Repository navigation
Releases
v1.13.1
What's Changed
- Add DTS Dashboard-compatible v2 Blob references and explicit access scopes by Bernd Verst (@berndverst) in #336
Warning
Older Python SDK versions cannot read new blob:v2 references. Upgrade all Python readers before enabling new writers. Existing references outside configured payload access scopes or containing . / .. path segments now fail during replay and retrieval. Review the upgrade and migration guidance before upgrading.
PyPI: durabletask 1.13.1
Full Changelog: v1.13.0...v1.13.1
azuremanaged-v1.13.1
What's Changed
- Add DTS Dashboard-compatible v2 Blob references and explicit access scopes by Bernd Verst (@berndverst) in #336
Warning
Older Python SDK versions cannot read new blob:v2 references. Upgrade all Python readers before enabling new writers. Existing references outside configured payload access scopes or containing . / .. path segments now fail during replay and retrieval. Review the upgrade and migration guidance before upgrading.
PyPI: durabletask.azuremanaged 1.13.1
Full Changelog: azuremanaged-v1.13.0...azuremanaged-v1.13.1
azurefunctions-v2.0.0rc5
What's Changed
- Add DTS Dashboard-compatible v2 Blob references and explicit access scopes by Bernd Verst (@berndverst) in #336
Warning
Older Python SDK versions cannot read new blob:v2 references. Upgrade all Python readers before enabling new writers. Existing references outside configured payload access scopes or containing . / .. path segments now fail during replay and retrieval. Review the upgrade and migration guidance before upgrading.
PyPI: azure-functions-durable 2.0.0rc5
Full Changelog: azurefunctions-v2.0.0rc4...azurefunctions-v2.0.0rc5
v1.13.0
What's Changed
- fix(worker): close orchestrator generators when a dispatch ends by Tamir Dresher (@tamirdresher) in #300
- fix(worker): preserve dispatch outcomes when generator teardown raises CancelledError by Tamir Dresher (@tamirdresher) in #303
- Preflight all externalized payload sizes before uploads by andystaples in #305
- Complete permanent payload failures with entity-lock cleanup by andystaples in #306
- feat(client): add task hub name filter to orchestration queries by Hardik K (@hardikkaurani) in #295
- fix(worker): recover unexpected orchestration dispatch exits by andystaples in #307
- fix(entities): implement hash and repr for EntityInstanceId by Hardik K (@hardikkaurani) in #309
- fix(serialization): support set and frozenset generic containers in JsonDataConverter by Hardik K (@hardikkaurani) in #310
- Optimize deterministic set serialization without JSON parse round trips by andystaples in #313
- Drop Python 3.10 support and adopt Python 3.11 features by Bernd Verst (@berndverst) in #302
- Honor non-retriable activity and sub-orchestration failures by andystaples in #318
New Contributors
- Tamir Dresher (@tamirdresher) made their first contribution in #300
- Hardik K (@hardikkaurani) made their first contribution in #295
PyPI: durabletask 1.13.0
Full Changelog: v1.12.0...v1.13.0
azuremanaged-v1.13.0
What's Changed
- fix(worker): close orchestrator generators when a dispatch ends by Tamir Dresher (@tamirdresher) in #300
- fix(worker): preserve dispatch outcomes when generator teardown raises CancelledError by Tamir Dresher (@tamirdresher) in #303
- Preflight all externalized payload sizes before uploads by andystaples in #305
- Complete permanent payload failures with entity-lock cleanup by andystaples in #306
- feat(client): add task hub name filter to orchestration queries by Hardik K (@hardikkaurani) in #295
- fix(worker): recover unexpected orchestration dispatch exits by andystaples in #307
- fix(entities): implement hash and repr for EntityInstanceId by Hardik K (@hardikkaurani) in #309
- fix(serialization): support set and frozenset generic containers in JsonDataConverter by Hardik K (@hardikkaurani) in #310
- Optimize deterministic set serialization without JSON parse round trips by andystaples in #313
- Drop Python 3.10 support and adopt Python 3.11 features by Bernd Verst (@berndverst) in #302
- Honor non-retriable activity and sub-orchestration failures by andystaples in #318
New Contributors
- Tamir Dresher (@tamirdresher) made their first contribution in #300
- Hardik K (@hardikkaurani) made their first contribution in #295
PyPI: durabletask.azuremanaged 1.13.0
Full Changelog: azuremanaged-v1.12.0...azuremanaged-v1.13.0
azurefunctions-v2.0.0rc4
What's Changed
- [Azure Functions] Add opt-in Unicode JSON serialization by Bernd Verst (@berndverst) in #292
- fix(worker): close orchestrator generators when a dispatch ends by Tamir Dresher (@tamirdresher) in #300
- fix(worker): preserve dispatch outcomes when generator teardown raises CancelledError by Tamir Dresher (@tamirdresher) in #303
- Preflight all externalized payload sizes before uploads by andystaples in #305
- Complete permanent payload failures with entity-lock cleanup by andystaples in #306
- feat(client): add task hub name filter to orchestration queries by Hardik K (@hardikkaurani) in #295
- fix(worker): recover unexpected orchestration dispatch exits by andystaples in #307
- fix(entities): implement hash and repr for EntityInstanceId by Hardik K (@hardikkaurani) in #309
- fix(serialization): support set and frozenset generic containers in JsonDataConverter by Hardik K (@hardikkaurani) in #310
- Optimize deterministic set serialization without JSON parse round trips by andystaples in #313
- Drop Python 3.10 support and adopt Python 3.11 features by Bernd Verst (@berndverst) in #302
- Honor non-retriable activity and sub-orchestration failures by andystaples in #318
New Contributors
- Tamir Dresher (@tamirdresher) made their first contribution in #300
- Hardik K (@hardikkaurani) made their first contribution in #295
PyPI: azure-functions-durable 2.0.0rc4
Full Changelog: azurefunctions-v2.0.0rc3...azurefunctions-v2.0.0rc4
v1.12.0
What's Changed
- Add Python 3.15 support, by Bernd Verst (@berndverst) in #284
- Fix
when_allhanging when pending tasks are shared between groups or appear more than once in a group. Results preserve input order and duplicate positions without scheduling additional activities. Completedwhen_anygroups stop observing losing children without disturbing other groups, and throwing completion callbacks no longer prevent other observers from being notified. By andystaples in #286 - Reconstruct dataclasses with
init=Falsefields as their declared type rather than falling back to a dictionary. Handwritten constructors still receive serialized non-init fields they accept as keywords; other non-init fields use defaults, default factories, or__post_init__and may reset recorded values. Use an explicitfrom_json()hook to preserve recorded state that the constructor cannot accept. By Shubham Padkonde (@Shubham-Padkonde) in #283 - Explicitly target orchestrations rather than entities in single-instance
purge_orchestration()requests from both synchronous and asynchronous clients, by andystaples in #279 - Prevent timer callbacks from scheduling additional long-timer chunks, retrying activities or sub-orchestrations, or resuming orchestrator code after completion, failure, termination, or continue-as-new. Work scheduled before the terminal state is preserved. By andystaples in #280
- Preserve the global arrival order of unconsumed buffered external events across different event names when using
continue_as_new(..., save_events=True), by andystaples in #278 - Carry external events arriving after
continue_as_new(..., save_events=True)into the next execution instead of losing them to abandoned waits. Events already delivered to live waits are not carried over, andsave_events=Falsestill discards unprocessed events. By andystaples in #281
Warning
Replay-breaking bug fix: Shared nested races can select a different direct child during replay. For example, with first = when_any([timer, work]) and second = when_any([first, timer]), timer completion now notifies first, allowing second to select first rather than timer. Existing downstream history can then fail replay with NonDeterminismError. Disjoint task groups are unchanged.
Drain affected instances on the previous SDK, or keep their task hub on the previous SDK and use a separate task hub for new instances. Already-stuck instances may need deliberate recovery. Pin or lock the core durabletask dependency, including when using a provider. The coordinated provider releases require durabletask v1.12.0 or later and adopt this correction.
External Links
New Contributors
- Shubham Padkonde (@Shubham-Padkonde) made their first contribution in #283
Full Changelog: v1.11.0...v1.12.0
azuremanaged-v1.12.0
What's Changed
- Add Python 3.15 support, by Bernd Verst (@berndverst) in #284
- Inherit corrected shared and duplicate pending-task completion from the core SDK:
when_allgroups no longer remain waiting after their children complete. By andystaples in #286 - Explicitly target orchestrations rather than entities in single-instance
purge_orchestration()requests from synchronous and asynchronous Azure Managed clients, by andystaples in #279 - Prevent timer callbacks from retrying activities or sub-orchestrations, or resuming orchestrator code after completion, failure, termination, or continue-as-new. Azure Managed continues to use native long timers without chunking. By andystaples in #280
- Preserve the global arrival order of unconsumed buffered external events across different event names when using
continue_as_new(..., save_events=True), by andystaples in #278 - Carry external events arriving after
continue_as_new(..., save_events=True)into the next execution instead of losing them to abandoned waits, by andystaples in #281 - Update the minimum
durabletaskdependency to 1.12.0, including theazure-blob-payloadsextra, by andystaples in #287
Warning
Replay-breaking bug fix: The shared-task correction can change shared nested race winners during replay and cause NonDeterminismError for existing downstream histories. Disjoint task groups are unchanged. See the core changelog for the affected scenario and upgrade guidance.
Drain affected instances on the previous SDK, or keep their task hub on the previous SDK and use a separate task hub for new instances. Already-stuck instances may need deliberate recovery. This release requires durabletask v1.12.0 or later and adopts this correction. Pin or lock the core durabletask dependency as well as the provider to control future core upgrades.
External Links
New Contributors
- Shubham Padkonde (@Shubham-Padkonde) made their first contribution in #283
Full Changelog: azurefunctions-v2.0.0rc2...azuremanaged-v1.12.0
azurefunctions-v2.0.0rc3
What's Changed
- Add Python 3.15 support. The minimum supported Python version remains 3.13. By Bernd Verst (@berndverst) in #284
- Inherit corrected shared and duplicate pending-task completion from the core SDK: native
when_alland compatibilitycontext.task_all()groups no longer remain waiting after their children complete. By andystaples in #286 - Explicitly target orchestrations rather than entities in single-instance
purge_orchestration()requests fromDurableFunctionsClientandSyncDurableFunctionsClient, including thepurge_instance_history()compatibility alias, by andystaples in #279 - Prevent timer callbacks from scheduling additional long-timer chunks, retrying activities or sub-orchestrations, or resuming orchestrator code after completion, failure, termination, or continue-as-new. This applies to both native and compatibility orchestration APIs. By andystaples in #280
- Preserve the global arrival order of unconsumed buffered external events across different event names when using
continue_as_new(..., save_events=True), by andystaples in #278 - Preserve external events arriving after
continue_as_new(..., save_events=True)in native durabletask orchestrators instead of losing them to abandoned waits, by andystaples in #281 - Update the minimum
durabletaskdependency to 1.12.0, by andystaples in #287
Warning
Replay-breaking bug fix: The shared-task correction can change shared nested when_any / context.task_any() race winners during replay and cause NonDeterminismError for existing downstream histories. Disjoint task groups are unchanged. This warning also applies when upgrading from earlier 2.x prereleases, including release candidates. See the core changelog for the affected scenario and upgrade guidance.
Drain affected instances on the previous SDK, or keep their task hub on the previous SDK and use a separate task hub for new instances. Already-stuck instances may need deliberate recovery. This release requires durabletask v1.12.0 or later and adopts this correction. Pin or lock the core durabletask dependency as well as the provider to control future core upgrades.
External Links
New Contributors
- Shubham Padkonde (@Shubham-Padkonde) made their first contribution in #283
Full Changelog: azurefunctions-v2.0.0rc2...azurefunctions-v2.0.0rc3
v1.11.0
What's Changed
- Keep asynchronous Azure Blob payload uploads and downloads responsive by moving gzip compression and decompression off the calling event loop, by andystaples in #270
- Fix nested
when_allandwhen_anytasks leaving their enclosing composite waiting, includingwhen_any([cancel, when_all(tasks)]), by Tsuyoshi Ushio (@TsuyoshiUshio) in #267
Warning
Replay-breaking bug fix: The nested-composite correction can change the winner of a nested race when replaying an existing orchestration. An inner when_all could previously finish without completing its enclosing when_any, allowing a later cancellation to win. If downstream actions for that cancellation branch were already recorded, replay can now take a different branch and fail with NonDeterminismError. Ordinary, non-nested when_all and when_any usage is unchanged; not every nested-composite history is affected.
Before upgrading, allow affected instances to finish on the previous SDK, or keep their task hub on the previous SDK and use a separate task hub for new instances. Instances already stuck because of this bug may need deliberate recovery rather than waiting to drain. Pin or lock durabletask while planning the transition, including when using it through a provider.
External Links
Full Changelog: v1.10.1...v1.11.0