import asyncio import collections import logging import random import socket import threading import time import warnings import weakref from abc import ABC, abstractmethod from collections import defaultdict from copy import copy from itertools import chain from types import MethodType from typing import ( TYPE_CHECKING, Any, Callable, Coroutine, Deque, Dict, Generator, List, Literal, Mapping, Optional, Set, Tuple, Type, TypeVar, Union, ) if TYPE_CHECKING: from redis.asyncio.keyspace_notifications import ( AsyncClusterKeyspaceNotifications, ) from redis._defaults import ( DEFAULT_RETRY_BASE, DEFAULT_RETRY_CAP, DEFAULT_RETRY_COUNT, DEFAULT_SOCKET_CONNECT_TIMEOUT, DEFAULT_SOCKET_READ_SIZE, DEFAULT_SOCKET_TIMEOUT, ) from redis._parsers import AsyncCommandsParser, Encoder from redis._parsers.commands import CommandPolicies, RequestPolicy, ResponsePolicy from redis._parsers.helpers import get_response_callbacks from redis.asyncio.client import PubSub, ResponseCallbackT from redis.asyncio.connection import ( AbstractConnection, Connection, ConnectionPoolInterface, SSLConnection, parse_url, ) from redis.asyncio.lock import Lock from redis.asyncio.observability.recorder import ( record_error_count, record_operation_duration, ) from redis.asyncio.retry import Retry from redis.auth.token import TokenInterface from redis.backoff import ExponentialWithJitterBackoff, NoBackoff from redis.client import EMPTY_RESPONSE, NEVER_DECODE, AbstractRedis from redis.cluster import ( PIPELINE_BLOCKED_COMMANDS, PRIMARY, REPLICA, SLOT_ID, AbstractRedisCluster, LoadBalancer, LoadBalancingStrategy, block_pipeline_command, get_node_name, parse_cluster_shards, parse_cluster_shards_unified, parse_cluster_shards_with_str_keys, parse_cluster_slots, ) from redis.commands import READ_COMMANDS, AsyncRedisClusterCommands from redis.commands.helpers import list_or_args, parse_pubsub_subscriptions from redis.commands.policies import AsyncPolicyResolver, AsyncStaticPolicyResolver from redis.crc import REDIS_CLUSTER_HASH_SLOTS, key_slot from redis.credentials import CredentialProvider from redis.driver_info import DriverInfo, resolve_driver_info from redis.event import ( AfterAsyncClusterInstantiationEvent, AsyncAfterSlotsCacheRefreshEvent, AsyncEventListenerInterface, EventDispatcher, ) from redis.exceptions import ( AskError, BusyLoadingError, ClusterDownError, ClusterError, ConnectionError, CrossSlotTransactionError, DataError, ExecAbortError, InvalidPipelineStack, MaxConnectionsError, MovedError, RedisClusterException, RedisError, ResponseError, SlotNotCoveredError, TimeoutError, TryAgainError, WatchError, ) from redis.typing import ( AnyKeyT, ChannelT, EncodableT, KeyT, PubSubHandler, Subscription, ) from redis.utils import ( SENTINEL, SSL_AVAILABLE, deprecated_args, deprecated_function, safe_str, str_if_bytes, truncate_text, ) if SSL_AVAILABLE: from ssl import TLSVersion, VerifyFlags, VerifyMode else: TLSVersion = None VerifyMode = None VerifyFlags = None logger = logging.getLogger(__name__) TargetNodesT = TypeVar( "TargetNodesT", str, "ClusterNode", List["ClusterNode"], Dict[Any, "ClusterNode"] ) class RedisCluster(AbstractRedis, AbstractRedisCluster, AsyncRedisClusterCommands): """ Create a new RedisCluster client. Pass one of parameters: - `host` & `port` - `startup_nodes` | Use ``await`` :meth:`initialize` to find cluster nodes & create connections. | Use ``await`` :meth:`close` to disconnect connections & close client. Many commands support the target_nodes kwarg. It can be one of the :attr:`NODE_FLAGS`: - :attr:`PRIMARIES` - :attr:`REPLICAS` - :attr:`ALL_NODES` - :attr:`RANDOM` - :attr:`DEFAULT_NODE` Note: This client is not thread/process/fork safe. :param host: | Can be used to point to a startup node :param port: | Port used if **host** is provided :param startup_nodes: | :class:`~.ClusterNode` to used as a startup node :param require_full_coverage: | When set to ``False``: the client will not require a full coverage of the slots. However, if not all slots are covered, and at least one node has ``cluster-require-full-coverage`` set to ``yes``, the server will throw a :class:`~.ClusterDownError` for some key-based commands. | When set to ``True``: all slots must be covered to construct the cluster client. If not all slots are covered, :class:`~.RedisClusterException` will be thrown. | See: https://redis.io/docs/manual/scaling/#redis-cluster-configuration-parameters :param read_from_replicas: | @deprecated - please use load_balancing_strategy instead | Enable read from replicas in READONLY mode. When set to true, read commands will be assigned between the primary and its replications in a Round-Robin manner. The data read from replicas is eventually consistent with the data in primary nodes. :param load_balancing_strategy: | Enable read from replicas in READONLY mode and defines the load balancing strategy that will be used for cluster node selection. The data read from replicas is eventually consistent with the data in primary nodes. :param dynamic_startup_nodes: | Set the RedisCluster's startup nodes to all the discovered nodes. If true (default value), the cluster's discovered nodes will be used to determine the cluster nodes-slots mapping in the next topology refresh. It will remove the initial passed startup nodes if their endpoints aren't listed in the CLUSTER SLOTS output. If you use dynamic DNS endpoints for startup nodes but CLUSTER SLOTS lists specific IP addresses, it is best to set it to false. :param reinitialize_steps: | Specifies the number of MOVED errors that need to occur before reinitializing the whole cluster topology. If a MOVED error occurs and the cluster does not need to be reinitialized on this current error handling, only the MOVED slot will be patched with the redirected node. To reinitialize the cluster on every MOVED error, set reinitialize_steps to 1. To avoid reinitializing the cluster on moved errors, set reinitialize_steps to 0. :param cluster_error_retry_attempts: | @deprecated - Please configure the 'retry' object instead In case 'retry' object is set - this argument is ignored! Number of times to retry before raising an error when :class:`~.TimeoutError`, :class:`~.ConnectionError`, :class:`~.SlotNotCoveredError` or :class:`~.ClusterDownError` are encountered :param retry: | A retry object that defines the retry strategy and the number of retries for the cluster client. In current implementation for the cluster client (starting form redis-py version 6.0.0) the retry object is not yet fully utilized, instead it is used just to determine the number of retries for the cluster client. In the future releases the retry object will be used to handle the cluster client retries! :param max_connections: | Maximum number of connections per node. If there are no free connections & the maximum number of connections are already created, a :class:`~.MaxConnectionsError` is raised. :param socket_keepalive: | If ``True``, TCP keepalive is enabled for TCP socket connections. :param socket_keepalive_options: | Mapping of TCP keepalive socket option constants to values, for example ``{socket.TCP_KEEPIDLE: 30}``. If left unspecified, redis-py uses TCP keepalive defaults when ``socket_keepalive`` is enabled: idle 30 seconds, interval 5 seconds, and 3 probes. Platform-specific options that are not available are skipped. Pass ``None`` or ``{}`` to avoid setting additional TCP keepalive options. :param address_remap: | An optional callable which, when provided with an internal network address of a node, e.g. a `(host, port)` tuple, will return the address where the node is reachable. This can be used to map the addresses at which the nodes _think_ they are, to addresses at which a client may reach them, such as when they sit behind a proxy. | Rest of the arguments will be passed to the :class:`~redis.asyncio.connection.Connection` instances when created :raises RedisClusterException: if any arguments are invalid or unknown. Eg: - `db` != 0 or None - `path` argument for unix socket connection - none of the `host`/`port` & `startup_nodes` were provided """ @classmethod def from_url(cls, url: str, **kwargs: Any) -> "RedisCluster": """ Return a Redis client object configured from the given URL. For example:: redis://[[username]:[password]]@localhost:6379/0 rediss://[[username]:[password]]@localhost:6379/0 Three URL schemes are supported: - `redis://` creates a TCP socket connection. See more at: - `rediss://` creates a SSL wrapped TCP socket connection. See more at: The username, password, hostname, path and all querystring values are passed through ``urllib.parse.unquote`` in order to replace any percent-encoded values with their corresponding characters. All querystring options are cast to their appropriate Python types. Boolean arguments can be specified with string values "True"/"False" or "Yes"/"No". Values that cannot be properly cast cause a ``ValueError`` to be raised. Once parsed, the querystring arguments and keyword arguments are passed to :class:`~redis.asyncio.connection.Connection` when created. In the case of conflicting arguments, querystring arguments are used. """ kwargs.update(parse_url(url)) if kwargs.pop("connection_class", None) is SSLConnection: kwargs["ssl"] = True return cls(**kwargs) # Type discrimination marker for @overload self-type pattern _is_async_client: Literal[True] = True __slots__ = ( "_initialize", "_lock", "retry", "command_flags", "commands_parser", "connection_kwargs", "encoder", "node_flags", "nodes_manager", "read_from_replicas", "reinitialize_counter", "reinitialize_steps", "response_callbacks", "result_callbacks", ) @deprecated_args( args_to_warn=["read_from_replicas"], reason="Please configure the 'load_balancing_strategy' instead", version="5.3.0", ) @deprecated_args( args_to_warn=[ "cluster_error_retry_attempts", ], reason="Please configure the 'retry' object instead", version="6.0.0", ) @deprecated_args( args_to_warn=["lib_name", "lib_version"], reason="Use 'driver_info' parameter instead. " "lib_name and lib_version will be removed in a future version.", ) def __init__( self, host: str | None = None, port: str | int = 6379, # Cluster related kwargs startup_nodes: List["ClusterNode"] | None = None, require_full_coverage: bool = True, read_from_replicas: bool = False, load_balancing_strategy: LoadBalancingStrategy | None = None, dynamic_startup_nodes: bool = True, reinitialize_steps: int = 5, cluster_error_retry_attempts: int = DEFAULT_RETRY_COUNT, max_connections: int = 100, retry: Retry | None = None, retry_on_error: List[Type[Exception]] | None = None, # Client related kwargs db: str | int = 0, path: str | None = None, credential_provider: CredentialProvider | None = None, username: str | None = None, password: str | None = None, client_name: str | None = None, lib_name: str | object | None = SENTINEL, lib_version: str | object | None = SENTINEL, driver_info: DriverInfo | object | None = SENTINEL, # Encoding related kwargs encoding: str = "utf-8", encoding_errors: str = "strict", decode_responses: bool = False, # Connection related kwargs health_check_interval: float = 0, socket_timeout: float | None = DEFAULT_SOCKET_TIMEOUT, socket_connect_timeout: float | None = DEFAULT_SOCKET_CONNECT_TIMEOUT, socket_read_size: int = DEFAULT_SOCKET_READ_SIZE, socket_keepalive: bool = True, socket_keepalive_options: Mapping[int, int | bytes] | object | None = SENTINEL, # SSL related kwargs ssl: bool = False, ssl_ca_certs: str | None = None, ssl_ca_data: str | None = None, ssl_cert_reqs: "str | VerifyMode" = "required", ssl_include_verify_flags: List["VerifyFlags"] | None = None, ssl_exclude_verify_flags: List["VerifyFlags"] | None = None, ssl_certfile: str | None = None, ssl_check_hostname: bool = True, ssl_keyfile: str | None = None, ssl_min_version: "TLSVersion | None" = None, ssl_ciphers: str | None = None, protocol: int | None = None, legacy_responses: bool = True, address_remap: Callable[[Tuple[str, int]], Tuple[str, int]] | None = None, event_dispatcher: EventDispatcher | None = None, policy_resolver: AsyncPolicyResolver = AsyncStaticPolicyResolver(), ) -> None: if db: raise RedisClusterException( "Argument 'db' must be 0 or None in cluster mode" ) if path: raise RedisClusterException( "Unix domain socket is not supported in cluster mode" ) if (not host or not port) and not startup_nodes: raise RedisClusterException( "RedisCluster requires at least one node to discover the cluster.\n" "Please provide one of the following or use RedisCluster.from_url:\n" ' - host and port: RedisCluster(host="localhost", port=6379)\n' " - startup_nodes: RedisCluster(startup_nodes=[" 'ClusterNode("localhost", 6379), ClusterNode("localhost", 6380)])' ) computed_driver_info = resolve_driver_info(driver_info, lib_name, lib_version) kwargs: Dict[str, Any] = { "max_connections": max_connections, "connection_class": Connection, # Client related kwargs "credential_provider": credential_provider, "username": username, "password": password, "client_name": client_name, "driver_info": computed_driver_info, # Encoding related kwargs "encoding": encoding, "encoding_errors": encoding_errors, "decode_responses": decode_responses, # Connection related kwargs "health_check_interval": health_check_interval, "socket_connect_timeout": socket_connect_timeout, "socket_keepalive": socket_keepalive, "socket_keepalive_options": socket_keepalive_options, "socket_read_size": socket_read_size, "socket_timeout": socket_timeout, "protocol": protocol, "legacy_responses": legacy_responses, } if ssl: # SSL related kwargs kwargs.update( { "connection_class": SSLConnection, "ssl_ca_certs": ssl_ca_certs, "ssl_ca_data": ssl_ca_data, "ssl_cert_reqs": ssl_cert_reqs, "ssl_include_verify_flags": ssl_include_verify_flags, "ssl_exclude_verify_flags": ssl_exclude_verify_flags, "ssl_certfile": ssl_certfile, "ssl_check_hostname": ssl_check_hostname, "ssl_keyfile": ssl_keyfile, "ssl_min_version": ssl_min_version, "ssl_ciphers": ssl_ciphers, } ) if read_from_replicas or load_balancing_strategy: # Call our on_connect function to configure READONLY mode kwargs["redis_connect_func"] = self.on_connect if retry: self.retry = retry else: self.retry = Retry( backoff=ExponentialWithJitterBackoff( base=DEFAULT_RETRY_BASE, cap=DEFAULT_RETRY_CAP ), retries=cluster_error_retry_attempts, ) if retry_on_error: self.retry.update_supported_errors(retry_on_error) kwargs["response_callbacks"] = get_response_callbacks( user_protocol=kwargs.get("protocol"), legacy_responses=kwargs.get("legacy_responses", True), ) if not kwargs.get("legacy_responses", True): kwargs["response_callbacks"]["CLUSTER SHARDS"] = ( parse_cluster_shards_unified ) elif kwargs.get("protocol") is None: kwargs["response_callbacks"]["CLUSTER SHARDS"] = ( parse_cluster_shards_with_str_keys ) else: kwargs["response_callbacks"]["CLUSTER SHARDS"] = parse_cluster_shards self.connection_kwargs = kwargs if startup_nodes: passed_nodes = [] for node in startup_nodes: passed_nodes.append( ClusterNode(node.host, node.port, **self.connection_kwargs) ) startup_nodes = passed_nodes else: startup_nodes = [] if host and port: startup_nodes.append(ClusterNode(host, port, **self.connection_kwargs)) if event_dispatcher is None: self._event_dispatcher = EventDispatcher() else: self._event_dispatcher = event_dispatcher self.startup_nodes = startup_nodes self.nodes_manager = NodesManager( startup_nodes, require_full_coverage, kwargs, dynamic_startup_nodes=dynamic_startup_nodes, address_remap=address_remap, event_dispatcher=self._event_dispatcher, ) self.encoder = Encoder(encoding, encoding_errors, decode_responses) self.read_from_replicas = read_from_replicas self.load_balancing_strategy = load_balancing_strategy self.reinitialize_steps = reinitialize_steps self.reinitialize_counter = 0 # For backward compatibility, mapping from existing policies to new one self._command_flags_mapping: dict[str, Union[RequestPolicy, ResponsePolicy]] = { self.__class__.RANDOM: RequestPolicy.DEFAULT_KEYLESS, self.__class__.PRIMARIES: RequestPolicy.ALL_SHARDS, self.__class__.ALL_NODES: RequestPolicy.ALL_NODES, self.__class__.REPLICAS: RequestPolicy.ALL_REPLICAS, self.__class__.DEFAULT_NODE: RequestPolicy.DEFAULT_NODE, SLOT_ID: RequestPolicy.DEFAULT_KEYED, } self._policies_callback_mapping: dict[ Union[RequestPolicy, ResponsePolicy], Callable ] = { RequestPolicy.DEFAULT_KEYLESS: lambda command_name: [ self.get_random_primary_or_all_nodes(command_name) ], RequestPolicy.DEFAULT_KEYED: self.get_nodes_from_slot, RequestPolicy.DEFAULT_NODE: lambda: [self.get_default_node()], RequestPolicy.ALL_SHARDS: self.get_primaries, RequestPolicy.ALL_NODES: self.get_nodes, RequestPolicy.ALL_REPLICAS: self.get_replicas, RequestPolicy.SPECIAL: self.get_special_nodes, ResponsePolicy.DEFAULT_KEYLESS: lambda res: res, ResponsePolicy.DEFAULT_KEYED: lambda res: res, } self._policy_resolver = policy_resolver self.commands_parser = AsyncCommandsParser() self._aggregate_nodes = None self.node_flags = self.__class__.NODE_FLAGS.copy() self.command_flags = self.__class__.COMMAND_FLAGS.copy() self.response_callbacks = kwargs["response_callbacks"] self.result_callbacks = self.__class__.RESULT_CALLBACKS.copy() self.result_callbacks["CLUSTER SLOTS"] = ( lambda cmd, res, **kwargs: parse_cluster_slots( list(res.values())[0], **kwargs ) ) self._initialize = True self._lock: Optional[asyncio.Lock] = None # When used as an async context manager, we need to increment and decrement # a usage counter so that we can close the connection pool when no one is # using the client. self._usage_counter = 0 self._usage_lock = asyncio.Lock() async def initialize( self, additional_startup_nodes_info: Optional[List[Tuple[str, int]]] = None, last_failed_node_name: Optional[str] = None, ) -> "RedisCluster": """Get all nodes from startup nodes & creates connections if not initialized.""" if self._initialize: if not self._lock: self._lock = asyncio.Lock() async with self._lock: if self._initialize: try: await self.nodes_manager.initialize( additional_startup_nodes_info=additional_startup_nodes_info, last_failed_node_name=last_failed_node_name, ) await self.commands_parser.initialize( self.nodes_manager.default_node ) self._initialize = False except BaseException: await self.nodes_manager.aclose() await self.nodes_manager.aclose("startup_nodes") raise return self async def aclose(self) -> None: """Close all connections & client if initialized.""" if not self._initialize: if not self._lock: self._lock = asyncio.Lock() async with self._lock: if not self._initialize: self._initialize = True await self.nodes_manager.aclose() await self.nodes_manager.aclose("startup_nodes") @deprecated_function(version="5.0.0", reason="Use aclose() instead", name="close") async def close(self) -> None: """alias for aclose() for backwards compatibility""" await self.aclose() async def __aenter__(self) -> "RedisCluster": """ Async context manager entry. Increments a usage counter so that the connection pool is only closed (via aclose()) when no context is using the client. """ await self._increment_usage() try: # Initialize the client (i.e. establish connection, etc.) return await self.initialize() except Exception: # If initialization fails, decrement the counter to keep it in sync await self._decrement_usage() raise async def _increment_usage(self) -> int: """ Helper coroutine to increment the usage counter while holding the lock. Returns the new value of the usage counter. """ async with self._usage_lock: self._usage_counter += 1 return self._usage_counter async def _decrement_usage(self) -> int: """ Helper coroutine to decrement the usage counter while holding the lock. Returns the new value of the usage counter. """ async with self._usage_lock: self._usage_counter -= 1 return self._usage_counter async def __aexit__(self, exc_type, exc_value, traceback): """ Async context manager exit. Decrements a usage counter. If this is the last exit (counter becomes zero), the client closes its connection pool. """ current_usage = await asyncio.shield(self._decrement_usage()) if current_usage == 0: # This was the last active context, so disconnect the pool. await asyncio.shield(self.aclose()) def __await__(self) -> Generator[Any, None, "RedisCluster"]: return self.initialize().__await__() _DEL_MESSAGE = "Unclosed RedisCluster client" def __del__( self, _warn: Any = warnings.warn, _grl: Any = asyncio.get_running_loop, ) -> None: if hasattr(self, "_initialize") and not self._initialize: _warn(f"{self._DEL_MESSAGE} {self!r}", ResourceWarning, source=self) try: context = {"client": self, "message": self._DEL_MESSAGE} _grl().call_exception_handler(context) except RuntimeError: pass async def on_connect(self, connection: Connection) -> None: await connection.on_connect() # Sending READONLY command to server to configure connection as # readonly. Since each cluster node may change its server type due # to a failover, we should establish a READONLY connection # regardless of the server type. If this is a primary connection, # READONLY would not affect executing write commands. await connection.send_command("READONLY") if str_if_bytes(await connection.read_response()) != "OK": raise ConnectionError("READONLY command failed") def get_nodes(self) -> List["ClusterNode"]: """Get all nodes of the cluster.""" return list(self.nodes_manager.nodes_cache.values()) def get_primaries(self) -> List["ClusterNode"]: """Get the primary nodes of the cluster.""" return self.nodes_manager.get_nodes_by_server_type(PRIMARY) def get_replicas(self) -> List["ClusterNode"]: """Get the replica nodes of the cluster.""" return self.nodes_manager.get_nodes_by_server_type(REPLICA) def get_random_node(self) -> "ClusterNode": """Get a random node of the cluster.""" return random.choice(list(self.nodes_manager.nodes_cache.values())) def get_default_node(self) -> "ClusterNode": """Get the default node of the client.""" return self.nodes_manager.default_node def set_default_node(self, node: "ClusterNode") -> None: """ Set the default node of the client. :raises DataError: if None is passed or node does not exist in cluster. """ if not node or not self.get_node(node_name=node.name): raise DataError("The requested node does not exist in the cluster.") self.nodes_manager.default_node = node def get_node( self, host: Optional[str] = None, port: Optional[int] = None, node_name: Optional[str] = None, ) -> Optional["ClusterNode"]: """Get node by (host, port) or node_name.""" return self.nodes_manager.get_node(host, port, node_name) def get_node_from_key( self, key: str, replica: bool = False ) -> Optional["ClusterNode"]: """ Get the cluster node corresponding to the provided key. :param key: :param replica: | Indicates if a replica should be returned | None will returned if no replica holds this key :raises SlotNotCoveredError: if the key is not covered by any slot. """ slot = self.keyslot(key) slot_cache = self.nodes_manager.slots_cache.get(slot) if not slot_cache: raise SlotNotCoveredError(f'Slot "{slot}" is not covered by the cluster.') if replica: if len(self.nodes_manager.slots_cache[slot]) < 2: return None node_idx = 1 else: node_idx = 0 return slot_cache[node_idx] def get_random_primary_or_all_nodes(self, command_name): """ Returns random primary or all nodes depends on READONLY mode. """ if self.read_from_replicas and command_name in READ_COMMANDS: return self.get_random_node() return self.get_random_primary_node() def get_random_primary_node(self) -> "ClusterNode": """ Returns a random primary node """ return random.choice(self.get_primaries()) async def get_nodes_from_slot(self, command: str, *args): """ Returns a list of nodes that hold the specified keys' slots. """ # get the node that holds the key's slot return [ self.nodes_manager.get_node_from_slot( await self._determine_slot(command, *args), self.read_from_replicas and command in READ_COMMANDS, self.load_balancing_strategy if command in READ_COMMANDS else None, ) ] def get_special_nodes(self) -> Optional[list["ClusterNode"]]: """ Returns a list of nodes for commands with a special policy. """ if not self._aggregate_nodes: raise RedisClusterException( "Cannot execute FT.CURSOR commands without FT.AGGREGATE" ) return self._aggregate_nodes def keyslot(self, key: EncodableT) -> int: """ Find the keyslot for a given key. See: https://redis.io/docs/manual/scaling/#redis-cluster-data-sharding """ return key_slot(self.encoder.encode(key)) def get_encoder(self) -> Encoder: """Get the encoder object of the client.""" return self.encoder def get_connection_kwargs(self) -> Dict[str, Optional[Any]]: """Get the kwargs passed to :class:`~redis.asyncio.connection.Connection`.""" return self.connection_kwargs def set_retry(self, retry: Retry) -> None: self.retry = retry def set_response_callback(self, command: str, callback: ResponseCallbackT) -> None: """Set a custom response callback.""" self.response_callbacks[command] = callback async def _determine_nodes( self, command: str, *args: Any, request_policy: RequestPolicy, node_flag: Optional[str] = None, ) -> List["ClusterNode"]: # Determine which nodes should be executed the command on. # Returns a list of target nodes. if not node_flag: # get the nodes group for this command if it was predefined node_flag = self.command_flags.get(command) if node_flag in self._command_flags_mapping: request_policy = self._command_flags_mapping[node_flag] policy_callback = self._policies_callback_mapping[request_policy] if request_policy == RequestPolicy.DEFAULT_KEYED: nodes = await policy_callback(command, *args) elif request_policy == RequestPolicy.DEFAULT_KEYLESS: nodes = policy_callback(command) else: nodes = policy_callback() if command.lower() == "ft.aggregate": self._aggregate_nodes = nodes return nodes async def _determine_slot(self, command: str, *args: Any) -> int: if self.command_flags.get(command) == SLOT_ID: # The command contains the slot ID return int(args[0]) # Get the keys in the command # EVAL and EVALSHA are common enough that it's wasteful to go to the # redis server to parse the keys. Besides, there is a bug in redis<7.0 # where `self._get_command_keys()` fails anyway. So, we special case # EVAL/EVALSHA. # - issue: https://github.com/redis/redis/issues/9493 # - fix: https://github.com/redis/redis/pull/9733 if command.upper() in ("EVAL", "EVALSHA"): # command syntax: EVAL "script body" num_keys ... if len(args) < 2: raise RedisClusterException( f"Invalid args in command: {command, *args}" ) keys = args[2 : 2 + int(args[1])] # if there are 0 keys, that means the script can be run on any node # so we can just return a random slot if not keys: return random.randrange(0, REDIS_CLUSTER_HASH_SLOTS) else: keys = await self.commands_parser.get_keys(command, *args) if not keys: # FCALL can call a function with 0 keys, that means the function # can be run on any node so we can just return a random slot if command.upper() in ("FCALL", "FCALL_RO"): return random.randrange(0, REDIS_CLUSTER_HASH_SLOTS) raise RedisClusterException( "No way to dispatch this command to Redis Cluster. " "Missing key.\nYou can execute the command by specifying " f"target nodes.\nCommand: {args}" ) # single key command if len(keys) == 1: return self.keyslot(keys[0]) # multi-key command; we need to make sure all keys are mapped to # the same slot slots = {self.keyslot(key) for key in keys} if len(slots) != 1: raise RedisClusterException( f"{command} - all keys must map to the same key slot" ) return slots.pop() def _is_node_flag(self, target_nodes: Any) -> bool: return isinstance(target_nodes, str) and target_nodes in self.node_flags def _parse_target_nodes(self, target_nodes: Any) -> List["ClusterNode"]: if isinstance(target_nodes, list): nodes = target_nodes elif isinstance(target_nodes, ClusterNode): # Supports passing a single ClusterNode as a variable nodes = [target_nodes] elif isinstance(target_nodes, dict): # Supports dictionaries of the format {node_name: node}. # It enables to execute commands with multi nodes as follows: # rc.cluster_save_config(rc.get_primaries()) nodes = list(target_nodes.values()) else: raise TypeError( "target_nodes type can be one of the following: " "node_flag (PRIMARIES, REPLICAS, RANDOM, ALL_NODES)," "ClusterNode, list, or dict. " f"The passed type is {type(target_nodes)}" ) return nodes async def _record_error_metric( self, error: Exception, connection: Union[Connection, "ClusterNode"], is_internal: bool = True, retry_attempts: Optional[int] = None, ): """ Records error count metric directly. Accepts either a Connection or ClusterNode object. """ await record_error_count( server_address=connection.host, server_port=connection.port, network_peer_address=connection.host, network_peer_port=connection.port, error_type=error, retry_attempts=retry_attempts if retry_attempts is not None else 0, is_internal=is_internal, ) async def _record_command_metric( self, command_name: str, duration_seconds: float, connection: Union[Connection, "ClusterNode"], error: Optional[Exception] = None, ): """ Records operation duration metric directly. Accepts either a Connection or ClusterNode object. """ # Connection has db attribute, ClusterNode has connection_kwargs if hasattr(connection, "db"): db = connection.db else: db = connection.connection_kwargs.get("db", 0) await record_operation_duration( command_name=command_name, duration_seconds=duration_seconds, server_address=connection.host, server_port=connection.port, db_namespace=str(db) if db is not None else None, error=error, ) async def execute_command(self, *args: EncodableT, **kwargs: Any) -> Any: """ Execute a raw command on the appropriate cluster node or target_nodes. It will retry the command as specified by the retries property of the :attr:`retry` & then raise an exception. :param args: | Raw command args :param kwargs: - target_nodes: :attr:`NODE_FLAGS` or :class:`~.ClusterNode` or List[:class:`~.ClusterNode`] or Dict[Any, :class:`~.ClusterNode`] - Rest of the kwargs are passed to the Redis connection :raises RedisClusterException: if target_nodes is not provided & the command can't be mapped to a slot """ command = args[0] target_nodes = [] target_nodes_specified = False retry_attempts = self.retry.get_retries() passed_targets = kwargs.pop("target_nodes", None) if passed_targets and not self._is_node_flag(passed_targets): target_nodes = self._parse_target_nodes(passed_targets) target_nodes_specified = True retry_attempts = 0 command_policies = await self._policy_resolver.resolve(args[0].lower()) if not command_policies and not target_nodes_specified: command_flag = self.command_flags.get(command) if not command_flag: # Fallback to default policy if not self.get_default_node(): slot = None else: slot = await self._determine_slot(*args) if slot is None: command_policies = CommandPolicies() else: command_policies = CommandPolicies( request_policy=RequestPolicy.DEFAULT_KEYED, response_policy=ResponsePolicy.DEFAULT_KEYED, ) else: if command_flag in self._command_flags_mapping: command_policies = CommandPolicies( request_policy=self._command_flags_mapping[command_flag] ) else: command_policies = CommandPolicies() elif not command_policies and target_nodes_specified: command_policies = CommandPolicies() # Add one for the first execution execute_attempts = 1 + retry_attempts failure_count = 0 # Start timing for observability start_time = time.monotonic() last_failed_node_name = None for _ in range(execute_attempts): if self._initialize: await self.initialize(last_failed_node_name=last_failed_node_name) last_failed_node_name = None if ( len(target_nodes) == 1 and target_nodes[0] == self.get_default_node() ): # Replace the default cluster node self.replace_default_node() try: if not target_nodes_specified: # Determine the nodes to execute the command on target_nodes = await self._determine_nodes( *args, request_policy=command_policies.request_policy, node_flag=passed_targets, ) if not target_nodes: raise RedisClusterException( f"No targets were found to execute {args} command on" ) if len(target_nodes) == 1: # Return the processed result ret = await self._execute_command(target_nodes[0], *args, **kwargs) if command in self.result_callbacks: ret = self.result_callbacks[command]( command, {target_nodes[0].name: ret}, **kwargs ) return self._policies_callback_mapping[ command_policies.response_policy ](ret) else: keys = [node.name for node in target_nodes] values = await asyncio.gather( *( asyncio.create_task( self._execute_command(node, *args, **kwargs) ) for node in target_nodes ) ) if command in self.result_callbacks: return self.result_callbacks[command]( command, dict(zip(keys, values)), **kwargs ) return self._policies_callback_mapping[ command_policies.response_policy ](dict(zip(keys, values))) except Exception as e: if retry_attempts > 0 and type(e) in self.__class__.ERRORS_ALLOW_RETRY: # The nodes and slots cache were should be reinitialized. # Try again with the new cluster setup. retry_attempts -= 1 failure_count += 1 last_failed_node_name = getattr(e, "last_failed_node_name", None) if hasattr(e, "connection"): await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=e.connection, error=e, ) await self._record_error_metric( error=e, connection=e.connection, retry_attempts=failure_count, ) continue else: # raise the exception if hasattr(e, "connection"): await self._record_error_metric( error=e, connection=e.connection, retry_attempts=failure_count, is_internal=False, ) raise e async def _execute_command( self, target_node: "ClusterNode", *args: Union[KeyT, EncodableT], **kwargs: Any ) -> Any: asking = moved = False redirect_addr = None ttl = self.RedisClusterRequestTTL command = args[0] start_time = time.monotonic() while ttl > 0: ttl -= 1 try: if asking: target_node = self.get_node(node_name=redirect_addr) await target_node.execute_command("ASKING") asking = False elif moved: # MOVED occurred and the slots cache was updated, # refresh the target node slot = await self._determine_slot(*args) target_node = self.nodes_manager.get_node_from_slot( slot, self.read_from_replicas and args[0] in READ_COMMANDS, self.load_balancing_strategy if args[0] in READ_COMMANDS else None, ) moved = False response = await target_node.execute_command(*args, **kwargs) await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, ) return response except BusyLoadingError as e: e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise except MaxConnectionsError as e: # MaxConnectionsError indicates client-side resource exhaustion # (too many connections in the pool), not a node failure. # Don't treat this as a node failure - just re-raise the error # without reinitializing the cluster. e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise except (ConnectionError, TimeoutError) as e: # Connection retries are being handled in the node's # Retry object. # Mark active connections for reconnect and disconnect free ones # This handles connection state (like READONLY) that may be stale target_node.update_active_connections_for_reconnect() await target_node.disconnect_free_connections() # Move the failed node to the end of the cached nodes list # so it's tried last during reinitialization self.nodes_manager.move_node_to_end_of_cached_nodes(target_node.name) e.last_failed_node_name = target_node.name # Signal that reinitialization is needed # The retry loop will handle initialize() AND replace_default_node() self._initialize = True e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise except (ClusterDownError, SlotNotCoveredError) as e: # ClusterDownError can occur during a failover and to get # self-healed, we will try to reinitialize the cluster layout # and retry executing the command # SlotNotCoveredError can occur when the cluster is not fully # initialized or can be temporary issue. # We will try to reinitialize the cluster topology # and retry executing the command await self.aclose() await asyncio.sleep(0.25) e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise except MovedError as e: # First, we will try to patch the slots/nodes cache with the # redirected node output and try again. If MovedError exceeds # 'reinitialize_steps' number of times, we will force # reinitializing the tables, and then try again. # 'reinitialize_steps' counter will increase faster when # the same client object is shared between multiple threads. To # reduce the frequency you can set this variable in the # RedisCluster constructor. self.reinitialize_counter += 1 if ( self.reinitialize_steps and self.reinitialize_counter % self.reinitialize_steps == 0 ): await self.aclose() # Reset the counter self.reinitialize_counter = 0 else: await self.nodes_manager.move_slot(e) moved = True await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) await self._record_error_metric( error=e, connection=target_node, ) except AskError as e: redirect_addr = get_node_name(host=e.host, port=e.port) asking = True await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) await self._record_error_metric( error=e, connection=target_node, ) except TryAgainError as e: if ttl < self.RedisClusterRequestTTL / 2: await asyncio.sleep(0.05) await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) await self._record_error_metric( error=e, connection=target_node, ) except ResponseError as e: e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise except Exception as e: e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise e = ClusterError("TTL exhausted.") e.connection = target_node await self._record_command_metric( command_name=command, duration_seconds=time.monotonic() - start_time, connection=target_node, error=e, ) raise e def pipeline( self, transaction: Optional[Any] = None, shard_hint: Optional[Any] = None ) -> "ClusterPipeline": """ Create & return a new :class:`~.ClusterPipeline` object. Cluster implementation of pipeline does not support transaction or shard_hint. :raises RedisClusterException: if transaction or shard_hint are truthy values """ if shard_hint: raise RedisClusterException("shard_hint is deprecated in cluster mode") return ClusterPipeline(self, transaction) def pubsub( self, node: Optional["ClusterNode"] = None, host: Optional[str] = None, port: Optional[int] = None, **kwargs: Any, ) -> "ClusterPubSub": """ Create and return a ClusterPubSub instance. Allows passing a ClusterNode, or host&port, to get a pubsub instance connected to the specified node :param node: ClusterNode to connect to :param host: Host of the node to connect to :param port: Port of the node to connect to :param kwargs: Additional keyword arguments :return: ClusterPubSub instance """ return ClusterPubSub(self, node=node, host=host, port=port, **kwargs) def keyspace_notifications( self, key_prefix: Union[str, bytes, None] = None, ignore_subscribe_messages: bool = True, ) -> "AsyncClusterKeyspaceNotifications": """ Return an :class:`~redis.asyncio.keyspace_notifications.AsyncClusterKeyspaceNotifications` object for subscribing to keyspace and keyevent notifications across all primary nodes in the cluster. Note: Keyspace notifications must be enabled on all Redis cluster nodes via the ``notify-keyspace-events`` configuration option. Args: key_prefix: Optional prefix to filter and strip from keys in notifications. ignore_subscribe_messages: If True, subscribe/unsubscribe confirmations are not returned by get_message/listen. """ from redis.asyncio.keyspace_notifications import ( AsyncClusterKeyspaceNotifications, ) return AsyncClusterKeyspaceNotifications( self, key_prefix=key_prefix, ignore_subscribe_messages=ignore_subscribe_messages, ) def lock( self, name: KeyT, timeout: Optional[float] = None, sleep: float = 0.1, blocking: bool = True, blocking_timeout: Optional[float] = None, lock_class: Optional[Type[Lock]] = None, thread_local: bool = True, raise_on_release_error: bool = True, ) -> Lock: """ Return a new Lock object using key ``name`` that mimics the behavior of threading.Lock. If specified, ``timeout`` indicates a maximum life for the lock. By default, it will remain locked until release() is called. ``sleep`` indicates the amount of time to sleep per loop iteration when the lock is in blocking mode and another client is currently holding the lock. ``blocking`` indicates whether calling ``acquire`` should block until the lock has been acquired or to fail immediately, causing ``acquire`` to return False and the lock not being acquired. Defaults to True. Note this value can be overridden by passing a ``blocking`` argument to ``acquire``. ``blocking_timeout`` indicates the maximum amount of time in seconds to spend trying to acquire the lock. A value of ``None`` indicates continue trying forever. ``blocking_timeout`` can be specified as a float or integer, both representing the number of seconds to wait. ``lock_class`` forces the specified lock implementation. Note that as of redis-py 3.0, the only lock class we implement is ``Lock`` (which is a Lua-based lock). So, it's unlikely you'll need this parameter, unless you have created your own custom lock class. ``thread_local`` indicates whether the lock token is placed in thread-local storage. By default, the token is placed in thread local storage so that a thread only sees its token, not a token set by another thread. Consider the following timeline: time: 0, thread-1 acquires `my-lock`, with a timeout of 5 seconds. thread-1 sets the token to "abc" time: 1, thread-2 blocks trying to acquire `my-lock` using the Lock instance. time: 5, thread-1 has not yet completed. redis expires the lock key. time: 5, thread-2 acquired `my-lock` now that it's available. thread-2 sets the token to "xyz" time: 6, thread-1 finishes its work and calls release(). if the token is *not* stored in thread local storage, then thread-1 would see the token value as "xyz" and would be able to successfully release the thread-2's lock. ``raise_on_release_error`` indicates whether to raise an exception when the lock is no longer owned when exiting the context manager. By default, this is True, meaning an exception will be raised. If False, the warning will be logged and the exception will be suppressed. In some use cases it's necessary to disable thread local storage. For example, if you have code where one thread acquires a lock and passes that lock instance to a worker thread to release later. If thread local storage isn't disabled in this case, the worker thread won't see the token set by the thread that acquired the lock. Our assumption is that these cases aren't common and as such default to using thread local storage.""" if lock_class is None: lock_class = Lock return lock_class( self, name, timeout=timeout, sleep=sleep, blocking=blocking, blocking_timeout=blocking_timeout, thread_local=thread_local, raise_on_release_error=raise_on_release_error, ) async def transaction( self, func: Coroutine[None, "ClusterPipeline", Any], *watches, **kwargs ): """ Convenience method for executing the callable `func` as a transaction while watching all keys specified in `watches`. The 'func' callable should expect a single argument which is a Pipeline object. """ shard_hint = kwargs.pop("shard_hint", None) value_from_callable = kwargs.pop("value_from_callable", False) watch_delay = kwargs.pop("watch_delay", None) async with self.pipeline(True, shard_hint) as pipe: while True: try: if watches: await pipe.watch(*watches) func_value = await func(pipe) exec_value = await pipe.execute() return func_value if value_from_callable else exec_value except WatchError: if watch_delay is not None and watch_delay > 0: time.sleep(watch_delay) continue class ClusterNode: """ Create a new ClusterNode. Each ClusterNode manages multiple :class:`~redis.asyncio.connection.Connection` objects for the (host, port). """ __slots__ = ( "_background_tasks", "_connections", "_free", "_lock", "_event_dispatcher", "connection_class", "connection_kwargs", "host", "max_connections", "name", "port", "response_callbacks", "server_type", ) def __init__( self, host: str, port: Union[str, int], server_type: Optional[str] = None, *, max_connections: int = 100, connection_class: Type[Connection] = Connection, **connection_kwargs: Any, ) -> None: if host == "localhost": host = socket.gethostbyname(host) connection_kwargs["host"] = host connection_kwargs["port"] = port self.host = host self.port = port self.name = get_node_name(host, port) self.server_type = server_type self.max_connections = max_connections self.connection_class = connection_class self.connection_kwargs = connection_kwargs self.response_callbacks = connection_kwargs.pop("response_callbacks", {}) self._connections: List[Connection] = [] self._free: Deque[Connection] = collections.deque(maxlen=self.max_connections) self._background_tasks: Set[asyncio.Task] = set() self._event_dispatcher = self.connection_kwargs.get("event_dispatcher", None) if self._event_dispatcher is None: self._event_dispatcher = EventDispatcher() def __repr__(self) -> str: return ( f"[host={self.host}, port={self.port}, " f"name={self.name}, server_type={self.server_type}]" ) def __eq__(self, obj: Any) -> bool: return isinstance(obj, ClusterNode) and obj.name == self.name def __hash__(self) -> int: return hash(self.name) _DEL_MESSAGE = "Unclosed ClusterNode object" def __del__( self, _warn: Any = warnings.warn, _grl: Any = asyncio.get_running_loop, ) -> None: for connection in self._connections: if connection.is_connected: _warn(f"{self._DEL_MESSAGE} {self!r}", ResourceWarning, source=self) try: context = {"client": self, "message": self._DEL_MESSAGE} _grl().call_exception_handler(context) except RuntimeError: pass break async def disconnect(self) -> None: ret = await asyncio.gather( *( asyncio.create_task(connection.disconnect()) for connection in self._connections ), return_exceptions=True, ) exc = next((res for res in ret if isinstance(res, Exception)), None) if exc: raise exc def acquire_connection(self) -> Connection: try: return self._free.popleft() except IndexError: if len(self._connections) < self.max_connections: # We are configuring the connection pool not to retry # connections on lower level clients to avoid retrying # connections to nodes that are not reachable # and to avoid blocking the connection pool. # The only error that will have some handling in the lower # level clients is ConnectionError which will trigger disconnection # of the socket. # The retries will be handled on cluster client level # where we will have proper handling of the cluster topology retry = Retry( backoff=NoBackoff(), retries=0, supported_errors=(ConnectionError,), ) connection_kwargs = self.connection_kwargs.copy() connection_kwargs["retry"] = retry connection = self.connection_class(**connection_kwargs) self._connections.append(connection) return connection raise MaxConnectionsError() async def disconnect_if_needed(self, connection: Connection) -> None: """ Disconnect a connection if it's marked for reconnect. This implements lazy disconnection to avoid race conditions. The connection will auto-reconnect on next use. """ if connection.should_reconnect(): await connection.disconnect() def release(self, connection: Connection) -> None: """ Release connection back to free queue. If the connection is marked for reconnect, disconnect it before returning it to the free queue. """ if connection.should_reconnect(): task = asyncio.create_task(self._disconnect_and_release(connection)) self._background_tasks.add(task) task.add_done_callback(self._background_tasks.discard) return self._free.append(connection) async def _disconnect_and_release(self, connection: Connection) -> None: try: await connection.disconnect() except Exception as exc: logger.debug( "disconnecting released cluster connection failed: %r", exc, exc_info=True, ) try: self._connections.remove(connection) except ValueError: pass return self._free.append(connection) def get_encoder(self) -> Encoder: """Return an :class:`Encoder` derived from this node's connection kwargs.""" kwargs = self.connection_kwargs encoder_class = kwargs.get("encoder_class", Encoder) return encoder_class( encoding=kwargs.get("encoding", "utf-8"), encoding_errors=kwargs.get("encoding_errors", "strict"), decode_responses=kwargs.get("decode_responses", False), ) def update_active_connections_for_reconnect(self) -> None: """ Mark all in-use (active) connections for reconnect. In-use connections are those in _connections but not currently in _free. They will be disconnected after their current operation completes. """ free_set = set(self._free) for connection in self._connections: if connection not in free_set: connection.mark_for_reconnect() async def disconnect_free_connections(self) -> None: """ Disconnect all free/idle connections in the pool. This is useful after topology changes (e.g., failover) to clear stale connection state like READONLY mode. The connections remain in the pool and will reconnect on next use. """ if self._free: # Take a snapshot to avoid issues if _free changes during await await asyncio.gather( *(connection.disconnect() for connection in tuple(self._free)), return_exceptions=True, ) async def parse_response( self, connection: Connection, command: str, **kwargs: Any ) -> Any: try: if NEVER_DECODE in kwargs: response = await connection.read_response(disable_decoding=True) kwargs.pop(NEVER_DECODE) else: response = await connection.read_response() except ResponseError: if EMPTY_RESPONSE in kwargs: return kwargs[EMPTY_RESPONSE] raise if EMPTY_RESPONSE in kwargs: kwargs.pop(EMPTY_RESPONSE) # Remove keys entry, it needs only for cache. kwargs.pop("keys", None) # Return response if command in self.response_callbacks: return self.response_callbacks[command](response, **kwargs) return response async def execute_command(self, *args: Any, **kwargs: Any) -> Any: # Acquire connection connection = self.acquire_connection() try: # Handle lazy disconnect for connections marked for reconnect await self.disconnect_if_needed(connection) # Execute command await connection.send_packed_command(connection.pack_command(*args)) # Read response return await self.parse_response(connection, args[0], **kwargs) finally: try: await self.disconnect_if_needed(connection) finally: # Release connection self.release(connection) async def execute_pipeline(self, commands: List["PipelineCommand"]) -> bool: # Acquire connection connection = self.acquire_connection() try: # Handle lazy disconnect for connections marked for reconnect await self.disconnect_if_needed(connection) # Execute command await connection.send_packed_command( connection.pack_commands(cmd.args for cmd in commands) ) # Read responses ret = False for cmd in commands: try: cmd.result = await self.parse_response( connection, cmd.args[0], **cmd.kwargs ) except Exception as e: cmd.result = e ret = True return ret finally: try: await self.disconnect_if_needed(connection) finally: # Release connection self.release(connection) async def re_auth_callback(self, token: TokenInterface): tmp_queue = collections.deque() while self._free: conn = self._free.popleft() await conn.retry.call_with_retry( lambda: conn.send_command( "AUTH", token.try_get("oid"), token.get_value() ), lambda error: self._mock(error), ) await conn.retry.call_with_retry( lambda: conn.read_response(), lambda error: self._mock(error) ) tmp_queue.append(conn) while tmp_queue: conn = tmp_queue.popleft() self._free.append(conn) async def _mock(self, error: RedisError): """ Dummy functions, needs to be passed as error callback to retry object. :param error: :return: """ pass class NodesManager: __slots__ = ( "_dynamic_startup_nodes", "_event_dispatcher", "_background_tasks", "connection_kwargs", "default_node", "nodes_cache", "_epoch", "read_load_balancer", "_initialize_lock", "require_full_coverage", "slots_cache", "startup_nodes", "address_remap", ) def __init__( self, startup_nodes: List["ClusterNode"], require_full_coverage: bool, connection_kwargs: Dict[str, Any], dynamic_startup_nodes: bool = True, address_remap: Optional[Callable[[Tuple[str, int]], Tuple[str, int]]] = None, event_dispatcher: Optional[EventDispatcher] = None, ) -> None: self.startup_nodes = {node.name: node for node in startup_nodes} self.require_full_coverage = require_full_coverage self.connection_kwargs = connection_kwargs self.address_remap = address_remap self.default_node: "ClusterNode" = None self.nodes_cache: Dict[str, "ClusterNode"] = {} self.slots_cache: Dict[int, List["ClusterNode"]] = {} self._epoch: int = 0 self.read_load_balancer = LoadBalancer() self._initialize_lock: asyncio.Lock = asyncio.Lock() self._background_tasks: Set[asyncio.Task] = set() self._dynamic_startup_nodes: bool = dynamic_startup_nodes if event_dispatcher is None: self._event_dispatcher = EventDispatcher() else: self._event_dispatcher = event_dispatcher def get_node( self, host: Optional[str] = None, port: Optional[int] = None, node_name: Optional[str] = None, ) -> Optional["ClusterNode"]: if host and port: # the user passed host and port if host == "localhost": host = socket.gethostbyname(host) return self.nodes_cache.get(get_node_name(host=host, port=port)) elif node_name: return self.nodes_cache.get(node_name) else: raise DataError( "get_node requires one of the following: 1. node name 2. host and port" ) def set_nodes( self, old: Dict[str, "ClusterNode"], new: Dict[str, "ClusterNode"], remove_old: bool = False, ) -> None: if remove_old: for name in list(old.keys()): if name not in new: # Node is removed from cache before disconnect starts, # so it won't be found in lookups during disconnect # Mark active connections so in-flight commands can # finish, then disconnect them when their current # operation completes. Free connections can be # disconnected immediately. removed_node = old.pop(name) removed_node.update_active_connections_for_reconnect() task = asyncio.create_task( removed_node.disconnect_free_connections() ) self._background_tasks.add(task) task.add_done_callback(self._background_tasks.discard) for name, node in new.items(): if name in old: # Preserve the existing node but mark connections for reconnect. # This method is sync so we can't call disconnect_free_connections() # which is async. Instead, we mark free connections for reconnect # and they will be lazily disconnected when acquired via # disconnect_if_needed() to avoid race conditions. # TODO: Make this method async in the next major release to allow # immediate disconnection of free connections. existing_node = old[name] existing_node.server_type = node.server_type existing_node.update_active_connections_for_reconnect() for conn in existing_node._free: conn.mark_for_reconnect() continue # New node is detected and should be added to the pool old[name] = node def move_node_to_end_of_cached_nodes(self, node_name: str) -> None: """ Move a failing node to the end of startup_nodes and nodes_cache so it's tried last during reinitialization and when selecting the default node. If the node is not in the respective list, nothing is done. """ # Move in startup_nodes if node_name in self.startup_nodes and len(self.startup_nodes) > 1: node = self.startup_nodes.pop(node_name) self.startup_nodes[node_name] = node # Re-insert at end # Move in nodes_cache - this affects get_nodes_by_server_type ordering # which is used to select the default_node during initialize() if node_name in self.nodes_cache and len(self.nodes_cache) > 1: node = self.nodes_cache.pop(node_name) self.nodes_cache[node_name] = node # Re-insert at end async def move_slot(self, e: AskError | MovedError): node_changed = False redirected_node = self.get_node(host=e.host, port=e.port) if redirected_node: # The node already exists if redirected_node.server_type != PRIMARY: # Update the node's server type redirected_node.server_type = PRIMARY else: # This is a new node, we will add it to the nodes cache redirected_node = ClusterNode( e.host, e.port, PRIMARY, **self.connection_kwargs ) self.set_nodes(self.nodes_cache, {redirected_node.name: redirected_node}) slot_nodes = self.slots_cache[e.slot_id] if redirected_node not in slot_nodes: # The new slot owner is a new server, or a server from a different # shard. We need to remove all current nodes from the slot's list # (including replications) and add just the new node. self.slots_cache[e.slot_id] = [redirected_node] node_changed = True elif redirected_node is not slot_nodes[0]: # The MOVED error resulted from a failover, and the new slot owner # had previously been a replica. old_primary = slot_nodes[0] # Update the old primary to be a replica and add it to the end of # the slot's node list old_primary.server_type = REPLICA slot_nodes.append(old_primary) # Remove the old replica, which is now a primary, from the slot's # node list slot_nodes.remove(redirected_node) # Override the old primary with the new one slot_nodes[0] = redirected_node if self.default_node == old_primary: # Update the default node with the new primary self.default_node = redirected_node node_changed = True # else: circular MOVED to current primary -> no-op # Dispatch so listeners can run shard-pubsub reconciliation; skipped on # the no-op branch to avoid needless walks under MOVED storms. A # listener must not break slots-cache refresh; log and continue so a # single buggy listener cannot starve the rest. if node_changed: try: await self._event_dispatcher.dispatch_async( AsyncAfterSlotsCacheRefreshEvent() ) except Exception as exc: # Don't shadow the method parameter ``e``: ``except as`` binds # the listener exception in the function scope and ``del``s # the name on block exit (PEP 3134), which would also wipe # out the original AskError/MovedError parameter. logger.exception( "listener raised during slots-cache refresh: %s: %s", type(exc).__name__, exc, ) def get_node_from_slot( self, slot: int, read_from_replicas: bool = False, load_balancing_strategy=None, ) -> "ClusterNode": if read_from_replicas is True and load_balancing_strategy is None: load_balancing_strategy = LoadBalancingStrategy.ROUND_ROBIN try: if len(self.slots_cache[slot]) > 1 and load_balancing_strategy: # get the server index using the strategy defined in load_balancing_strategy primary_name = self.slots_cache[slot][0].name node_idx = self.read_load_balancer.get_server_index( primary_name, len(self.slots_cache[slot]), load_balancing_strategy ) return self.slots_cache[slot][node_idx] return self.slots_cache[slot][0] except (IndexError, TypeError): raise SlotNotCoveredError( f'Slot "{slot}" not covered by the cluster. ' f'"require_full_coverage={self.require_full_coverage}"' ) def get_nodes_by_server_type(self, server_type: str) -> List["ClusterNode"]: return [ node for node in self.nodes_cache.values() if node.server_type == server_type ] async def initialize( self, additional_startup_nodes_info: Optional[List[Tuple[str, int]]] = None, last_failed_node_name: Optional[str] = None, ) -> None: self.read_load_balancer.reset() tmp_nodes_cache: Dict[str, "ClusterNode"] = {} tmp_slots: Dict[int, List["ClusterNode"]] = {} disagreements = [] startup_nodes_reachable = False fully_covered = False exception = None epoch = self._epoch if additional_startup_nodes_info is None: additional_startup_nodes_info = [] async with self._initialize_lock: if self._epoch != epoch: # another initialize call has already reinitialized the # nodes since we started waiting for the lock; # we don't need to do it again. return # Copy to a list to prevent RuntimeError if self.startup_nodes # is modified during iteration, then shuffle the iteration order. startup_nodes = list(self.startup_nodes.values()) deferred_failed_nodes = [] if last_failed_node_name is not None: for index, node in enumerate(startup_nodes): if node.name == last_failed_node_name: deferred_failed_nodes.append(startup_nodes.pop(index)) break if len(startup_nodes) > 1: # Vary which startup node is queried first so clients do not # all reinitialize through the same node. random.shuffle(startup_nodes) additional_startup_nodes = [ ClusterNode(host, port, **self.connection_kwargs) for host, port in additional_startup_nodes_info ] if last_failed_node_name is not None: for index, node in enumerate(additional_startup_nodes): if node.name == last_failed_node_name: if not deferred_failed_nodes: deferred_failed_nodes.append(node) additional_startup_nodes.pop(index) break for startup_node in chain( startup_nodes, additional_startup_nodes, deferred_failed_nodes, ): try: # Make sure cluster mode is enabled on this node try: self._event_dispatcher.dispatch( AfterAsyncClusterInstantiationEvent( self.nodes_cache, self.connection_kwargs.get("credential_provider", None), ) ) cluster_slots = await startup_node.execute_command( "CLUSTER SLOTS" ) except ResponseError: raise RedisClusterException( "Cluster mode is not enabled on this node" ) startup_nodes_reachable = True except Exception as e: # Try the next startup node. # The exception is saved and raised only if we have no more nodes. exception = e continue # CLUSTER SLOTS command results in the following output: # [[slot_section[from_slot,to_slot,master,replica1,...,replicaN]]] # where each node contains the following list: [IP, port, node_id] # Therefore, cluster_slots[0][2][0] will be the IP address of the # primary node of the first slot section. # If there's only one server in the cluster, its ``host`` is '' # Fix it to the host in startup_nodes if ( len(cluster_slots) == 1 and not cluster_slots[0][2][0] and len(self.startup_nodes) == 1 ): cluster_slots[0][2][0] = startup_node.host for slot in cluster_slots: for i in range(2, len(slot)): slot[i] = [str_if_bytes(val) for val in slot[i]] primary_node = slot[2] host = primary_node[0] if host == "": host = startup_node.host port = int(primary_node[1]) host, port = self.remap_host_port(host, port) nodes_for_slot = [] target_node = tmp_nodes_cache.get(get_node_name(host, port)) if not target_node: target_node = ClusterNode( host, port, PRIMARY, **self.connection_kwargs ) # add this node to the nodes cache tmp_nodes_cache[target_node.name] = target_node nodes_for_slot.append(target_node) replica_nodes = slot[3:] for replica_node in replica_nodes: host = replica_node[0] port = replica_node[1] host, port = self.remap_host_port(host, port) target_replica_node = tmp_nodes_cache.get( get_node_name(host, port) ) if not target_replica_node: target_replica_node = ClusterNode( host, port, REPLICA, **self.connection_kwargs ) # add this node to the nodes cache tmp_nodes_cache[target_replica_node.name] = target_replica_node nodes_for_slot.append(target_replica_node) for i in range(int(slot[0]), int(slot[1]) + 1): if i not in tmp_slots: tmp_slots[i] = nodes_for_slot else: # Validate that 2 nodes want to use the same slot cache # setup tmp_slot = tmp_slots[i][0] if tmp_slot.name != target_node.name: disagreements.append( f"{tmp_slot.name} vs {target_node.name} on slot: {i}" ) if len(disagreements) > 5: raise RedisClusterException( f"startup_nodes could not agree on a valid " f"slots cache: {', '.join(disagreements)}" ) # Validate if all slots are covered or if we should try next startup node fully_covered = True for i in range(REDIS_CLUSTER_HASH_SLOTS): if i not in tmp_slots: fully_covered = False break if fully_covered: break if not startup_nodes_reachable: raise RedisClusterException( f"Redis Cluster cannot be connected. Please provide at least " f"one reachable node: {str(exception)}" ) from exception # Check if the slots are not fully covered if not fully_covered and self.require_full_coverage: # Despite the requirement that the slots be covered, there # isn't a full coverage raise RedisClusterException( f"All slots are not covered after query all startup_nodes. " f"{len(tmp_slots)} of {REDIS_CLUSTER_HASH_SLOTS} " f"covered..." ) # Set the tmp variables to the real variables self.set_nodes(self.nodes_cache, tmp_nodes_cache, remove_old=True) # tmp_slots was built from CLUSTER SLOTS responses and can contain # newly-created ClusterNode objects for nodes we already know about. # Rebuild the slots cache with the preserved nodes_cache instances # so existing per-node connection pools stay in use after refresh. # Keep the shared node-list-per-slot-range shape from tmp_slots to # avoid allocating a separate list for every slot. node_lists_by_id: Dict[int, List["ClusterNode"]] = {} new_slots_cache: Dict[int, List["ClusterNode"]] = {} for slot, nodes in tmp_slots.items(): node_list_id = id(nodes) slot_nodes = node_lists_by_id.get(node_list_id) if slot_nodes is None: slot_nodes = [self.nodes_cache[node.name] for node in nodes] node_lists_by_id[node_list_id] = slot_nodes new_slots_cache[slot] = slot_nodes self.slots_cache = new_slots_cache if self._dynamic_startup_nodes: # Populate the startup nodes with all discovered nodes self.set_nodes(self.startup_nodes, self.nodes_cache, remove_old=True) # Set the default node self.default_node = self.get_nodes_by_server_type(PRIMARY)[0] self._epoch += 1 # Dispatch so listeners (e.g. ClusterPubSub) can reconcile per-node # state after slot ownership may have changed. A listener must not # break slots-cache refresh; log and continue so a single buggy # listener cannot starve the rest. try: await self._event_dispatcher.dispatch_async( AsyncAfterSlotsCacheRefreshEvent() ) except Exception as e: logger.exception( "listener raised during slots-cache refresh: %s: %s", type(e).__name__, e, ) async def aclose(self, attr: str = "nodes_cache") -> None: self.default_node = None await asyncio.gather( *( asyncio.create_task(node.disconnect()) for node in getattr(self, attr).values() ) ) def remap_host_port(self, host: str, port: int) -> Tuple[str, int]: """ Remap the host and port returned from the cluster to a different internal value. Useful if the client is not connecting directly to the cluster. """ if self.address_remap: return self.address_remap((host, port)) return host, port class ClusterPipeline(AbstractRedis, AbstractRedisCluster, AsyncRedisClusterCommands): """ Create a new ClusterPipeline object. Usage:: result = await ( rc.pipeline() .set("A", 1) .get("A") .hset("K", "F", "V") .hgetall("K") .mset_nonatomic({"A": 2, "B": 3}) .get("A") .get("B") .delete("A", "B", "K") .execute() ) # result = [True, "1", 1, {"F": "V"}, True, True, "2", "3", 1, 1, 1] Note: For commands `DELETE`, `EXISTS`, `TOUCH`, `UNLINK`, `mset_nonatomic`, which are split across multiple nodes, you'll get multiple results for them in the array. Retryable errors: - :class:`~.ClusterDownError` - :class:`~.ConnectionError` - :class:`~.TimeoutError` Redirection errors: - :class:`~.TryAgainError` - :class:`~.MovedError` - :class:`~.AskError` :param client: | Existing :class:`~.RedisCluster` client """ __slots__ = ( "cluster_client", "_transaction", "_execution_strategy", ) # Type discrimination marker for @overload self-type pattern _is_async_client: Literal[True] = True def __init__( self, client: RedisCluster, transaction: Optional[bool] = None ) -> None: self.cluster_client = client self._transaction = transaction self._execution_strategy: ExecutionStrategy = ( PipelineStrategy(self) if not self._transaction else TransactionStrategy(self) ) @property def nodes_manager(self) -> "NodesManager": """Get the nodes manager from the cluster client.""" return self.cluster_client.nodes_manager def set_response_callback(self, command: str, callback: ResponseCallbackT) -> None: """Set a custom response callback on the cluster client.""" self.cluster_client.set_response_callback(command, callback) async def initialize(self) -> "ClusterPipeline": await self._execution_strategy.initialize() return self async def __aenter__(self) -> "ClusterPipeline": return await self.initialize() async def __aexit__(self, exc_type: None, exc_value: None, traceback: None) -> None: await self.reset() def __await__(self) -> Generator[Any, None, "ClusterPipeline"]: return self.initialize().__await__() def __bool__(self) -> bool: "Pipeline instances should always evaluate to True on Python 3+" return True def __len__(self) -> int: return len(self._execution_strategy) def execute_command( self, *args: Union[KeyT, EncodableT], **kwargs: Any ) -> "ClusterPipeline": """ Append a raw command to the pipeline. :param args: | Raw command args :param kwargs: - target_nodes: :attr:`NODE_FLAGS` or :class:`~.ClusterNode` or List[:class:`~.ClusterNode`] or Dict[Any, :class:`~.ClusterNode`] - Rest of the kwargs are passed to the Redis connection """ return self._execution_strategy.execute_command(*args, **kwargs) async def execute( self, raise_on_error: bool = True, allow_redirections: bool = True ) -> List[Any]: """ Execute the pipeline. It will retry the commands as specified by retries specified in :attr:`retry` & then raise an exception. :param raise_on_error: | Raise the first error if there are any errors :param allow_redirections: | Whether to retry each failed command individually in case of redirection errors :raises RedisClusterException: if target_nodes is not provided & the command can't be mapped to a slot """ try: return await self._execution_strategy.execute( raise_on_error, allow_redirections ) finally: await self.reset() def _split_command_across_slots( self, command: str, *keys: KeyT ) -> "ClusterPipeline": for slot_keys in self.cluster_client._partition_keys_by_slot(keys).values(): self.execute_command(command, *slot_keys) return self async def reset(self): """ Reset back to empty pipeline. """ await self._execution_strategy.reset() def multi(self): """ Start a transactional block of the pipeline after WATCH commands are issued. End the transactional block with `execute`. """ self._execution_strategy.multi() async def discard(self): """ """ await self._execution_strategy.discard() async def watch(self, *names): """Watches the values at keys ``names``""" await self._execution_strategy.watch(*names) async def unwatch(self): """Unwatches all previously specified keys""" await self._execution_strategy.unwatch() async def unlink(self, *names): await self._execution_strategy.unlink(*names) def mset_nonatomic( self, mapping: Mapping[AnyKeyT, EncodableT] ) -> "ClusterPipeline": return self._execution_strategy.mset_nonatomic(mapping) for command in PIPELINE_BLOCKED_COMMANDS: command = command.replace(" ", "_").lower() if command == "mset_nonatomic": continue setattr(ClusterPipeline, command, block_pipeline_command(command)) class PipelineCommand: def __init__(self, position: int, *args: Any, **kwargs: Any) -> None: self.args = args self.kwargs = kwargs self.position = position self.result: Union[Any, Exception] = None self.command_policies: Optional[CommandPolicies] = None def __repr__(self) -> str: return f"[{self.position}] {self.args} ({self.kwargs})" class ExecutionStrategy(ABC): @abstractmethod async def initialize(self) -> "ClusterPipeline": """ Initialize the execution strategy. See ClusterPipeline.initialize() """ pass @abstractmethod def execute_command( self, *args: Union[KeyT, EncodableT], **kwargs: Any ) -> "ClusterPipeline": """ Append a raw command to the pipeline. See ClusterPipeline.execute_command() """ pass @abstractmethod async def execute( self, raise_on_error: bool = True, allow_redirections: bool = True ) -> List[Any]: """ Execute the pipeline. It will retry the commands as specified by retries specified in :attr:`retry` & then raise an exception. See ClusterPipeline.execute() """ pass @abstractmethod def mset_nonatomic( self, mapping: Mapping[AnyKeyT, EncodableT] ) -> "ClusterPipeline": """ Executes multiple MSET commands according to the provided slot/pairs mapping. See ClusterPipeline.mset_nonatomic() """ pass @abstractmethod async def reset(self): """ Resets current execution strategy. See: ClusterPipeline.reset() """ pass @abstractmethod def multi(self): """ Starts transactional context. See: ClusterPipeline.multi() """ pass @abstractmethod async def watch(self, *names): """ Watch given keys. See: ClusterPipeline.watch() """ pass @abstractmethod async def unwatch(self): """ Unwatches all previously specified keys See: ClusterPipeline.unwatch() """ pass @abstractmethod async def discard(self): pass @abstractmethod async def unlink(self, *names): """ "Unlink a key specified by ``names``" See: ClusterPipeline.unlink() """ pass @abstractmethod def __len__(self) -> int: pass class AbstractStrategy(ExecutionStrategy): def __init__(self, pipe: ClusterPipeline) -> None: self._pipe: ClusterPipeline = pipe self._command_queue: List["PipelineCommand"] = [] async def initialize(self) -> "ClusterPipeline": if self._pipe.cluster_client._initialize: await self._pipe.cluster_client.initialize() self._command_queue = [] return self._pipe def execute_command( self, *args: Union[KeyT, EncodableT], **kwargs: Any ) -> "ClusterPipeline": self._command_queue.append( PipelineCommand(len(self._command_queue), *args, **kwargs) ) return self._pipe def _annotate_exception(self, exception, number, command): """ Provides extra context to the exception prior to it being handled """ cmd = " ".join(map(safe_str, command)) msg = ( f"Command # {number} ({truncate_text(cmd)}) of pipeline " f"caused error: {exception.args[0]}" ) exception.args = (msg,) + exception.args[1:] @abstractmethod def mset_nonatomic( self, mapping: Mapping[AnyKeyT, EncodableT] ) -> "ClusterPipeline": pass @abstractmethod async def execute( self, raise_on_error: bool = True, allow_redirections: bool = True ) -> List[Any]: pass @abstractmethod async def reset(self): pass @abstractmethod def multi(self): pass @abstractmethod async def watch(self, *names): pass @abstractmethod async def unwatch(self): pass @abstractmethod async def discard(self): pass @abstractmethod async def unlink(self, *names): pass def __len__(self) -> int: return len(self._command_queue) class PipelineStrategy(AbstractStrategy): def __init__(self, pipe: ClusterPipeline) -> None: super().__init__(pipe) def mset_nonatomic( self, mapping: Mapping[AnyKeyT, EncodableT] ) -> "ClusterPipeline": encoder = self._pipe.cluster_client.encoder slots_pairs = {} for pair in mapping.items(): slot = key_slot(encoder.encode(pair[0])) slots_pairs.setdefault(slot, []).extend(pair) for pairs in slots_pairs.values(): self.execute_command("MSET", *pairs) return self._pipe async def execute( self, raise_on_error: bool = True, allow_redirections: bool = True ) -> List[Any]: if not self._command_queue: return [] try: retry_attempts = self._pipe.cluster_client.retry.get_retries() while True: try: if self._pipe.cluster_client._initialize: await self._pipe.cluster_client.initialize() return await self._execute( self._pipe.cluster_client, self._command_queue, raise_on_error=raise_on_error, allow_redirections=allow_redirections, ) except RedisCluster.ERRORS_ALLOW_RETRY as e: if retry_attempts > 0: # Try again with the new cluster setup. All other errors # should be raised. retry_attempts -= 1 await self._pipe.cluster_client.aclose() await asyncio.sleep(0.25) else: # All other errors should be raised. raise e finally: await self.reset() async def _execute( self, client: "RedisCluster", stack: List["PipelineCommand"], raise_on_error: bool = True, allow_redirections: bool = True, ) -> List[Any]: todo = [ cmd for cmd in stack if not cmd.result or isinstance(cmd.result, Exception) ] nodes = {} for cmd in todo: passed_targets = cmd.kwargs.pop("target_nodes", None) command_policies = await client._policy_resolver.resolve( cmd.args[0].lower() ) if passed_targets and not client._is_node_flag(passed_targets): target_nodes = client._parse_target_nodes(passed_targets) if not command_policies: command_policies = CommandPolicies() else: if not command_policies: command_flag = client.command_flags.get(cmd.args[0]) if not command_flag: # Fallback to default policy if not client.get_default_node(): slot = None else: slot = await client._determine_slot(*cmd.args) if slot is None: command_policies = CommandPolicies() else: command_policies = CommandPolicies( request_policy=RequestPolicy.DEFAULT_KEYED, response_policy=ResponsePolicy.DEFAULT_KEYED, ) else: if command_flag in client._command_flags_mapping: command_policies = CommandPolicies( request_policy=client._command_flags_mapping[ command_flag ] ) else: command_policies = CommandPolicies() target_nodes = await client._determine_nodes( *cmd.args, request_policy=command_policies.request_policy, node_flag=passed_targets, ) if not target_nodes: raise RedisClusterException( f"No targets were found to execute {cmd.args} command on" ) cmd.command_policies = command_policies if len(target_nodes) > 1: raise RedisClusterException(f"Too many targets for command {cmd.args}") node = target_nodes[0] if node.name not in nodes: nodes[node.name] = (node, []) nodes[node.name][1].append(cmd) # Start timing for observability start_time = time.monotonic() errors = await asyncio.gather( *( asyncio.create_task(node[0].execute_pipeline(node[1])) for node in nodes.values() ) ) # Record operation duration for each node for node_name, (node, commands) in nodes.items(): # Find the first error in this node's commands, if any node_error = None for cmd in commands: if isinstance(cmd.result, Exception): node_error = cmd.result break db = node.connection_kwargs.get("db", 0) await record_operation_duration( command_name="PIPELINE", duration_seconds=time.monotonic() - start_time, server_address=node.host, server_port=node.port, db_namespace=str(db) if db is not None else None, error=node_error, ) if any(errors): if allow_redirections: # send each errored command individually for cmd in todo: if isinstance(cmd.result, (TryAgainError, MovedError, AskError)): try: cmd.result = client._policies_callback_mapping[ cmd.command_policies.response_policy ](await client.execute_command(*cmd.args, **cmd.kwargs)) except Exception as e: cmd.result = e if raise_on_error: for cmd in todo: result = cmd.result if isinstance(result, Exception): command = " ".join(map(safe_str, cmd.args)) msg = ( f"Command # {cmd.position + 1} " f"({truncate_text(command)}) " f"of pipeline caused error: {result.args}" ) result.args = (msg,) + result.args[1:] raise result default_cluster_node = client.get_default_node() # Check whether the default node was used. In some cases, # 'client.get_default_node()' may return None. The check below # prevents a potential AttributeError. if default_cluster_node is not None: default_node = nodes.get(default_cluster_node.name) if default_node is not None: # This pipeline execution used the default node, check if we need # to replace it. # Note: when the error is raised we'll reset the default node in the # caller function. for cmd in default_node[1]: # Check if it has a command that failed with a relevant # exception if type(cmd.result) in RedisCluster.ERRORS_ALLOW_RETRY: client.replace_default_node() break return [cmd.result for cmd in stack] async def reset(self): """ Reset back to empty pipeline. """ self._command_queue = [] def multi(self): raise RedisClusterException( "method multi() is not supported outside of transactional context" ) async def watch(self, *names): raise RedisClusterException( "method watch() is not supported outside of transactional context" ) async def unwatch(self): raise RedisClusterException( "method unwatch() is not supported outside of transactional context" ) async def discard(self): raise RedisClusterException( "method discard() is not supported outside of transactional context" ) async def unlink(self, *names): if len(names) != 1: raise RedisClusterException( "unlinking multiple keys is not implemented in pipeline command" ) return self.execute_command("UNLINK", names[0]) class TransactionStrategy(AbstractStrategy): NO_SLOTS_COMMANDS = {"UNWATCH"} IMMEDIATE_EXECUTE_COMMANDS = {"WATCH", "UNWATCH"} UNWATCH_COMMANDS = {"DISCARD", "EXEC", "UNWATCH"} SLOT_REDIRECT_ERRORS = (AskError, MovedError) CONNECTION_ERRORS = ( ConnectionError, OSError, ClusterDownError, SlotNotCoveredError, ) def __init__(self, pipe: ClusterPipeline) -> None: super().__init__(pipe) self._explicit_transaction = False self._watching = False self._pipeline_slots: Set[int] = set() self._transaction_node: Optional[ClusterNode] = None self._transaction_connection: Optional[Connection] = None self._executing = False self._retry = copy(self._pipe.cluster_client.retry) self._retry.update_supported_errors( RedisCluster.ERRORS_ALLOW_RETRY + self.SLOT_REDIRECT_ERRORS ) def _get_client_and_connection_for_transaction( self, ) -> Tuple[ClusterNode, Connection]: """ Find a connection for a pipeline transaction. For running an atomic transaction, watch keys ensure that contents have not been altered as long as the watch commands for those keys were sent over the same connection. So once we start watching a key, we fetch a connection to the node that owns that slot and reuse it. """ if not self._pipeline_slots: raise RedisClusterException( "At least a command with a key is needed to identify a node" ) node: ClusterNode = self._pipe.cluster_client.nodes_manager.get_node_from_slot( list(self._pipeline_slots)[0], False ) self._transaction_node = node if not self._transaction_connection: connection: Connection = self._transaction_node.acquire_connection() self._transaction_connection = connection return self._transaction_node, self._transaction_connection def execute_command(self, *args: Union[KeyT, EncodableT], **kwargs: Any) -> "Any": # Given the limitation of ClusterPipeline sync API, we have to run it in thread. response = None error = None def runner(): nonlocal response nonlocal error try: response = asyncio.run(self._execute_command(*args, **kwargs)) except Exception as e: error = e thread = threading.Thread(target=runner) thread.start() thread.join() if error: raise error return response async def _execute_command( self, *args: Union[KeyT, EncodableT], **kwargs: Any ) -> Any: if self._pipe.cluster_client._initialize: await self._pipe.cluster_client.initialize() slot_number: Optional[int] = None if args[0] not in self.NO_SLOTS_COMMANDS: slot_number = await self._pipe.cluster_client._determine_slot(*args) if ( self._watching or args[0] in self.IMMEDIATE_EXECUTE_COMMANDS ) and not self._explicit_transaction: if args[0] == "WATCH": self._validate_watch() if slot_number is not None: if self._pipeline_slots and slot_number not in self._pipeline_slots: raise CrossSlotTransactionError( "Cannot watch or send commands on different slots" ) self._pipeline_slots.add(slot_number) elif args[0] not in self.NO_SLOTS_COMMANDS: raise RedisClusterException( f"Cannot identify slot number for command: {args[0]}," "it cannot be triggered in a transaction" ) return self._immediate_execute_command(*args, **kwargs) else: if slot_number is not None: self._pipeline_slots.add(slot_number) return super().execute_command(*args, **kwargs) def _validate_watch(self): if self._explicit_transaction: raise RedisError("Cannot issue a WATCH after a MULTI") self._watching = True async def _immediate_execute_command(self, *args, **options): return await self._retry.call_with_retry( lambda: self._get_connection_and_send_command(*args, **options), self._reinitialize_on_error, with_failure_count=True, ) async def _get_connection_and_send_command(self, *args, **options): redis_node, connection = self._get_client_and_connection_for_transaction() # Only disconnect if not watching - disconnecting would lose WATCH state if not self._watching: await redis_node.disconnect_if_needed(connection) # Start timing for observability start_time = time.monotonic() try: response = await self._send_command_parse_response( connection, redis_node, args[0], *args, **options ) await record_operation_duration( command_name=args[0], duration_seconds=time.monotonic() - start_time, server_address=connection.host, server_port=connection.port, db_namespace=str(connection.db), ) return response except Exception as e: e.connection = connection await record_operation_duration( command_name=args[0], duration_seconds=time.monotonic() - start_time, server_address=connection.host, server_port=connection.port, db_namespace=str(connection.db), error=e, ) raise async def _send_command_parse_response( self, connection: Connection, redis_node: ClusterNode, command_name, *args, **options, ): """ Send a command and parse the response """ await connection.send_command(*args) output = await redis_node.parse_response(connection, command_name, **options) if command_name in self.UNWATCH_COMMANDS: self._watching = False return output async def _reinitialize_on_error(self, error, failure_count): if hasattr(error, "connection"): await record_error_count( server_address=error.connection.host, server_port=error.connection.port, network_peer_address=error.connection.host, network_peer_port=error.connection.port, error_type=error, retry_attempts=failure_count, is_internal=True, ) if self._watching: if type(error) in self.SLOT_REDIRECT_ERRORS and self._executing: raise WatchError("Slot rebalancing occurred while watching keys") if ( type(error) in self.SLOT_REDIRECT_ERRORS or type(error) in self.CONNECTION_ERRORS ): if self._transaction_connection and self._transaction_node: # Disconnect and release back to pool await self._transaction_connection.disconnect() self._transaction_node.release(self._transaction_connection) self._transaction_connection = None self._pipe.cluster_client.reinitialize_counter += 1 if ( self._pipe.cluster_client.reinitialize_steps and self._pipe.cluster_client.reinitialize_counter % self._pipe.cluster_client.reinitialize_steps == 0 ): await self._pipe.cluster_client.nodes_manager.initialize() self.reinitialize_counter = 0 else: if isinstance(error, AskError): await self._pipe.cluster_client.nodes_manager.move_slot(error) self._executing = False async def _raise_first_error(self, responses, stack, start_time): """ Raise the first exception on the stack """ for r, cmd in zip(responses, stack): if isinstance(r, Exception): self._annotate_exception(r, cmd.position + 1, cmd.args) await record_operation_duration( command_name="TRANSACTION", duration_seconds=time.monotonic() - start_time, server_address=self._transaction_connection.host, server_port=self._transaction_connection.port, db_namespace=str(self._transaction_connection.db), error=r, ) raise r def mset_nonatomic( self, mapping: Mapping[AnyKeyT, EncodableT] ) -> "ClusterPipeline": raise NotImplementedError("Method is not supported in transactional context.") async def execute( self, raise_on_error: bool = True, allow_redirections: bool = True ) -> List[Any]: stack = self._command_queue if not stack and (not self._watching or not self._pipeline_slots): return [] return await self._execute_transaction_with_retries(stack, raise_on_error) async def _execute_transaction_with_retries( self, stack: List["PipelineCommand"], raise_on_error: bool ): return await self._retry.call_with_retry( lambda: self._execute_transaction(stack, raise_on_error), lambda error, failure_count: self._reinitialize_on_error( error, failure_count ), with_failure_count=True, ) async def _execute_transaction( self, stack: List["PipelineCommand"], raise_on_error: bool ): if len(self._pipeline_slots) > 1: raise CrossSlotTransactionError( "All keys involved in a cluster transaction must map to the same slot" ) self._executing = True redis_node, connection = self._get_client_and_connection_for_transaction() # Only disconnect if not watching - disconnecting would lose WATCH state if not self._watching: await redis_node.disconnect_if_needed(connection) stack = chain( [PipelineCommand(0, "MULTI")], stack, [PipelineCommand(0, "EXEC")], ) commands = [c.args for c in stack if EMPTY_RESPONSE not in c.kwargs] packed_commands = connection.pack_commands(commands) # Start timing for observability start_time = time.monotonic() await connection.send_packed_command(packed_commands) errors = [] # parse off the response for MULTI # NOTE: we need to handle ResponseErrors here and continue # so that we read all the additional command messages from # the socket try: await redis_node.parse_response(connection, "MULTI") except ResponseError as e: self._annotate_exception(e, 0, "MULTI") errors.append(e) except self.CONNECTION_ERRORS as cluster_error: self._annotate_exception(cluster_error, 0, "MULTI") cluster_error.connection = connection raise # and all the other commands for i, command in enumerate(self._command_queue): if EMPTY_RESPONSE in command.kwargs: errors.append((i, command.kwargs[EMPTY_RESPONSE])) else: try: _ = await redis_node.parse_response(connection, "_") except self.SLOT_REDIRECT_ERRORS as slot_error: self._annotate_exception(slot_error, i + 1, command.args) errors.append(slot_error) except self.CONNECTION_ERRORS as cluster_error: self._annotate_exception(cluster_error, i + 1, command.args) cluster_error.connection = connection raise except ResponseError as e: self._annotate_exception(e, i + 1, command.args) errors.append(e) response = None # parse the EXEC. try: response = await redis_node.parse_response(connection, "EXEC") except ExecAbortError: if errors: raise errors[0] raise self._executing = False # EXEC clears any watched keys self._watching = False if response is None: raise WatchError("Watched variable changed.") # put any parse errors into the response for i, e in errors: response.insert(i, e) if len(response) != len(self._command_queue): raise InvalidPipelineStack( "Unexpected response length for cluster pipeline EXEC." " Command stack was {} but response had length {}".format( [c.args[0] for c in self._command_queue], len(response) ) ) # find any errors in the response and raise if necessary if raise_on_error or len(errors) > 0: await self._raise_first_error( response, self._command_queue, start_time, ) # We have to run response callbacks manually data = [] for r, cmd in zip(response, self._command_queue): if not isinstance(r, Exception): command_name = cmd.args[0] if command_name in self._pipe.cluster_client.response_callbacks: r = self._pipe.cluster_client.response_callbacks[command_name]( r, **cmd.kwargs ) data.append(r) await record_operation_duration( command_name="TRANSACTION", duration_seconds=time.monotonic() - start_time, server_address=connection.host, server_port=connection.port, db_namespace=str(connection.db), ) return data async def reset(self): self._command_queue = [] try: # make sure to reset the connection state in the event that we # were watching something if self._transaction_connection: try: if self._watching: # call this manually since our unwatch or # immediate_execute_command methods can call reset() await self._transaction_connection.send_command("UNWATCH") await self._transaction_connection.read_response() except self.CONNECTION_ERRORS: # disconnect will also remove any previous WATCHes if self._transaction_connection: await self._transaction_connection.disconnect() except asyncio.CancelledError: # Disconnect so any unread UNWATCH reply does not get # served to the next caller that takes the connection. if self._transaction_connection: await self._transaction_connection.disconnect() raise else: # On the happy path, honor lazy reconnect before release. await self._transaction_node.disconnect_if_needed( self._transaction_connection ) finally: # Always return the connection to the node's free queue, even on # cancellation, so cancelled resets do not leak pooled # connections. Detach the reference before releasing so the # strategy never holds a pointer to a returned connection. # ClusterNode.release is synchronous, so no shield is required. if self._transaction_connection and self._transaction_node: connection, self._transaction_connection = ( self._transaction_connection, None, ) self._transaction_node.release(connection) # clean up the other instance attributes self._transaction_connection = None self._transaction_node = None self._watching = False self._explicit_transaction = False self._pipeline_slots = set() self._executing = False def multi(self): if self._explicit_transaction: raise RedisError("Cannot issue nested calls to MULTI") if self._command_queue: raise RedisError( "Commands without an initial WATCH have already been issued" ) self._explicit_transaction = True async def watch(self, *names): if self._explicit_transaction: raise RedisError("Cannot issue a WATCH after a MULTI") return await self.execute_command("WATCH", *names) async def unwatch(self): if self._watching: return await self.execute_command("UNWATCH") return True async def discard(self): await self.reset() async def unlink(self, *names): return self.execute_command("UNLINK", *names) class _ClusterNodePoolAdapter(ConnectionPoolInterface): """Thin adapter exposing the :class:`ConnectionPoolInterface` that :class:`PubSub` requires, backed by a :class:`ClusterNode`'s own connection pool. Connections are acquired from the node via :meth:`ClusterNode.acquire_connection` and returned via :meth:`ClusterNode.release`. :meth:`PubSub.aclose` already disconnects the connection *before* calling :meth:`release`, so the connection is returned to the node's free-queue in a disconnected state — guaranteeing that a subscribed socket is never silently reused for regular commands. Methods that do not apply to this adapter (the underlying node's lifecycle is managed by the cluster, not by individual PubSub instances) are implemented as no-ops so the adapter remains a valid :class:`ConnectionPoolInterface`. """ def __init__(self, node: "ClusterNode") -> None: self._node = node self.connection_kwargs = node.connection_kwargs # -- methods used by PubSub ------------------------------------------------ def get_encoder(self) -> Encoder: return self._node.get_encoder() async def get_connection( self, command_name: Optional[str] = None, *keys: Any, **options: Any ) -> AbstractConnection: connection = self._node.acquire_connection() try: await connection.connect() except BaseException: # connect() may fail mid-handshake (e.g. after the TCP socket # is established but before AUTH/HELLO completes) leaving the # connection in a partially-connected state. Disconnect before # returning it to the node's free queue so it is not reused. await connection.disconnect() self._node.release(connection) raise return connection async def release(self, connection: AbstractConnection) -> None: # PubSub.aclose() disconnects the connection before calling # release(), so it is safe to put it back in the node's free # queue – it will reconnect lazily on next use. await self._node.disconnect_if_needed(connection) self._node.release(connection) # -- no-op stubs for the rest of ConnectionPoolInterface ------------------- # The node's connections are shared with regular cluster traffic and its # lifecycle is managed by RedisCluster / NodesManager, so the adapter must # not reset, disconnect, retry-configure or re-auth them on behalf of a # single PubSub instance. def get_protocol(self): return self.connection_kwargs.get("protocol", None) def reset(self) -> None: pass async def disconnect(self, inuse_connections: bool = True) -> None: pass async def aclose(self) -> None: pass def set_retry(self, retry: "Retry") -> None: pass async def re_auth_callback(self, token: TokenInterface) -> None: pass def get_connection_count(self) -> List[Tuple[int, dict]]: return [] def _unregister_slots_cache_listener( dispatcher_ref: "weakref.ref[EventDispatcher]", listener: AsyncEventListenerInterface, event_type: Type[object], ) -> None: # Module-level finalizer callback. Kept free of strong references to the # owning ClusterPubSub so attaching it via weakref.finalize does not # extend the pubsub's lifetime. dispatcher = dispatcher_ref() if dispatcher is not None: dispatcher.unregister_listeners({event_type: [listener]}) class ClusterPubSubSlotsCacheListener(AsyncEventListenerInterface): """ Async listener that forwards AsyncAfterSlotsCacheRefreshEvent to a ClusterPubSub. Holds a weak reference to the pubsub so it does not keep the instance alive. Deterministic cleanup of the dispatcher's strong reference to this listener is performed by a ``weakref.finalize`` attached to the owning ClusterPubSub in ``ClusterPubSub.__init__``. """ def __init__(self, pubsub: "ClusterPubSub") -> None: self._pubsub_ref: "weakref.ref[ClusterPubSub]" = weakref.ref(pubsub) async def listen(self, event: object) -> None: pubsub = self._pubsub_ref() if pubsub is None: # Race window between pubsub GC and the finalizer running; safe # no-op, finalizer will remove this listener shortly. return try: await pubsub.on_slots_changed() except Exception as e: # Listeners must not break slots-cache refresh; log and continue so # a single buggy pubsub cannot starve the rest. logger.exception( "pubsub %r raised during slots-cache change: %s: %s", pubsub, type(e).__name__, e, ) class ClusterPubSub(PubSub): """ Async cluster implementation for pub/sub. IMPORTANT: before using ClusterPubSub, read about the known limitations with pubsub in Cluster mode and learn how to workaround them: https://redis.readthedocs.io/en/stable/clustering.html#known-pubsub-limitations """ def __init__( self, redis_cluster: "RedisCluster", node: Optional["ClusterNode"] = None, host: Optional[str] = None, port: Optional[int] = None, push_handler_func: Optional[Callable] = None, event_dispatcher: Optional[EventDispatcher] = None, **kwargs: Any, ) -> None: """ When a pubsub instance is created without specifying a node, a single node will be transparently chosen for the pubsub connection on the first command execution. The node will be determined by: 1. Hashing the channel name in the request to find its keyslot 2. Selecting a node that handles the keyslot: If read_from_replicas is set to true or load_balancing_strategy is set, a replica can be selected. :param redis_cluster: RedisCluster instance :param node: ClusterNode to connect to :param host: Host of the node to connect to :param port: Port of the node to connect to :param push_handler_func: Optional push handler function :param event_dispatcher: Optional event dispatcher :param kwargs: Additional keyword arguments """ self.node = None self.set_pubsub_node(redis_cluster, node, host, port) # Borrow the node's own connection pool via an adapter rather than # creating a second, detached ConnectionPool for pubsub. if self.node is not None: connection_pool = _ClusterNodePoolAdapter(self.node) else: connection_pool = None self.cluster = redis_cluster self.node_pubsub_mapping: Dict[str, PubSub] = {} # Reverse index: shard channel (normalized) -> owning node.name. Used to # route sunsubscribe calls and reconcile subscriptions after slot # migration / failover. self._shard_channel_to_node: Dict[Any, str] = {} # Dedicated lock for shard-subscription bookkeeping. Distinct from # PubSub.self._lock (which serializes wire I/O on the cluster-level # connection used by aclose / send_command / regular subscribe) so # that reconciliation cannot starve those unrelated coroutines # during long per-channel migrations. self._shard_state_lock: asyncio.Lock = asyncio.Lock() # Background tasks created by on_slots_changed; kept to prevent GC. self._reconcile_tasks: Set[asyncio.Task] = set() self._pubsubs_generator = self._pubsubs_generator() if event_dispatcher is None: self._event_dispatcher = EventDispatcher() else: self._event_dispatcher = event_dispatcher super().__init__( connection_pool=connection_pool, encoder=redis_cluster.encoder, push_handler_func=push_handler_func, event_dispatcher=self._event_dispatcher, **kwargs, ) # Subscribe to slots-cache change notifications so shard subscriptions # can be reconciled automatically after topology refreshes. nm_dispatcher = redis_cluster.nodes_manager._event_dispatcher self._slots_cache_listener = ClusterPubSubSlotsCacheListener(self) nm_dispatcher.register_listeners( {AsyncAfterSlotsCacheRefreshEvent: [self._slots_cache_listener]} ) # Deterministic GC-time cleanup so short-lived pubsubs do not leak # listeners in the dispatcher when no slots-refresh event ever fires. weakref.finalize( self, _unregister_slots_cache_listener, weakref.ref(nm_dispatcher), self._slots_cache_listener, AsyncAfterSlotsCacheRefreshEvent, ) def set_pubsub_node( self, cluster: "RedisCluster", node: Optional["ClusterNode"] = None, host: Optional[str] = None, port: Optional[int] = None, ) -> None: """ The pubsub node will be set according to the passed node, host and port When none of the node, host, or port are specified - the node is set to None and will be determined by the keyslot of the channel in the first command to be executed. RedisClusterException will be thrown if the passed node does not exist in the cluster. If host is passed without port, or vice versa, a DataError will be thrown. """ if node is not None: # node is passed by the user self._raise_on_invalid_node(cluster, node, node.host, node.port) pubsub_node = node elif host is not None and port is not None: # host and port passed by the user node = cluster.get_node(host=host, port=port) self._raise_on_invalid_node(cluster, node, host, port) pubsub_node = node elif host is not None or port is not None: # only one of host and port is specified raise DataError("Specify both host and port") else: # nothing specified by the user pubsub_node = None self.node = pubsub_node def get_pubsub_node(self) -> Optional["ClusterNode"]: """ Get the node that is being used as the pubsub connection. :return: The ClusterNode being used for pubsub, or None if not yet determined """ return self.node async def _resubscribe_shard_channels(self) -> None: # A single node can own multiple slot ranges, so a batched # ``SSUBSCRIBE`` covering every tracked channel would be rejected by # Redis with a ``CROSSSLOT`` error. Group by hash slot and emit one # ``SSUBSCRIBE`` per slot. by_slot: defaultdict[int, dict] = defaultdict(dict) for k, v in self.shard_channels.items(): by_slot[key_slot(self.encoder.encode(k))][k] = v for subscriptions in by_slot.values(): await self._resubscribe(subscriptions, self.ssubscribe) def _get_node_pubsub(self, node: "ClusterNode") -> PubSub: """Get or create a PubSub instance for the given node.""" try: return self.node_pubsub_mapping[node.name] except KeyError: pubsub = PubSub( connection_pool=_ClusterNodePoolAdapter(node), encoder=self.cluster.encoder, push_handler_func=self.push_handler_func, event_dispatcher=self._event_dispatcher, ) # Replay shard subscriptions on reconnect with slot-aware grouping # so that channels spanning multiple slots owned by this node do # not trigger a CROSSSLOT error. pubsub._resubscribe_shard_channels = MethodType( ClusterPubSub._resubscribe_shard_channels, pubsub ) self.node_pubsub_mapping[node.name] = pubsub return pubsub def _find_node_name_for_pubsub(self, pubsub: PubSub) -> Optional[str]: for name, candidate in self.node_pubsub_mapping.items(): if candidate is pubsub: return name return None async def _sharded_message_generator( self, timeout: float = 0.0 ) -> Tuple[Optional[PubSub], Optional[Dict[str, Any]]]: """Generate messages from shard channels across all nodes.""" for _ in range(len(self.node_pubsub_mapping)): pubsub = next(self._pubsubs_generator) # Don't pass ignore_subscribe_messages here - let get_sharded_message # handle the filtering after processing subscription state changes message = await pubsub.get_message( ignore_subscribe_messages=False, timeout=timeout ) if message is not None: return pubsub, message return None, None def _pubsubs_generator(self) -> Generator[PubSub, None, None]: """Generator that yields PubSub instances in round-robin fashion.""" while True: current_nodes = list(self.node_pubsub_mapping.values()) if not current_nodes: return # Avoid infinite loop when no subscriptions exist yield from current_nodes async def get_sharded_message( self, ignore_subscribe_messages: bool = False, timeout: float = 0.0, target_node: Optional["ClusterNode"] = None, ) -> Optional[Dict[str, Any]]: """ Get a message from shard channels. :param ignore_subscribe_messages: Whether to ignore subscribe messages :param timeout: Timeout for message retrieval :param target_node: Specific node to get message from :return: Message dictionary or None """ pubsub: Optional[PubSub] if target_node: pubsub = self.node_pubsub_mapping.get(target_node.name) if pubsub: # Don't pass ignore_subscribe_messages here - let get_sharded_message # handle the filtering after processing subscription state changes message = await pubsub.get_message( ignore_subscribe_messages=False, timeout=timeout ) else: message = None else: pubsub, message = await self._sharded_message_generator(timeout=timeout) if message is None: return None # Only sunsubscribe mutates cluster-level shard state; bypassing the # lock on the data-message hot path keeps smessage delivery from # competing with the reconciliation task for _shard_state_lock. if str_if_bytes(message["type"]) == "sunsubscribe": # Serialize state mutation against reinitialize_shard_subscriptions # (background task). The blocking get_message above intentionally # runs outside the lock so reconciliation is not stalled by long # polls. async with self._shard_state_lock: if message["channel"] in self.pending_unsubscribe_shard_channels: # User-initiated sunsubscribe: drop from cluster-level tracking. self.pending_unsubscribe_shard_channels.remove(message["channel"]) self.shard_channels.pop(message["channel"], None) self._shard_channel_to_node.pop(message["channel"], None) # Drop the per-node pubsub that delivered the confirmation once # it no longer holds any shard subscriptions, regardless of # whether the sunsubscribe was user-initiated or driven by # slot-migration reconciliation (_migrate_shard_channel, which # intentionally does not add the channel to # pending_unsubscribe_shard_channels). This releases the # dedicated connection that would otherwise linger. # Identifying the receiving pubsub directly (rather than via # the cluster's current slot map) is required after slot # migration, where the channel's owner is no longer the node # that received our original SSUBSCRIBE. if pubsub is not None and not pubsub.subscribed: name = self._find_node_name_for_pubsub(pubsub) if name is not None: try: await pubsub.aclose() except Exception: pass self.node_pubsub_mapping.pop(name, None) # Only suppress subscribe/unsubscribe messages, not data messages (smessage) if str_if_bytes(message["type"]) in ("ssubscribe", "sunsubscribe"): if self.ignore_subscribe_messages or ignore_subscribe_messages: return None return message async def ssubscribe( self, *args: ChannelT | Subscription, **kwargs: PubSubHandler ) -> None: """ Subscribe to shard channels. :param args: Channel names or ``Subscription`` objects :param kwargs: Channel names with handlers """ s_channels = parse_pubsub_subscriptions(args, kwargs) # Serialize against reinitialize_shard_subscriptions (background # task) so the reverse index, shard_channels, and node_pubsub_mapping # are not mutated concurrently. _migrate_shard_channel below does not # re-acquire this lock (asyncio.Lock is non-reentrant). async with self._shard_state_lock: for s_channel, handler in s_channels.items(): node = self.cluster.get_node_from_key(s_channel) if not node: continue # Lazy re-route: if this channel is already tracked against a # different node (e.g. after a slot migration), migrate it now # so the caller's intent is applied on the current owner. normalized_key = next(iter(self._normalize_keys({s_channel: None}))) old_name = self._shard_channel_to_node.get(normalized_key) if old_name and old_name != node.name: # Match PubSub.ssubscribe() dict.update() semantics: the # caller's newly supplied handler (including None) always # overrides any previously registered handler. await self._migrate_shard_channel( normalized_key, handler, old_name, node, ) continue pubsub = self._get_node_pubsub(node) if handler: await pubsub.ssubscribe(Subscription(s_channel, handler)) else: await pubsub.ssubscribe(s_channel) self.shard_channels.update(pubsub.shard_channels) self._shard_channel_to_node[normalized_key] = node.name self.pending_unsubscribe_shard_channels.difference_update( self._normalize_keys({s_channel: None}) ) async def sunsubscribe(self, *args: Any) -> None: """ Unsubscribe from shard channels. :param args: Channel names to unsubscribe from. If empty, unsubscribe from all. """ if args: args = list_or_args(args[0], args[1:]) else: args = list(self.shard_channels.keys()) # Serialize against reinitialize_shard_subscriptions: the reverse # index and node_pubsub_mapping must not change between the lookup # and the per-node sunsubscribe call below. async with self._shard_state_lock: for s_channel in args: normalized_key = next(iter(self._normalize_keys({s_channel: None}))) # Route via the reverse index so we unsubscribe on the node # that actually holds the subscription. After a slot migration # the cluster's current owner may no longer be that node. name = self._shard_channel_to_node.get(normalized_key) if name and name in self.node_pubsub_mapping: pubsub = self.node_pubsub_mapping[name] else: node = self.cluster.get_node_from_key(s_channel) if not node or node.name not in self.node_pubsub_mapping: continue pubsub = self.node_pubsub_mapping[node.name] await pubsub.sunsubscribe(s_channel) self.pending_unsubscribe_shard_channels.update( pubsub.pending_unsubscribe_shard_channels ) async def reinitialize_shard_subscriptions(self) -> None: """ Reconcile per-node shard subscriptions against the cluster's current slot ownership map. For each tracked shard channel whose owning node has changed (e.g. after CLUSTER SETSLOT / failover), sunsubscribe on the old node's pubsub and ssubscribe on the new owner's pubsub, preserving any registered handler. """ uncovered: list = [] made_progress = False first_migrate_error: Optional[BaseException] = None async with self._shard_state_lock: for channel, handler in list(self.shard_channels.items()): try: new_node = self.cluster.get_node_from_key(channel) except SlotNotCoveredError: # Slot is transiently uncovered (mid-migration / partial # topology refresh). Defer this channel so coverable # siblings still reconcile this pass; we surface the # error below so the caller (and logs) know not every # channel was reconciled. Retry happens on the next # slots-cache change notification. uncovered.append(channel) continue old_name = self._shard_channel_to_node.get(channel) if old_name == new_node.name: continue try: await self._migrate_shard_channel( channel, handler, old_name, new_node ) made_progress = True except (ConnectionError, TimeoutError, OSError) as e: # Transient connectivity error while subscribing on the # new owner (or unsubscribing on the old owner if its # handler chose to re-raise). Do not abort reconciliation # for sibling channels: _shard_channel_to_node was not # advanced for this channel, so the next slots-cache # change notification will retry it. logger.warning( "shard channel %r migration deferred: %s: %s", channel, type(e).__name__, e, ) if first_migrate_error is None: first_migrate_error = e continue # Garbage-collect per-node pubsubs that no longer hold any # subscription so their connections are released. for name, pubsub in list(self.node_pubsub_mapping.items()): if not pubsub.subscribed: try: await pubsub.aclose() except Exception: pass self.node_pubsub_mapping.pop(name, None) if uncovered: # Surface the uncovered channels so the caller (and observer # notification path) knows reconciliation was incomplete. All # coverable siblings have already been migrated above. raise SlotNotCoveredError( f"{len(uncovered)} shard channel(s) left unreconciled; " f"slot(s) not covered by the cluster: {uncovered!r}" ) if first_migrate_error is not None and not made_progress: # Every migration attempted in this pass failed transiently and # nothing else made progress. Re-raise the first caught error # (typically the root cause; later failures are often downstream # symptoms of the same unreachable node) so the task's done- # callback surfaces a single representative failure through the # same logger channel used for SlotNotCoveredError. Per-channel # WARNINGs above preserve the full forensic detail. raise first_migrate_error async def _migrate_shard_channel( self, channel: Any, handler: Optional[Callable], old_name: Optional[str], new_node: "ClusterNode", ) -> None: # Detach from the old per-node pubsub, best-effort: the old node may # already be unreachable during migration / failover. if old_name and old_name in self.node_pubsub_mapping: old_pubsub = self.node_pubsub_mapping[old_name] try: await old_pubsub.sunsubscribe(channel) except (ConnectionError, TimeoutError, OSError): # redis-py's Connection has already called ``disconnect()`` # before raising (see Connection.read_response / # send_packed_command with ``disconnect_on_error=True``), # so ``old_pubsub``'s dedicated socket is gone. Two cases: # # 1. The old node is no longer in the cluster topology # (e.g. removed by failover / topology refresh): no # reconnect target exists, so ``old_pubsub.subscribed`` # would stay True forever and the end-of-pass GC block # would skip it. Drop it eagerly so the round-robin # generator does not keep yielding a dead pubsub that # produces periodic errors from ``get_sharded_message``. # 2. The old node is still known (transiently slow / # unreachable): ``PubSub._execute`` auto-reconnects and # ``on_connect`` re-subscribes to remaining channels, # so other subscriptions on the same pubsub recover # naturally. Leave it alone. if self.cluster.get_node(node_name=old_name) is None: try: await old_pubsub.aclose() except Exception: pass self.node_pubsub_mapping.pop(old_name, None) # Attach to the new per-node pubsub, preserving the handler. Decode to # a text key only when we must pass it as a kwarg (handler present). new_pubsub = self._get_node_pubsub(new_node) if handler: await new_pubsub.ssubscribe(Subscription(channel, handler)) else: await new_pubsub.ssubscribe(channel) self.shard_channels.update(new_pubsub.shard_channels) normalized_key = next(iter(self._normalize_keys({channel: None}))) self._shard_channel_to_node[normalized_key] = new_node.name self.pending_unsubscribe_shard_channels.difference_update( self._normalize_keys({channel: None}) ) async def on_slots_changed(self) -> None: # Observer hook invoked by NodesManager after a slots-cache refresh. # Schedule reconciliation as a separate task so the caller's code # path (typically MovedError handling in _execute_command) is not # blocked on the network I/O performed by reinitialize_shard_ # subscriptions. No-op when there are no shard subscriptions to # reconcile. if not self.shard_channels: return task = asyncio.create_task(self.reinitialize_shard_subscriptions()) self._reconcile_tasks.add(task) task.add_done_callback(self._reconcile_tasks.discard) # Consume the task's exception (if any) so Python does not emit a # "Task exception was never retrieved" warning. reinitialize_shard_ # subscriptions surfaces SlotNotCoveredError when a slot is still # transiently uncovered; route it through the same logger channel # as sync ClusterPubSubSlotsCacheListener for consistent observability. task.add_done_callback(self._log_reconcile_task_exception) @staticmethod def _log_reconcile_task_exception(task: "asyncio.Task") -> None: if task.cancelled(): return exc = task.exception() if exc is not None: logger.error( "shard subscription reconciliation failed: %r", exc, exc_info=exc ) def get_redis_connection(self) -> Optional["AbstractConnection"]: """ Get the Redis connection of the pubsub connected node. Returns the pubsub's dedicated connection (acquired from its own connection pool), not from the ClusterNode's connection pool. This avoids the connection pool resource leak that would occur if we called node.acquire_connection() without releasing. """ # Return the pubsub's own dedicated connection, which is acquired # from self.connection_pool when executing pubsub commands. # This is safe because it's the connection dedicated to this pubsub # instance, not a shared pool connection from the ClusterNode. return self.connection async def aclose(self) -> None: """ Disconnect the pubsub connection. """ # Cancel and gather in-flight reconciliation tasks BEFORE acquiring # _shard_state_lock. The tasks themselves take that lock inside # reinitialize_shard_subscriptions; since asyncio.Lock is non- # reentrant, gathering while holding it would deadlock. Awaiting # each task with suppressed CancelledError also avoids unhandled- # exception warnings if the task was created but not yet scheduled. if self._reconcile_tasks: tasks = list(self._reconcile_tasks) for task in tasks: task.cancel() await asyncio.gather(*tasks, return_exceptions=True) # Hold _shard_state_lock across the rest of the teardown so it # observes the same mutual-exclusion discipline as ssubscribe / # sunsubscribe / get_sharded_message / reinitialize_shard_ # subscriptions, which all mutate shard_channels, # _shard_channel_to_node, and node_pubsub_mapping under this lock. # Without it, super().aclose() rebinds shard_channels and # pending_unsubscribe_shard_channels in parallel with a concurrent # user-coroutine mutation that resumes during one of the awaits # below, silently dropping subscription intent. async with self._shard_state_lock: self._reconcile_tasks.clear() # Close all shard pubsub instances first for pubsub in self.node_pubsub_mapping.values(): await pubsub.aclose() # Drop the now-dead per-node pubsubs from the mapping so the # round-robin in _pubsubs_generator / _sharded_message_generator # cannot yield them between teardown and re-subscription. self.node_pubsub_mapping.clear() # _pubsubs_generator captures node_pubsub_mapping.values() into # a local list inside ``yield from``; clearing the mapping does # not reach references already held by that captured snapshot, # so a generator suspended mid-yield-from would still surface # the now-aclose()'d per-node pubsubs after re-subscription. # Recreate it to drop the captured list. type(self) bypasses # the instance-level self-shadow established at __init__ # (self._pubsubs_generator = self._pubsubs_generator()). self._pubsubs_generator = type(self)._pubsubs_generator( # type: ignore[method-assign] self ) # Let parent handle self.connection disconnect under the lock # (includes disconnect, release to pool, and clearing # self.connection) await super().aclose() # Clear the reverse index so a reused instance doesn't route # against stale mappings. super().aclose() has already cleared # shard_channels. self._shard_channel_to_node.clear() def _raise_on_invalid_node( self, redis_cluster: "RedisCluster", node: Optional["ClusterNode"], host: Optional[str], port: Optional[int], ) -> None: """ Raise a RedisClusterException if the node is None or doesn't exist in the cluster. """ if node is None or redis_cluster.get_node(node_name=node.name) is None: raise RedisClusterException( f"Node {host}:{port} doesn't exist in the cluster" ) async def execute_command(self, *args: Any, **kwargs: Any) -> Any: """ Execute a command on the appropriate cluster node. Taken code from redis-py and tweaked to make it work within a cluster. """ # NOTE: don't parse the response in this function -- it could pull a # legitimate message off the stack if the connection is already # subscribed to one or more channels # For shard commands, route to appropriate node command = args[0].upper() if args else "" if command in ("SSUBSCRIBE", "SUNSUBSCRIBE", "SPUBLISH"): if len(args) > 1: channel = args[1] node = self.cluster.get_node_from_key(channel) if node: pubsub = self._get_node_pubsub(node) return await pubsub.execute_command(*args, **kwargs) # For other commands, use the set node or lazily discover one if self.connection is None: if self.connection_pool is None: if len(args) > 1: # Hash the first channel and get one of the nodes holding # this slot channel = args[1] slot = self.cluster.keyslot(channel) node = self.cluster.nodes_manager.get_node_from_slot( slot, self.cluster.read_from_replicas, self.cluster.load_balancing_strategy, ) else: # Get a random node node = self.cluster.get_random_node() self.node = node self.connection_pool = _ClusterNodePoolAdapter(node) # Now we have a connection_pool, use parent's execute_command return await super().execute_command(*args, **kwargs)