Feature

• Extend read from replica functionality to unshareded cluster pubsub

Bug Fix

• Release lock when exception occcurs in lock context

Development

• Add --no-docker option to pytest to skip starting
  docker containers when running tests.

Bug Fix

• Revert backward incompatible changes in compilation units.

Bug Fix

• Fix memory leak due to incorrect property caching of command request objects
  introduced in 6.3.0
• Remove incorrectly added opentelemetry-api dependency

Telemetry

• Ensure connection pool name is always added to operation
  duration metric
• Add library version to opentelemetry attributes

Bug Fix

• Fix regression introduced in 6.5.0 that results in batch
  request cancellations being suppressed.

Feature

• Add OpenTelemetry Support

Performance

• Pipeline performance optimization for large pipeline batches
  by collapsing request resolution into a single await.

Bug Fix

• Add missing pool_timeout argument in client from_url factory
  methods
• Remove extra final call to failure_hook in retry policy

Compatibility

• Relax version constraint on packaging dependency

Feature

• Allow explicitly specifying (through coredis.commands.CommandRequest.route)
  the cluster node per command for methods that can't automatically be routed to one or many nodes.
• Enable SCAN in cluster when used with explicit coredis.commands.CommandRequest.route
  to perform the scan on a single node in the cluster
• Use replicas in cluster pipelines when all commands are readonly

Bug Fix

• Fix performance regression in cluster pubsub due to a new
  capacity limiter being used for each connection instead of
  being shared by the pool.
• Catch authorization errors on connection establishment when setting
  CLIENT SETINFO and allow it to be usable

Features

• Add cross slot implementation for:

  • MSET
  • MGET
  • JSON.MGET
  • JSON.MSET

Compatibility

• Add GEOSHAPE index type for search

Bug Fixes

• Clean up connection pool task group when __aenter__
  fails during initialization

  Failure to do so would leave the connection pool unusable as the counter
  would be stuck at 1 and would not trigger a reinitialization.

• Fix failure to recover from errors during cluster discovery

• Handle pool depletion (and therefore a subsequent stall on getting a
  connection from the pool) in case of repeated connection establishment
  errors.

• Ensure connections are reusable before returning them from the pool
  as a connection used for pubsub which still has messages being
  received could be used for a new pubsub consumer resulting in
  stale messages being consumed by the new consumer.

• Fix initialization races and add early failure for unsafe concurrent sharing

  • Add initialization lock preventing a race
    where a herd of concurrent tasks could try to use the pool before
    it was fully initialized

  • Since AnyIO strictly requires that a cancel scope (in this case the task group
    of the connection pool) be exited by the exact same task that created it, there
    was a very easily reproducable scenario where an uninitialized pool
    shared by concurrent tasks would be initialized (counter=1) by a short lived task
    which would exit before the other tasks sharing the pool do. This
    would result in the last task (the task that decrements counter to 0)
    to try and cancel the task group and fail.

    The pool now fails early with a better error.

• Fix stalled initialization when a client is initialized with a cache is pointing
  to an unreachable redis server.

Usability

• Improve cluster exception heirarchy

  • Rename RedisClusterException to coredis.exceptions.RedisClusterError
    for consistency. (Alias is maintained for backward compatibility)
  • Ensure all cluster exceptions do derive from the base
    coredis.exceptions.RedisClusterError

• Fix inconsistent wrapping of non single key cluster methods
  in coroutines instead of returning an awaitable.
  These commands would therefore not have been
  usable in a consistent manner as the basic redis client.

Reliability & Performance

• Improve resiliency of sentinel primary

  Using the SENTINEL_MASTERS is not the recommended
  approach in the redis sentinel client implementation guide
  (https://redis.io/docs/latest/develop/reference/sentinel-clients/#step-2-ask-for-the-master-address)
  and it has been recommended to use GET_MASTER_ADDR-BY-NAME.

  Testing under repeated failovers it appears that the SENTINEL MASTERS
  command would often return a stale primary that had been demoted to a
  replica which would result in ReadOnlyError in previous versions (thus
  leaving the client broken since the ReadOnlyError would not result in a
  disconnection). The GET_MASTER_ADDR_BY_NAME command appears to have a
  more consistent response - however it is still possible that a stale
  primary is returned. In this case a coredis.exceptions.StalePrimaryError
  will now be raised on a new connection (failover should result in all active connections
  being closed by the server, as noted in: https://redis.io/docs/latest/develop/reference/sentinel-clients/#sentinel-failover-disconnection)

  Automatically recovering from stale primary errors can easily be
  done by using a to a retry policy for the coredis.exceptions.StalePrimaryError error

• Remove runtime validation of iterable parameters for redis commands
  as there is already sufficient static type validation along with
  opt in runtime validation with beartype.

• Add explicit verification of unsubscription requests in pubsub to ensure
  that the connection's push message buffer is fully consumed before releasing
  the connection back to the pool.

Typing

• Fix typing errors caused by invariant list type
• Add fallback type overloads for client constructors

Chores

• Improve generic structure of connection pools and connection types
  to provide a clearer separation of which parameters are acceptable
  for initializing which connection / connection pool type.
  ConnectionPools are now generic around the connection type they
  pool (invariant)

• Standardize server location types across the code base to use
  coredis.connection.Location

• Clearly separate sentinel connection pool, connection and client
  and move them to appropriate sub packages (coredis.connection &
  coredis.pool)

• Replace NodeManager with a ClusterLayout that owns both the live
  topology view and a background monitor task to keep the local cluster
  layout cache refreshed. DiscoveryService becomes a pure query object.

• Extract command splitting from cluster layout

  Added RoutingStrategy to extract the logic of routing/splitting commands
  that affect either no keys or keys on more than one slot out of
  ClusterLayout & RedisCluster. Routing logic is now collapsed into a
  single "route_strategy" in the command decorators that specifies:

  • where to distribute
  • how to split
  • how to merge

• Updated all redis command methods to have consistent docstrings for parameters
  & return types

• Enable type checking for all response callback related private submodules

Bug Fix

• Fix incorrect initialization of connection capacity limiter resulting in module level shared capacity limiter.
• Fix broken recovery from moved errors in cluster retry mechanism
• Remove bloat in source distribution and only package relevant source files

Compatibility

• Add support for DELEX (coredis.Redis.delex)
• Add support for DIGEST (coredis.Redis.digest)
• Add support for MSETEX (coredis.Redis.msetex)
• Update coredis.Redis.set to support conditional setting of values and expiration based on value or digest equality
• Add support for idempotent stream producers in coredis.Redis.xadd
• Add support for XCFGET (coredis.Redis.xcfgset)

Usability

• Remove ambiguous "Any" kwargs in clients, connection pools and connection constructors and tighten the acceptable parameters.
• Broaden acceptable types for commands that take mappings as arguments to allow for covariance in the mapping key

This release is a major architectural rewrite focused on structured concurrency, predictable resource lifetimes, and safer defaults. See the Migration guide for upgrade instructions.

Features

• Migrate entire library to anyio, adding structured concurrency and trio support

• Command API

  • Allow chainable coredis.commands.CommandRequest.transform method to accept inline callbacks to transform command responses
  • Added coredis.commands.CommandRequest.retry chainable method to the response from all commands to allow using retry policies individually with a request instead of having to apply it on all requests issued by a client

• Pipeline

  • Pipelines execute automatically when their context exits - results are accessible in a type-safe way
  • Cluster pipelines now support coredis.commands.Script instances

• Scripting

  • coredis.commands.Script and coredis.commands.Function can now be called with optional callbacks to transform the raw response from redis before returning it. This is supported by their associated coredis.commands.Script.wraps decorators as well.
  • Responses from coredis.commands.Script and coredis.commands.Function instances can now also be used with coredis.commands.CommandRequest.transform or coredis.commands.CommandRequest.retry.
  • Add a coredis.commands.function.wraps.verify_existence flag to library coredis.commands.function.wraps to optionally skip local validation and optimistically call it

• Retry Policies

  • Retry policies now allow infinite retries & retry deadlines.
  • Retry policies now support retrying when the raised ExceptionGroup contains a matching exception
  • coredis.retry.ExponentialBackoffRetryPolicy now supports coredis.retry.ExponentialBackoffRetryPolicy.jitter

• Miscellaneous

  • coredis.Redis.lock added as a convenient accessor for the Lua-based lock: coredis.patterns.lock.Lock (Previously found in coredis.recipes.LuaLock)
  • Add coredis.Redis.xconsumer factory method to create a single or group stream consumer
  • Improve guarantees of calling subscription methods on coredis.patterns.pubsub.PubSub instances. Using a coredis.patterns.pubsub.PubSub.subscription_timeout, the acknowledgement of subscription to a topic or pattern can be guaranteed before proceeding with listening to messages

Breaking Changes

• All classes (clients, connection pools, PubSub, pipelines, stream consumers, Sentinel clients) must be used with their async context managers for initialization/cleanup. This ensures a predictable and unified resource lifecycle management model.

• Users should replace coredis.cache.TrackingCache with coredis.patterns.cache.LRUCache when providing a cache instance to the clients. Cache size can no longer be bound by byte size and only coredis.patterns.cache.LRUCache.max_keys is supported

• Connection pools are now always blocking. Non-blocking commands are multiplexed over shared connections and released back to the pool immediately after being queued and thus most workloads naturally converge to using a single active connection. Pool exhaustion mainly occurs when the workload relies on blocking operations (e.g. pubsub, pipelines, blocking list commands), making a unified blocking model simpler and more predictable. The explicit blocking variants coredis.BlockingConnectionPool and coredis.BlockingClusterConnectionPool have been removed.

• Connection pool acquisition timeouts now raise TimeoutError instead of coredis.exceptions.ConnectionError.

• Pipeline

  • Pipeline instances can no longer be awaited and must be used as async context managers
  • Pipelines no longer expose an explicit coredis.patterns.pipeline.Pipeline.execute method and instead execute automatically when their context exits
  • coredis.Redis.pipeline is no longer an async method
  • The watches parameter has been removed from the pipeline constructor
  • Drop support for explicit management of WATCH, UNWATCH, MULTI in pipelines. This is replaced by the coredis.patterns.pipeline.Pipeline.watch context manager.

• Stream consumers can no longer be awaited and must be used as async context managers to ensure initialization

• coredis.Redis.pubsub, coredis.RedisCluster.pubsub & coredis.RedisCluster.sharded_pubsub now only accept keyword arguments

• When defining type stubs for FFI for Lua scripts or library functions, keys can only be distinguished from arguments by annotating them with coredis.typing.KeyT

• Module reorganization

  • Move Pub/Sub, Pipeline, Stream, Cache and Lock submodules to coredis.patterns
  • coredis.commands.pubsub → coredis.patterns.pubsub
  • coredis.pipeline → coredis.patterns.pipeline
  • coredis.stream → coredis.patterns.streams
  • coredis.cache → coredis.patterns.cache
  • coredis.lock → coredis.patterns.lock

Removals

• Drop support for RESP version 2
• Remove coredis.commands.Monitor wrapper
• Remove coredis.modules.RedisGraph module support

Compatibility

• The clients no longer check if a module is loaded at initialization. Therefore a module command issued against a server that doesn't have the module will not fail early and instead result in an coredis.exceptions.UnknownCommandError exception being raised
• Add coredis.modules.response.aggregation.AggregationType.COUNTALL and coredis.modules.response.aggregation.AggregationType.COUNTNAN aggregation types to coredis.modules.TimeSeries
• Add coredis.modules.Search.create.indexall argument to coredis.modules.Search.create
• Add coredis.Redis.vsim.nothread argument to coredis.Redis.vsim
• Add support for VRANGE vector set command (coredis.Redis.vrange)
• Add support for FT.HYBRID search command (coredis.modules.Search.hybrid)

Renamed submodules

• Move Pub/Sub, Pipeline, Stream, Cache and Lock submodules to coredis.patterns.
  • coredis.commands.pubsub -> coredis.patterns.pubsub
  • coredis.pipeline -> coredis.patterns.pipeline
  • coredis.stream -> coredis.patterns.streams
  • coredis.cache -> coredis.patterns.cache
  • coredis.lock -> coredis.patterns.lock

Compatibility

• Add countall and countnan aggregation types to coredis.modules.TimeSeries
• Add indexall argument to coredis.modules.Search.create
• Add nothread argument to coredis.Redis.vsim

Performance

• Improve pub/sub performance that had regressed in 6.0.0rc1 & 6.0.0rc2 when compared with 5.x. The root of the slowdown was the change in reading from the connection explicitly versus being fed the data sequentially by the eventloop in the previous asyncio implementation. Performance is now back to (or better) than 5.x

Feature

• Add coredis.Redis.xconsumer factory method to create a single or group stream consumer
• Add a verify_existence flag to library coredis.commands.function.wraps to optionally skip local validation and optimistically call it.
• Improve guarantees of calling subscription methods on PubStub instances. Using a subscription_timeout the acknowledgement of subscription to a topic or pattern can be guaranteed before proceeding with listening to messages.
• Cluster pipelines now support Script instances
• Allow chaininable transform method to accept inline callbacks to transform the response from the redis server
• Added retry chainable method to the response from all commands to allow using retry policies individually with a requests instead of having to apply it on all requests issued by a client.
• Retry policies now allow infinite retries & retry deadlines.
• ExponentionBackoffRetryPolicy now supports jitter

Breaking change

• Stream consumers can no longer be awaited and must be used as context managers to ensure initialization
• coredis.Redis.pubsub, coredis.RedisCluster.pubsub & coredis.RedisCluster.sharded_pubsub now only accept keyword arguments.

Compatibility

• The clients no longer check if a module is loaded at initialization. Therefore a module command issued against a server that doesn't have the module will not fail early and instead result in an coredis.exceptions.UnknownCommandError exception by raised.
• Add support for VRANGE vector set command (coredis.Redis.vrange)
• Add support for FT.HYBRID search command (coredis.modules.Search.hybrid)

Performance

• Fixed performance regression introduced in 6.0.0rc1 with sending commands to the socket. Performance over concurrent workloads is now at par with 5.x or better

Features

• Migrate entire library to anyio, adding structured concurrency and trio support.
• Pipelines auto-execute when exiting their context manager; results are accessible in a type-safe way.
• Library.wraps renamed to coredis.commands.function.wraps and now supports callbacks.
• coredis.commands.Script and coredis.commands.Function can now be called with optional callbacks to transform the raw response from redis before returning it. This is supported by their associated wraps decorators as well.
• coredis.Redis.lock & coredis.RedisCluster.lock added as a convenient accessors for the Lua-based lock: coredis.lock.Lock (Previously found in coredis.recipes.LuaLock).

Breaking Changes

• Almost all classes (clients, connection pools, PubSub, pipelines) now require being used with their async context managers for initialization/cleanup.

• Users should replace TrackingCache with coredis.cache.LRUCache when providing a cache instance to the clients. Cache size can no longer be bound by byte size and only max_keys is supported.

• All connection pools are now blocking.

• Pipelines

  • Pipelines no longer expose an explicit execute() method and instead auto-execute when leaving their context manager.
  • Removed __len__ and __bool__ methods of coredis.pipeline.Pipeline
  • Drop support for explicit management of watch, unwatch, multi in pipelines. This is replaced by the coredis.pipeline.Pipeline.watch async context manager.

• When defining type stubs for FFI for Lua scripts or library functions, keys can only be distinguished from arguments by annotating them with coredis.typing.KeyT.

Removals

• Drop support for RESP2.
• Remove Monitor wrapper.
• Remove RedisGraph module support.

Bug Fix

• Ensure username and password provided as keyword arguments to from_url are used if there are none found in the url itself.

Bug Fix

• Fix a potential race condition with the Redis client where a non blocking command could be queued on a connection that is already acquired for a blocking command.

Compatibility

• Deprecate support for legacy RESP2 protocol

Bug Fix

• Fix recursion error when retrying pipeline commands on cluster moved errors

Compatibility

• Mark redis graph module support as deprecated
• Fix compatibility with beartype >= 0.22
• Add support for python 3.14

Feature

• Add support for new stream commands (xackdel & xdelex) in redis 8.2
• Add vismember command for vector sets

Bug Fix

• Allow client to gracefully proceed when modules can't be enumerated due to ACL restrictions

Development

• Switch entirely to pyproject.toml for project metadata
• Use hatch for build
• Use uv for development environment

Bug Fix

• Ensure ssl_context passed in kwargs of from_url factory method is respected.

Bug Fix

• Fix regression caused by 5.0.0 which completely broke the use of Sentinel with decode_responses=False.

Features

• Add support for using custom types with redis commands by registering serializers and deserializers
• Allow stacking pipeline commands synchronously
• Expose statically typed responses for pipeline commands

Compatibility

• Redis command methods are no longer coroutines and instead synchronous methods that return subclasses of Awaitable (CommandRequest) which can be awaited as before.
• Add support for redis 8.0 vector set commands
• Add support for redis 8.0 hash expiry commands
• Remove deprecated pubsub listen and threaded worker APIs
• Remove support for KeyDB

Performance

• Streamline client side cache shrinking

Bug Fix

• Fix duplicate command error in using transform with pipeline

Features

• Add support for using custom types with redis commands by registering serializers and deserializers
• Allow stacking pipeline commands syncronously
• Expose statically types responses for pipeline commands

Compatibility

• Add support for redis 8.0 vector set commands
• Add support for redis 8.0 hash expiry commands
• Remove deprecated pubsub listen and threaded worker APIs
• Remove support for KeyDB

Performance

• Streamline client side cache shrinking

Bug Fix

• Add support for using library functions with pipelines

Bug Fix

• Ensure pubsub consumer task cleanly exists when the pubsub instance has been shutdown with close or aclose

Bug Fix

• Fix busy loop in pubsub consumer when it isn't subscribed to any channel or pattern.

Bug Fix

• Disable incompatible scan method for redis cluster
• Fix use of scan_iter with redis cluster with uninitialized client

Features

• Improve API for pubsub instances to:

  • Be used as Async context manager for automatic cleanup
  • Be iterated on with async for to consume messages
  • Be awaited to ensure initialization
  • subscribe on instantiation

Bug Fix

• Correct type narrowing when constructing client with from_url class method

Compatibility

• Drop support for python 3.9
• Remove tests for RESP2

Deprecations

• Deprecate run_in_thread APIs for PubSub & Monitor
• Deprecate listen methods for PubSub instances

Bug Fix

• Ensure ssl context is correctly setup with the check_hostname & cert_reqs even when a keyfile / cert are not provided. This allows clients to connect to a redis instance that has tls-auth-client=no by simply setting ssl_cert_reqs=none

v4.20.0

v4.20.0

v4.20.0

v4.20.0

Features

• Add novalues argument to hscan

Bug Fix

• Improve handling of ssl parameters (especially from query string) when dealing with self signed certificates.

v4.19.0

v4.19.0

v4.19.0

v4.19.0

Features

• Added ability to use credential providers to supply authentication details when establishing connections to the server
• Added support for hash field expiry commands introduced in redis 7.4:
  • hexpire
  • hexpireat
  • hpexpire
  • hpexpireat
  • hpersist
  • hexpiretime
  • hpexpiretime
  • httl
• Added support for maxage parameter in client_kill method.

Compatibility

• Added support for python 3.13 wheels
• Dropped support for python 3.8
• Dropped support for pypy 3.8
• Added test coverage for pypy 3.10
• Added test coverage for redis 7.4.x
• Added test coverage against redict servers
• Added test coverage for Valkey 7.0 & 8.0

Chore

• Fix artifact download in github release workflow

Release Date: 2024-12-09

• Chore

  • Fix github release workflow

Features

• Add explicit exception types for locking recipe

Bug fix

• Fix incorrect use of CLIENT SET-INFO for lib-name & lib-ver when redis version < 7.2
• Fix various incorrect type annotations in return types

Compatibility

• Update documentation dependencies
• Update test dependencies (related to pytest)

Bug fix

• Fix intermittent errors due to mismatched responses when multiple couroutines access a new connection pool.

Compatibility

• Remove support for python 3.7
• Remove Redis 6.0 from CI
• Disable RedisGraph tests in CI as the module is now not part of Redis Stack
• Fix RESP3 compatibility for RedisSearch
• Mark json.resp as deprecated

Bug fix

• Fix intermittent errors due to mismatched responses when multiple couroutines access a new connection pool.

Bug fix

• Handle edge case of clearing a closed socket buffer during object destruction.

Chores

• Re-enable CI for dragonfly

Bug fix

• Improve cleanup on socket disconnect by clearing internal response buffer

Chores

• Add typing overload for lpop method
• Remove python 3.7 from CI due to EOL.
• Temporarily disable pre-release python 3.12 from CI due to dependency resolution issues.
• Update development dependencies

Features

• Improve parsing of TS.INFO response for rules section
• Broaden input parameter type annotations when expecting a mapping to use Mapping instead of Dict

Compatibility

• Update parsing of timeseries module responses to be compatible with RESP3 responses

Chores

• Update CI to test against 7.2-rc2
• Update mypy dependency

Feature

• Add json.mset command

Bug Fix

• Remove caching for json.mget
• Ensure hint from moved errors are used on next attempt

Feature

• Add json.merge command

Bug fix

• Fix exception message for unsupported commands

Hack

• Downgrade sphinx back to 6.x for theme compatibility

Bug Fixes

• Fix incorrect explicit command execution on cluster pipeline retries
• Fix inconsistent reset api signatures for pipeline across standalone & cluster clients
• Ensure pipeline requests respect explicit timeout
• Fix incomplete xinfo_streams response for groups/consumers details

Chores

• Expand coverage of modules tests to include RESP2
• Allow failures for "next" versions of redis in CI

Feature

• Add pure python wheel to release process

Chores

• Improve redis module documentation
• Update CI to use python 3.11 as default
• Fix README formatting

Chore

• Clean up changelog entries

Bug Fix

• Ensure task cancellation results in proper cleanup of a connection that might be blocked

Features

• Add support for RedisBloom module

• Add support for ReJSON module

• Add support for RedisSearch module

• Add support for RedisTimeSeries module

• Add support for RedisGraph module

• Check argument versions for compatibility and raise appropriate errors if an argument is used on an older server version which doesn't support it.

• Expose ~coredis.Redis.retry_policy to client constructors

• Expose ~coredis.Redis.noevict in client constructors

• Add initial support for redis 7.2

  • Expose ~coredis.Redis.notouch in client constructors
  • Add support for ~coredis.Redis.client_no_touch
  • Add support for ~coredis.Redis.client_setinfo
  • Add support for ~coredis.Redis.waitaof
  • Add new withscore argument for ~coredis.Redis.zrevrank
  • Add new context manager ~coredis.Redis.ensure_persistence

• Allow adding streams to stream consumers after initialization

• Improve cluster routing for commands that act on multiple slots but are handled by the same node.

• Allow overriding the default stream_timeout when using a pipeline

Bug Fix

• Ensure multiple properties returned from info command are collapsed into an array
• Fix leaked connections when using ~coredis.Redis.ensure_replication
• Improve handling of cancellation errors
• Improve handling of timeout errors
• Ensure cluster commands routed to random nodes use primaries by default
• Handle pause/resume callbacks from Transport and pause sending subsequent commands until the transport buffer is resumed.
• Handle RESP3 response for ~coredis.Redis.command
• Update ~coredis.ConnectionPool.from_url & ~coredis.ClusterConnectionPool.from_url to support all constructor arguments

Features

• Add support for RedisGraph module
• Allow overriding the default stream_timeout when using a pipeline
• Check argument versions for compatibility and raise appropriate errors if an argument is used on an older server version which doesn't support it.

Bug Fix

• Handle pause/resume callbacks from Transport and pause sending subsequent commands until the transport buffer is resumed.
• Handle RESP3 response for ~coredis.Redis.command
• Update ~coredis.ConnectionPool.from_url & ~coredis.ClusterConnectionPool.from_url to support all constructor arguments

Chores

• Add redis-stack@edge to compatibility matrix in CI

Features

• Add support for RedisSearch module
• Allow adding streams to stream consumers after initialization

Chores

• Update mypy

Features

• Add support for RedisBloom module
• Add support for ReJSON module
• Add support for RedisTimeSeries module
• Improve cluster routing for commands that act on multiple slots but are handled by the same node.

Bug Fix

• Ensure multiple properties returned from info command are collapsed into an array

Chores

• Handle external warnings in tests
• Improve docstrings & annotations for ensure_persistence
• Add reruns for test failures in CI
• Add python 3.12 to CI

Features

• Expose retry_policy to client constructors
• Expose noevict in client constructors
• Add initial support for redis 7.2
  • Expose notouch in client constructors
  • Add support for client_no_touch
  • Add support for `client_setinfo``
  • Add support for waitaof
  • Add new withscore argument for zrank & zrevrank
  • Add new context manager ensure_persistence

Bug Fix

• Fix leaked connections when using ensure_replication
• Improve handling of cancellation errors
• Improve handling of timeout errors
• Ensure cluster commands routed to random nodes use primaries by default

Bug Fix

• Ensure task cancellation results in proper cleanup of a connection that might be blocked

Bug Fix

• Ensure protocol_version is parsed as an int from url

Chores

• Update cibuildwheels action
• Update versioneer
• Migrate setup.cfg to pyproject
• Parallelize CI wheel build
• Reintroduce ruff for linting

Chores

• Add changelog link to pypi