hatchet/sdks/python/CHANGELOG.md

Changelog

All notable changes to Hatchet's Python SDK will be documented in this changelog.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[1.21.5] - 2025-12-06

Changed

• Task outputs that fail to serialize to JSON will now raise an IllegalTaskOutputError instead of being stringified. This pulls errors from the engine upstream to the SDK, and will allow users to catch and handle these errors more easily.

[1.21.4] - 2025-12-05

Added

• Adds support for dynamic rate limits using CEL expressions (strings) for the limit parameter.

Changed

• Fixes a serialization error caused by Pydantic sometimes being unable to encode bytes, reported here: https://github.com/hatchet-dev/hatchet/issues/2601
• Fixes a bug where string-based CEL expressions for limit were rejected due to the validation logic.

[1.21.3] - 2025-11-26

Added

• Adds GZIP compression for gRPC communication between the SDK and the Hatchet engine to reduce bandwidth usage.

[1.21.2] - 2025-11-13

Added

• Adds an OTel option to allow you to include the action name in the root span name for task runs.

Changed

• Span kinds (e.g. producer, consumer) have been added to OpenTelemetry spans created by the SDK to better reflect their roles.

[1.21.1] - 2025-11-08

Changed

• The list methods for the logs client now allow for pagination via the limit, since, and until params.

[1.21.0] - 2025-10-31

Added

• Adds support for dataclasses as input validators for workflows (and tasks), and also as output validators for tasks.

Changed

• Fixes a bug where an exception in a lifespan would cause the lifespan to hang indefinitely.

[1.20.2] - 2025-10-15

Added

• Adds a include_payloads parameter to the list methods on the runs client (defaults to true, so no change in behavior).

[1.20.1] - 2025-10-14

Added

• Adds wrapper methods for bulk cancelling / replaying large numbers of runs with pagination.

[1.20.0] - 2025-10-3

Removed

• Removes all references to get_group_key_* which is no longer available in V1
• Removes all checks + references to V0

[1.19.0] - 2025-09-24

Removed

• Removed the deprecated v0 client and all related code.
• Removed unused dependencies.

[1.18.1] - 2025-08-26

Changed

• Fixes an install issue caused by a misnamed optional dependency.

[1.18.0] - 2025-08-26

Added

• Adds a stubs client on the main Hatchet client, which allows for creating typed stub tasks and workflows. These are intended to be used for triggering workflows that are registered on other workers in either other services or other languages.
• Adds a config option force_shutdown_on_shutdown_signal which allows users to forcefully terminate all processes when a shutdown signal is received instead of waiting for them to exit gracefully.

[1.17.2] - 2025-08-20

Added

• Adds back an optional cel-python dependency for v0 compatibility, allowing users to use the v0 client with the v0-compatible features in the SDK.
• Adds dependencies to the mock_run methods on the Standalone.
• Removes aiostream dependency that was unused.
• Removes aiohttp-retry dependency that was unused.

[1.17.1] - 2025-08-18

Added

• Adds a HATCHET_CLIENT_LOG_QUEUE_SIZE environment variable to configure the size of the log queue used for capturing logs and forwarding them to Hatchet

[1.17.0] - 2025-08-12

Added

• Adds support for dependency injection in tasks via the Depends class.
• Deprecated fetch_task_run_error in favor of get_task_run_error, which returns a TaskRunError object instead of a string. This allows for better error handling and debugging.

Changed

• Uses logger.exception in place of logger.error in the action runner to improve (e.g.) Sentry error reporting
• Extends the TaskRunError to include the task_run_external_id, which is useful for debugging and tracing errors in task runs.
• Fixes an issue with logging which allows log levels to be respected over the API.

Removed

• Removes the cel-python dependency

[1.16.5] - 2025-08-07

Changed

• Relaxes constraint on Prometheus dependency

[1.16.4] - 2025-07-28

Added

• Adds a new config option grpc_enable_fork_support to allow users to enable or disable gRPC fork support. This is useful for environments where gRPC fork support is not needed or causes issues. Previously was set to False by default, which would cause issues with e.g. Gunicorn setups. Can also be set with the HATCHET_CLIENT_GRPC_ENABLE_FORK_SUPPORT environment variable.

Changed

• Changes ValidTaskReturnType to allow Mapping[str, Any] instead of dict[str, Any] to allow for more flexible return types in tasks, including using TypedDict.

[1.16.3] - 2025-07-23

Added

• Adds support for filters and formatters in the logger that's passed to the Hatchet client.
• Adds a flag to disable log capture.

Changed

• Fixes a bug in aio_sleep_for and the SleepCondition that did not allow duplicate sleeps to be awaited correctly.
• Stops retrying gRPC requests on 4XX failures, since retrying won't help

[1.16.2] - 2025-07-22

Added

• Adds an input_validator property to BaseWorkflow which returns a typechecker-aware version of the validator class.

[1.16.1] - 2025-07-18

Added

• Adds a CEL feature client for debugging CEL expressions

[1.16.0] - 2025-07-17

Added

• Adds new methods for unit testing tasks and standalones, called mock_run and aio_mock_run, which allow you to run tasks and standalones in a mocked environment without needing to start a worker or connect to the engine.
• Improves exception logs throughout the SDK to provide more context for what went wrong when an exception is thrown.
• Adds create_run_ref, get_result, and aio_get_result methods to the Standalone class, to allow for getting typed results of a run more easily.
• Adds return_exceptions option to the run_many and aio_run_many methods to be more similar to e.g. asyncio.gather. If True, exceptions will be returned as part of the results instead of raising them.

Changed

• Correctly propagates additional metadata through the various run methods to spawned children.

[1.15.3] - 2025-07-14

Changed

• remove_null_unicode_character now accepts any type of data, not just strings, dictionaries, lists, and tuples. If the data is not one of these types, it's returned as-is.

[1.15.2] - 2025-07-12

Changed

• Fixes an issue in capture_logs where the log call was blocking the event loop.

[1.15.1] - 2025-07-11

Added

• Correctly sends SDK info to the engine when a worker is created

[1.15.0] - 2025-07-10

Added

• The Metrics client now includes a method to scrape Prometheus metrics from the tenant.

Changed

• The Metrics client's get_task_metrics and get_queue_metrics now return better-shaped, correctly-fetched data from the API.

[1.14.4] - 2025-07-09

Added

• Adds delete and aio_delete methods to the workflows feature client and the corresponding Workflow and Standalone classes, allowing for deleting workflows and standalone tasks.

[1.14.3] - 2025-07-07

Added

• Adds remove_null_unicode_character utility function to remove null unicode characters from data structures.

Changed

• Task outputs that contain a null unicode character (\u0000) will now throw an exception instead of being serialized.
• OpenTelemetry instrumentor now correctly reports exceptions raised in tasks to the OTel collector.

[1.14.2] - 2025-07-03

Added

• The Runs client now has list_with_pagination and aio_list_with_pagination methods that allow for listing workflow runs with internal pagination. The wrappers on the Standalone and Workflow classes have been updated to use these methods.
• Added retries with backoff to all of the REST API wrapper methods on the feature clients.

[1.14.1] - 2025-07-03

Changed

• DurableContext.aio_wait_for can now accept an or group, in addition to sleep and event conditions.

[1.14.0] - 2025-06-25

Added

• Adds an IllegalTaskOutputError that handles cases where tasks return invalid outputs.
• Logs NonRetryableException as an info-level log so it doesn't get picked up by Sentry and similar tools.

Changed

• Exports NonRetryableException at the top level
• Fixes an issue with the status field throwing a Pydantic error when calling worker.get
• Fixes an issue with duplicate protobufs if you try to import both the v1 and v0 clients.

[1.13.0] - 2025-06-25

Added

• Documentation for the Context classes
• Allows for a worker to be terminated after a certain number of tasks by providing the terminate_worker_after_num_tasks config option

Changed

• Adds a number of helpful Ruff linting rules
• DedupeViolationErr is now DedupeViolationError
• Fixed events documentation to correctly have a skipped run example.
• Changed default arguments to many methods from mutable defaults like [] to None
• Changes JSONSerializableMapping from Mapping to dict
• Handles some potential bugs related to asyncio tasks being garbage collected.
• Improves exception printing with an ExceptionGroup implementation
• Fixes a bug with namespacing of user event conditions where the namespace was not respected so the task waiting for it would hang
• Fixes a memory leak in streaming and logging, and fixes some issues with log capture.

[1.12.3] - 2025-06-25

Changed

• Fixes a namespacing-related but in the workflow.id property that incorrectly (and inconsistently) returned incorrect IDs for namespaced workflows.

[1.12.2] - 2025-06-17

Changed

• Fixes a security vulnerability by bumping the protobuf library

[1.12.1] - 2025-06-13

Added

• Adds corresponding SDK changes from API changes to events (additional parameters to filter events by, additional data returned)

[1.12.0] - 2025-06-06

Added

• Adds a warning on client init if the SDK version is not compatible with the tenant (engine) version.
• Adds a default_filters parameter to the Hatchet.workflow and Hatchet.task methods to allow you to declaratively provide a list of filters that will be applied to the workflow by default when events are pushed.
• Adds get_status and aio_get_status methods to the Runs feature client, which return a workflow run's status by its ID.
• Adds a update methods to the Filters feature client.

Changed

• Allows the concurrency parameter to tasks to be a list.
• Fixes an internal bug with duplicate concurrency expressions being set when using Hatchet.task.
• Modifies existing datetime handling to use UTC timestamps everywhere.

[1.11.1] - 2025-06-05

Changed

• Fixes a couple of blocking calls buried in the admin client causing loop blockages on child spawning

[1.11.0] - 2025-05-29

Changed

• Significant improvements to the OpenTelemetry instrumentor, including:
  • Traceparents are automatically propagated through the metadata now so the client does not need to provide them manually.
  • Added a handful of attributes to the run_workflow, push_event, etc. spans, such as the workflow being run / event being pushed, the metadata, and so on. Ignoring
  • Added tracing for workflow scheduling

[1.10.2] - 2025-05-19

Changed

• Fixing an issue with the spawn index being set at the workflow_run_id level and not the (workflow_run_id, retry_count) level, causing children to be spawned multiple times on retry.

[1.10.1] - 2025-05-16

Added

• Adds an otel item to the ClientConfig and a excluded_attributes: list[OTelAttribute] there to allow users to exclude certain attributes from being sent to the OpenTelemetry collector.

[1.10.0] - 2025-05-16

Added

• The main Hatchet client now has a filters attribute (a Filters client) which wraps basic CRUD operations for managing filters.
• Events can now be pushed with a priority attribute, which sets the priority of the runs triggered by the event.
• There are new list and aio_list methods for the Events client, which allow listing events.
• Workflow runs can now be filtered by triggering_event_external_id, to allow for seeing runs triggered by a specific event.
• There is now an id property on all Workflow objects (Workflow created by hatchet.workflow and Standalone created by hatchet.task) that returns the ID (UUID) of the workflow.
• Events can now be pushed with a scope parameter, which is required for using filters to narrow down the filters to consider applying when triggering workflows from the event.

Changed

• The name parameter to hatchet.task and hatchet.durable_task is now optional. If not provided, the task name will be the same as the function name.