} return {HelloCompat.LEGACY_CMD: 1, "helloOk": True} def hello(self) -> Hello[dict[str, Any]]: return self._hello(None, None) def _hello( self, topology_version: Optional[Any], heartbeat_frequency: Optional[int], ) -> Hello[dict[str, Any]]: cmd = self.hello_cmd() performing_handshake = not self.performed_handshake awaitable = False cmd["backpressure"] = "2" if performing_handshake: self.performed_handshake = True cmd["client"] = self.opts.metadata if self.compression_settings: cmd["compression"] = self.compression_settings.compressors if self.opts.load_balanced: cmd["loadBalanced"] = True elif topology_version is not None: cmd["topologyVersion"] = topology_version assert heartbeat_frequency is not None cmd["maxAwaitTimeMS"] = int(heartbeat_frequency * 1000) awaitable = True # If connect_timeout is None there is no timeout. if self.opts.connect_timeout: self.set_conn_timeout(self.opts.connect_timeout + heartbeat_frequency) creds = self.opts._credentials if creds: if creds.mechanism == "DEFAULT" and creds.username: cmd["saslSupportedMechs"] = creds.source + "." + creds.username from pymongo.synchronous import auth auth_ctx = auth._AuthContext.from_credentials(creds, self.address) if auth_ctx: speculative_authenticate = auth_ctx.speculate_command() if speculative_authenticate is not None: cmd["speculativeAuthenticate"] = speculative_authenticate else: auth_ctx = None if performing_handshake: start = time.monotonic() doc = self.command("admin", cmd, publish_events=False, exhaust_allowed=awaitable) if performing_handshake: self.connect_rtt = time.monotonic() - start hello = Hello(doc, awaitable=awaitable) self.is_writable = hello.is_writable self.max_wire_version = hello.max_wire_version self.max_bson_size = hello.max_bson_size self.max_message_size = hello.max_message_size self.max_write_batch_size = hello.max_write_batch_size self.supports_sessions = ( hello.logical_session_timeout_minutes is not None and hello.is_readable ) self.logical_session_timeout_minutes: Optional[int] = hello.logical_session_timeout_minutes # hello_ok is set from helloOk, which is only returned from ismaster # don't overwrite this when connection switches to hello self.hello_ok = self.hello_ok or hello.hello_ok self.is_repl = hello.server_type in ( SERVER_TYPE.RSPrimary, SERVER_TYPE.RSSecondary, SERVER_TYPE.RSArbiter, SERVER_TYPE.RSOther, SERVER_TYPE.RSGhost, ) self.is_standalone = hello.server_type == SERVER_TYPE.Standalone self.is_mongos = hello.server_type == SERVER_TYPE.Mongos if performing_handshake and self.compression_settings: ctx = self.compression_settings.get_compression_context(hello.compressors) self.compression_context = ctx self.server_connection_id = hello.connection_id if creds: self.negotiated_mechs = hello.sasl_supported_mechs if auth_ctx: auth_ctx.parse_response(hello) # type:ignore[arg-type] if auth_ctx.speculate_succeeded(): self.auth_ctx = auth_ctx if self.opts.load_balanced: if not hello.service_id: raise ConfigurationError( "Driver attempted to initialize in load balancing mode," " but the server does not support this mode" ) self.service_id = hello.service_id self.generation = self.pool_gen.get(self.service_id) return hello def _next_reply(self) -> dict[str, Any]: reply = self.receive_message(None) self.more_to_come = reply.more_to_come unpacked_docs = reply.unpack_response() response_doc = unpacked_docs[0] helpers_shared._check_command_response(response_doc, self.max_wire_version) return response_doc @_handle_reauth def command( self, dbname: str, spec: MutableMapping[str, Any], read_preference: _ServerMode = ReadPreference.PRIMARY, codec_options: CodecOptions[Mapping[str, Any]] = DEFAULT_CODEC_OPTIONS, # type: ignore[assignment] check: bool = True, allowable_errors: Optional[Sequence[Union[str, int]]] = None, read_concern: Optional[ReadConcern] = None, write_concern: Optional[WriteConcern] = None, parse_write_concern_error: bool = False, collation: Optional[_CollationIn] = None, session: Optional[ClientSession] = None, client: Optional[MongoClient[Any]] = None, retryable_write: bool = False, publish_events: bool = True, user_fields: Optional[Mapping[str, Any]] = None, exhaust_allowed: bool = False, ) -> dict[str, Any]: """Execute a command or raise an error. :param dbname: name of the database on which to run the command :param spec: a command document as a dict, SON, or mapping object :param read_preference: a read preference :param codec_options: a CodecOptions instance :param check: raise OperationFailure if there are errors :param allowable_errors: errors to ignore if `check` is True :param read_concern: The read concern for this command. :param write_concern: The write concern for this command. :param parse_write_concern_error: Whether to parse the ``writeConcernError`` field in the command response. :param collation: The collation for this command. :param session: optional ClientSession instance. :param client: optional MongoClient for gossipping $clusterTime. :param retryable_write: True if this command is a retryable write. :param publish_events: Should we publish events for this command? :param user_fields: Response fields that should be decoded using the TypeDecoders from codec_options, passed to bson._decode_all_selective. """ self.validate_session(client, session) session = _validate_session_write_concern(session, write_concern) # Ensure command name remains in first place. if not isinstance(spec, ORDERED_TYPES): # type:ignore[arg-type] spec = dict(spec) if not (write_concern is None or write_concern.acknowledged or collation is None): raise ConfigurationError("Collation is unsupported for unacknowledged writes.") self.add_server_api(spec) if session: session._apply_to(spec, retryable_write, read_preference, self) self.send_cluster_time(spec, session, client) listeners = self.listeners if publish_events else None unacknowledged = bool(write_concern and not write_concern.acknowledged) if unacknowledged: self._raise_if_not_writable() try: if session is not None and session._starting_transaction: session._transaction.set_in_progress() return run_command( self, dbname, spec, read_preference, codec_options, # type: ignore[arg-type] session, client, check, allowable_errors, listeners, self.max_bson_size, read_concern, parse_write_concern_error=parse_write_concern_error, collation=collation, compression_ctx=self.compression_context, unacknowledged=unacknowledged, user_fields=user_fields, exhaust_allowed=exhaust_allowed, write_concern=write_concern, ) except (OperationFailure, NotPrimaryError): raise # Catch socket.error, KeyboardInterrupt, CancelledError, etc. and close ourselves. except BaseException as error: self._raise_connection_failure(error) def send_message(self, message: bytes, max_doc_size: int) -> None: """Send a raw BSON message or raise ConnectionFailure. If a network exception is raised, the socket is closed. """ if self.max_bson_size is not None and max_doc_size > self.max_bson_size: raise DocumentTooLarge( f"BSON document too large ({max_doc_size} bytes) - the connected server " f"supports BSON document sizes up to {self.max_bson_size} bytes." ) try: sendall(self.conn.get_conn, message) # Catch KeyboardInterrupt, CancelledError, etc. and cleanup. except BaseException as error: self._raise_connection_failure(error) def receive_message(self, request_id: Optional[int]) -> _OpMsg: """Receive a raw BSON message or raise ConnectionFailure. If any exception is raised, the socket is closed. """ try: return receive_message(self, request_id, self.max_message_size) # Catch KeyboardInterrupt, CancelledError, etc. and cleanup. except BaseException as error: self._raise_connection_failure(error) def _raise_if_not_writable(self) -> None: """Raise NotPrimaryError if this connection is not writable.""" if not self.is_writable: raise NotPrimaryError("not primary", {"ok": 0, "errmsg": "not primary", "code": 10107}) def authenticate(self, reauthenticate: bool = False) -> None: """Authenticate to the server if needed. Can raise ConnectionFailure or OperationFailure. """ # CMAP spec says to publish the ready event only after authenticating # the connection. if reauthenticate: if self.performed_handshake: # Existing auth_ctx is stale, remove it. self.auth_ctx = None self.ready = False if not self.ready: creds = self.opts._credentials if creds: from pymongo.synchronous import auth auth.authenticate(creds, self, reauthenticate=reauthenticate) self.ready = True self._telemetry.connection_ready(self.id, self.creation_time) def validate_session( self, client: Optional[MongoClient[Any]], session: Optional[ClientSession] ) -> None: """Validate this session before use with client. Raises error if the client is not the one that created the session. """ if session: if session._client is not client: raise InvalidOperation("Can only use session with the MongoClient that started it") def close_conn(self, reason: Optional[str]) -> None: """Close this connection with a reason.""" if self.closed: return self._close_conn() if reason: self._telemetry.connection_closed(self.id, reason) def _close_conn(self) -> None: """Close this connection.""" if self.closed: return self.closed = True self.cancel_context.cancel() # Note: We catch exceptions to avoid spurious errors on interpreter # shutdown. try: self.conn.close() except Exception: # noqa: S110 pass def conn_closed(self) -> bool: """Return True if we know socket has been closed, False otherwise.""" if _IS_SYNC: return self.socket_checker.socket_closed(self.conn.get_conn) else: return self.conn.is_closing() def send_cluster_time( self, command: MutableMapping[str, Any], session: Optional[ClientSession], client: Optional[MongoClient[Any]], ) -> None: """Add $clusterTime.""" if client: client._send_cluster_time(command, session) def add_server_api(self, command: MutableMapping[str, Any]) -> None: """Add server_api parameters.""" if self.opts.server_api: _add_to_command(command, self.opts.server_api) def update_last_checkin_time(self) -> None: self.last_checkin_time = time.monotonic() def update_is_writable(self, is_writable: bool) -> None: self.is_writable = is_writable def idle_time_seconds(self) -> float: """Seconds since this socket was last checked into its pool.""" return time.monotonic() - self.last_checkin_time def _raise_connection_failure(self, error: BaseException) -> NoReturn: # Catch *all* exceptions from socket methods and close the socket. In # regular Python, socket operations only raise socket.error, even if # the underlying cause was a Ctrl-C: a signal raised during socket.recv # is expressed as an EINTR error from poll. See internal_select_ex() in # socketmodule.c. All error codes from poll become socket.error at # first. Eventually in PyEval_EvalFrameEx the interpreter checks for # signals and throws KeyboardInterrupt into the current frame on the # main thread. # # But in Gevent, the polling mechanism (epoll, kqueue, # ..) is called in Python code, which experiences the signal as a # KeyboardInterrupt from the start, rather than as an initial # socket.error, so we catch that, close the socket, and reraise it. # # The connection closed event will be emitted later in checkin. if self.ready: reason = None else: reason = ConnectionClosedReason.ERROR self.close_conn(reason) # SSLError from PyOpenSSL inherits directly from Exception. if isinstance(error, (IOError, OSError, *SSLErrors)): details = _get_timeout_details(self.opts) _raise_connection_failure(self.address, error, timeout_details=details) else: raise error def __eq__(self, other: Any) -> bool: return self.conn == other.conn def __ne__(self, other: Any) -> bool: return not self == other def __hash__(self) -> int: return hash(self.conn) def __repr__(self) -> str: return "Connection({}){} at {}".format( repr(self.conn), (self.closed and " CLOSED") or "", id(self), ) class Pool: def __init__( self, address: _Address, options: PoolOptions, is_sdam: bool = False, client_id: Optional[ObjectId] = None, ): """ :param address: a (hostname, port) tuple :param options: a PoolOptions instance :param is_sdam: whether to call hello for each new Connection """ if options.pause_enabled: self.state = PoolState.PAUSED else: self.state = PoolState.READY # Check a socket's health with socket_closed() every once in a while. # Can override for testing: 0 to always check, None to never check. self._check_interval_seconds = 1 # LIFO pool. Sockets are ordered on idle time. Sockets claimed # and returned to pool from the left side. Stale sockets removed # from the right side. self.conns: collections.deque[Connection] = collections.deque() self.active_contexts: set[_CancellationContext] = set() self.lock = _create_lock() self._max_connecting_cond = _create_condition(self.lock) self.active_sockets = 0 # Monotonically increasing connection ID required for CMAP Events. self.next_connection_id = 1 # Track whether the sockets in this pool are writeable or not. self.is_writable: Optional[bool] = None # Keep track of resets, so we notice sockets created before the most # recent reset and close them. # self.generation = 0 self.gen = _PoolGeneration() self.pid = os.getpid() self.address = address self.opts = options self.is_sdam = is_sdam # The first portion of the wait queue. # Enforces: maxPoolSize # Also used for: clearing the wait queue self.size_cond = _create_condition(self.lock) self.requests = 0 self.max_pool_size = self.opts.max_pool_size if not self.max_pool_size: self.max_pool_size = float("inf") # The second portion of the wait queue. # Enforces: maxConnecting # Also used for: clearing the wait queue self._max_connecting_cond = _create_condition(self.lock) self._pending = 0 self._max_connecting = self.opts.max_connecting self._ssl_session_cache: Optional[list[Any]] = ( [None] if self.opts._ssl_context is not None else None ) self._telemetry = _CmapTelemetry( client_id, address, options._event_listeners, publish=not is_sdam, log=not is_sdam ) self._telemetry.pool_created(self.opts.non_default_options) # Similar to active_sockets but includes threads in the wait queue. self.operation_count: int = 0 # Retain references to pinned connections to prevent the CPython GC # from thinking that a cursor's pinned connection can be GC'd when the # cursor is GC'd (see PYTHON-2751). self._pinned_sockets: set[Connection] = set() self.ncursors = 0 self.ntxns = 0 def ready(self) -> None: # Take the lock to avoid the race condition described in PYTHON-2699. with self.lock: if self.state != PoolState.READY: self.state = PoolState.READY self._telemetry.pool_ready() @property def closed(self) -> bool: return self.state == PoolState.CLOSED def _reset( self, close: bool, pause: bool = True, service_id: Optional[ObjectId] = None, interrupt_connections: bool = False, ) -> None: old_state = self.state with self.size_cond: if self.closed: return if self.opts.pause_enabled and pause and not self.opts.load_balanced: old_state, self.state = self.state, PoolState.PAUSED self.gen.inc(service_id) newpid = os.getpid() if self.pid != newpid: self.pid = newpid self.active_sockets = 0 self.operation_count = 0 if service_id is None: sockets, self.conns = self.conns, collections.deque() else: discard: collections.deque = collections.deque() # type: ignore[type-arg] keep: collections.deque = collections.deque() # type: ignore[type-arg] for conn in self.conns: if conn.service_id == service_id: discard.append(conn) else: keep.append(conn) sockets = discard self.conns = keep if close: self.state = PoolState.CLOSED # Publish PoolClearedEvent while holding the lock and before # notify_all(). This guarantees it is recorded before any waiting # thread wakes up and emits ConnectionCheckOutFailedEvent, which # also requires size_cond. Without this ordering, a race on PyPy # and free-threaded Python causes ConnectionCheckOutFailedEvent to # arrive before PoolClearedEvent (PYTHON-3519). if not close and old_state != PoolState.PAUSED: self._telemetry.pool_cleared(service_id, interrupt_connections) # Clear the wait queue self._max_connecting_cond.notify_all() self.size_cond.notify_all() if interrupt_connections: for context in self.active_contexts: context.cancel() # CMAP spec says that close() MUST close sockets before publishing the # PoolClosedEvent but that reset() SHOULD close sockets *after* # publishing the PoolClearedEvent. if close: if not _IS_SYNC: asyncio.gather( *[conn.close_conn(ConnectionClosedReason.POOL_CLOSED) for conn in sockets], # type: ignore[func-returns-value] return_exceptions=True, ) else: for conn in sockets: conn.close_conn(ConnectionClosedReason.POOL_CLOSED) self._telemetry.pool_closed() else: if not _IS_SYNC: asyncio.gather( *[conn.close_conn(ConnectionClosedReason.STALE) for conn in sockets], # type: ignore[func-returns-value] return_exceptions=True, ) else: for conn in sockets: conn.close_conn(ConnectionClosedReason.STALE) def update_is_writable(self, is_writable: Optional[bool]) -> None: """Updates the is_writable attribute on all sockets currently in the Pool. """ self.is_writable = is_writable with self.lock: for _socket in self.conns: _socket.update_is_writable(self.is_writable) # type: ignore[arg-type] def reset( self, service_id: Optional[ObjectId] = None, interrupt_connections: bool = False ) -> None: self._reset(close=False, service_id=service_id, interrupt_connections=interrupt_connections) def reset_without_pause(self) -> None: self._reset(close=False, pause=False) def close(self) -> None: self._reset(close=True) def stale_generation(self, gen: int, service_id: Optional[ObjectId]) -> bool: return self.gen.stale(gen, service_id) def remove_stale_sockets(self, reference_generation: int) -> None: """Removes stale sockets then adds new ones if pool is too small and has not been reset. The `reference_generation` argument specifies the `generation` at the point in time this operation was requested on the pool. """ # Take the lock to avoid the race condition described in PYTHON-2699. with self.lock: if self.state != PoolState.READY: return if self.opts.max_idle_time_seconds is not None: close_conns = [] with self.lock: while ( self.conns and self.conns[-1].idle_time_seconds() > self.opts.max_idle_time_seconds ): close_conns.append(self.conns.pop()) if not _IS_SYNC: asyncio.gather( *[conn.close_conn(ConnectionClosedReason.IDLE) for conn in close_conns], # type: ignore[func-returns-value] return_exceptions=True, ) else: for conn in close_conns: conn.close_conn(ConnectionClosedReason.IDLE) while True: with self.size_cond: # There are enough sockets in the pool. if len(self.conns) + self.active_sockets >= self.opts.min_pool_size: return if self.requests >= self.opts.min_pool_size: return self.requests += 1 incremented = False try: with self._max_connecting_cond: # If maxConnecting connections are already being created # by this pool then try again later instead of waiting. if self._pending >= self._max_connecting: return self._pending += 1 incremented = True conn = self.connect() close_conn = False with self.lock: # Close connection and return if the pool was reset during # socket creation or while acquiring the pool lock. if self.gen.get_overall() != reference_generation: close_conn = True if not close_conn: self.conns.appendleft(conn) self.active_contexts.discard(conn.cancel_context) if close_conn: conn.close_conn(ConnectionClosedReason.STALE) return finally: if incremented: # Notify after adding the socket to the pool. with self._max_connecting_cond: self._pending -= 1 self._max_connecting_cond.notify() with self.size_cond: self.requests -= 1 self.size_cond.notify() def _handle_connection_error(self, error: BaseException) -> None: # Handle system overload condition for non-sdam pools. # Look for errors of type AutoReconnect and add error labels if appropriate. if self.is_sdam or type(error) not in (AutoReconnect, NetworkTimeout): return assert isinstance(error, AutoReconnect) # Appease type checker. # If the original error was a DNS, certificate, or SSL error, ignore it. if isinstance(error.__cause__, (_CertificateError, SSLErrors, socket.gaierror)): # End of file errors are excluded, because the server may have disconnected # during the handshake. if not isinstance(error.__cause__, SSL_EOF_ERRORS): return error._add_error_label("SystemOverloadedError") error._add_error_label("RetryableError") def connect(self, handler: Optional[_ClientCheckout] = None) -> Connection: """Connect to Mongo and return a new Connection. Can raise ConnectionFailure. Note that the pool does not keep a reference to the socket -- you must call checkin() when you're done with it. """ with self.lock: conn_id = self.next_connection_id self.next_connection_id += 1 # Use a temporary context so that interrupt_connections can cancel creating the socket. tmp_context = _CancellationContext() self.active_contexts.add(tmp_context) self._telemetry.connection_created(conn_id) try: networking_interface = _configured_socket_interface( self.address, self.opts, self._ssl_session_cache ) # Catch KeyboardInterrupt, CancelledError, etc. and cleanup. except BaseException as error: with self.lock: self.active_contexts.discard(tmp_context) self._telemetry.connection_closed(conn_id, ConnectionClosedReason.ERROR) if isinstance(error, (IOError, OSError, *SSLErrors)): details = _get_timeout_details(self.opts) # Wrap to AutoReconnect/NetworkTimeout BEFORE labeling so the # SystemOverloadedError label lands on the propagated exception. try: _raise_connection_failure(self.address, error, timeout_details=details) except (AutoReconnect, NetworkTimeout) as wrapped: self._handle_connection_error(wrapped) raise self._handle_connection_error(error) raise conn = Connection(networking_interface, self, self.address, conn_id, self.is_sdam) # type: ignore[arg-type] with self.lock: self.active_contexts.add(conn.cancel_context) self.active_contexts.discard(tmp_context) if tmp_context.cancelled: conn.cancel_context.cancel() completed_hello = False try: if not self.is_sdam: conn.hello() completed_hello = True self.is_writable = conn.is_writable if handler: handler.contribute_socket(conn, completed_handshake=False) conn.authenticate() # Catch KeyboardInterrupt, CancelledError, etc. and cleanup. except BaseException as e: with self.lock: self.active_contexts.discard(conn.cancel_context) if not completed_hello: self._handle_connection_error(e) conn.close_conn(ConnectionClosedReason.ERROR) raise if handler: handler.client._topology.receive_cluster_time(conn._cluster_time) return conn def checkout(self, handler: Optional[_ClientCheckout] = None) -> _PoolCheckout: """Get a connection from the pool. Always use in a with statement:: with pool.checkout() as connection: connection.send_message(msg) data = connection.receive_message(op_code, request_id) Can raise ConnectionFailure or OperationFailure. :param handler: A _ClientCheckout error handler. """ return _PoolCheckout(self, handler) def _raise_if_not_ready(self, checkout_started_time: float, emit_event: bool) -> None: if self.state != PoolState.READY: if emit_event: self._telemetry.checkout_failed( "An error occurred while trying to establish a new connection", ConnectionCheckOutFailedReason.CONN_ERROR, checkout_started_time, ) details = _get_timeout_details(self.opts) _raise_connection_failure( self.address, AutoReconnect("connection pool paused"), timeout_details=details ) def _get_conn( self, checkout_started_time: float, handler: Optional[_ClientCheckout] = None ) -> Connection: """Get or create a Connection. Can raise ConnectionFailure.""" # We use the pid here to avoid issues with fork / multiprocessing. # See test.test_client:TestClient.test_fork for an example of # what could go wrong otherwise if self.pid != os.getpid(): self.reset_without_pause() if self.closed: self._telemetry.checkout_failed( "Connection pool was closed", ConnectionCheckOutFailedReason.POOL_CLOSED, checkout_started_time, ) raise _PoolClosedError( "Attempted to check out a connection from closed connection pool" ) with self.lock: self.operation_count += 1 # Get a free socket or create one. if _csot.get_timeout(): deadline = _csot.get_deadline() elif self.opts.wait_queue_timeout: deadline = time.monotonic() + self.opts.wait_queue_timeout else: deadline = None with self.size_cond: self._raise_if_not_ready(checkout_started_time, emit_event=True) while not (self.requests < self.max_pool_size): timeout = deadline - time.monotonic() if deadline else None if not _cond_wait(self.size_cond, timeout): # Timed out, notify the next thread to ensure a # timeout doesn't consume the condition. if self.requests < self.max_pool_size: self.size_cond.notify() self._raise_wait_queue_timeout(checkout_started_time) self._raise_if_not_ready(checkout_started_time, emit_event=True) self.requests += 1 # We've now acquired the semaphore and must release it on error. conn = None incremented = False emitted_event = False is_new_conn = False try: with self.lock: self.active_sockets += 1 incremented = True while conn is None: # CMAP: we MUST wait for either maxConnecting OR for a socket # to be checked back into the pool. with self._max_connecting_cond: self._raise_if_not_ready(checkout_started_time, emit_event=False) while not (self.conns or self._pending < self._max_connecting): timeout = deadline - time.monotonic() if deadline else None if not _cond_wait(self._max_connecting_cond, timeout): # Timed out, notify the next thread to ensure a # timeout doesn't consume the condition. if self.conns or self._pending < self._max_connecting: self._max_connecting_cond.notify() emitted_event = True self._raise_wait_queue_timeout(checkout_started_time) self._raise_if_not_ready(checkout_started_time, emit_event=False) try: conn = self.conns.popleft() except IndexError: self._pending += 1 if conn: # We got a socket from the pool if self._perished(conn): conn = None continue else: # We need to create a new connection try: conn = self.connect(handler=handler) is_new_conn = True finally: with self._max_connecting_cond: self._pending -= 1 self._max_connecting_cond.notify() conn.active = True # connect() already adds cancel_context for new connections; only add # here for reused connections taken from the idle pool. if not is_new_conn: with self.lock: self.active_contexts.add(conn.cancel_context) # Catch KeyboardInterrupt, CancelledError, etc. and cleanup. except BaseException: if conn: # We checked out a socket but authentication failed. conn.close_conn(ConnectionClosedReason.ERROR) with self.size_cond: self.requests -= 1 if incremented: self.active_sockets -= 1 self.size_cond.notify() if not emitted_event: self._telemetry.checkout_failed( "An error occurred while trying to establish a new connection", ConnectionCheckOutFailedReason.CONN_ERROR, checkout_started_time, ) raise return conn def checkin(self, conn: Connection) -> None: """Return the connection to the pool, or if it's closed discard it. :param conn: The connection to check into the pool. """ txn = conn.pinned_txn cursor = conn.pinned_cursor conn.active = False conn.pinned_txn = False conn.pinned_cursor = False self._pinned_sockets.discard(conn) with self.lock: self.active_contexts.discard(conn.cancel_context) telemetry = self._telemetry if telemetry._should_publish or (telemetry._log and _is_debug_enabled(_CONNECTION_LOGGER)): telemetry.checked_in(conn.id) if self.pid != os.getpid(): self.reset_without_pause() else: if self.closed: conn.close_conn(ConnectionClosedReason.POOL_CLOSED) elif conn.closed: # CMAP requires the closed event be emitted after the check in. self._telemetry.connection_closed(conn.id, ConnectionClosedReason.ERROR) else: close_conn = False with self.lock: # Hold the lock to ensure this section does not race with # Pool.reset(). if self.stale_generation(conn.generation, conn.service_id): close_conn = True else: conn.update_last_checkin_time() conn.update_is_writable(bool(self.is_writable)) self.conns.appendleft(conn) # Notify any threads waiting to create a connection. self._max_connecting_cond.notify() if close_conn: conn.close_conn(ConnectionClosedReason.STALE) with self.size_cond: if txn: self.ntxns -= 1 elif cursor: self.ncursors -= 1 self.requests -= 1 self.active_sockets -= 1 self.operation_count -= 1 self.size_cond.notify() def _perished(self, conn: Connection) -> bool: """Return True and close the connection if it is "perished". This side-effecty function checks if this socket has been idle for for longer than the max idle time, or if the socket has been closed by some external network error, or if the socket's generation is outdated. Checking sockets lets us avoid seeing *some* :class:`~pymongo.errors.AutoReconnect` exceptions on server hiccups, etc. We only check if the socket was closed by an external error if it has been > 1 second since the socket was checked into the pool to keep performance reasonable - we can't avoid AutoReconnects completely anyway. """ idle_time_seconds = conn.idle_time_seconds() # If socket is idle, open a new one. if ( self.opts.max_idle_time_seconds is not None and idle_time_seconds > self.opts.max_idle_time_seconds ): conn.close_conn(ConnectionClosedReason.IDLE) return True check_interval_seconds = self._check_interval_seconds if check_interval_seconds is not None and ( check_interval_seconds == 0 or idle_time_seconds > check_interval_seconds ): if conn.conn_closed(): conn.close_conn(ConnectionClosedReason.ERROR) return True if self.stale_generation(conn.generation, conn.service_id): conn.close_conn(ConnectionClosedReason.STALE) return True return False def _raise_wait_queue_timeout(self, checkout_started_time: float) -> NoReturn: self._telemetry.checkout_failed( "Wait queue timeout elapsed without a connection becoming available", ConnectionCheckOutFailedReason.TIMEOUT, checkout_started_time, ) timeout = _csot.get_timeout() or self.opts.wait_queue_timeout if self.opts.load_balanced: other_ops = self.active_sockets - self.ncursors - self.ntxns raise WaitQueueTimeoutError( "Timeout waiting for connection from the connection pool. " f"maxPoolSize: {self.opts.max_pool_size}, connections in use by cursors: {self.ncursors}, " f"connections in use by transactions: {self.ntxns}, connections in use " f"by other operations: {other_ops}, timeout: {timeout}" ) raise WaitQueueTimeoutError( "Timed out while checking out a connection from connection pool. " f"maxPoolSize: {self.opts.max_pool_size}, timeout: {timeout}" ) def __del__(self) -> None: # Avoid ResourceWarnings in Python 3 # Close all sockets without calling reset() or close() because it is # not safe to acquire a lock in __del__. if _IS_SYNC: for conn in self.conns: conn.close_conn(None) # type: ignore[unused-coroutine] class _PoolCheckout: """Class-based context manager for pool connection checkout.""" __slots__ = ("_conn", "_handler", "_pool") def __init__( self, pool: Pool, handler: Optional[_ClientCheckout] = None, ) -> None: self._pool = pool self._handler = handler self._conn: Optional[Connection] = None def __enter__(self) -> Connection: pool = self._pool telemetry = pool._telemetry # Fast path: skip telemetry calls when CMAP events/logging are disabled if not telemetry._should_publish and not ( telemetry._log and _is_debug_enabled(_CONNECTION_LOGGER) ): conn = pool._get_conn(time.monotonic(), handler=self._handler) self._conn = conn return conn checkout_started_time = telemetry.checkout_started() conn = pool._get_conn(checkout_started_time, handler=self._handler) self._conn = conn try: telemetry.checkout_succeeded(conn.id, checkout_started_time) except BaseException: pool.checkin(conn) self._conn = None raise return conn def __exit__( self, exc_type: Optional[type[BaseException]], exc_val: Optional[BaseException], exc_tb: Optional[TracebackType], ) -> None: conn = self._conn if conn is None: return pool = self._pool if exc_type is not None: # Exception in caller. Ensure the connection gets returned. # Note that when pinned is True, the session owns the connection # and is responsible for checking it back into the pool. # SDAM error handling is performed by _ClientCheckout.__exit__ # before this method is called. pinned = conn.pinned_txn or conn.pinned_cursor if not pinned and conn.active: pool.checkin(conn) else: if conn.pinned_txn: with pool.lock: pool._pinned_sockets.add(conn) pool.ntxns += 1 elif conn.pinned_cursor: with pool.lock: pool._pinned_sockets.add(conn) pool.ncursors += 1 elif conn.active: pool.checkin(conn)