when_all waits for all tasks before surfacing the first failure

[acroca]

Description

when_all now waits for every task to finish and then surfaces the first error, instead of completing on the first failure while sibling tasks were still running.

This fixes several rough edges: completions arriving after a failure were silently dropped, so get_completed_tasks() and pending_tasks reported stale counts from the moment a child failed; and an orchestrator that caught the failure (e.g. to run compensation) would do so while fanned-out activities were still executing. Now the counts stay accurate for the composite's whole lifetime, and by the time the error is raised into the orchestrator, all fanned-out work is known to be finished. When multiple children fail, the first failure is the one reported. The failure is staged internally until the last child completes, so the task never reports is_failed while it is still incomplete.

Note this is a behavior change: a fan-out that hits a failure now runs its remaining activities to completion before the workflow fails, instead of failing immediately.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82.83%. Comparing base (3457d0a) to head (6780640).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1166      +/-   ##
==========================================
+ Coverage   82.78%   82.83%   +0.05%     
==========================================
  Files         123      123              
  Lines       10127    10128       +1     
==========================================
+ Hits         8384     8390       +6     
+ Misses       1743     1738       -5     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR changes the durabletask when_all composite task semantics so that it waits for all child tasks to finish before completing, and only then surfaces the first observed failure (if any). This aligns task completion accounting with real runtime state and ensures orchestrators that catch failures do so only after all fan-out work has concluded.

Changes:

• Update WhenAllTask to stage the first child failure internally and defer failing the composite until all children complete.
• Expand/adjust unit tests and executor-level tests to assert the new “defer failure” behavior and first-failure selection.
• Update public-facing docstrings for when_all to reflect the behavior change.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
dapr/ext/workflow/_durabletask/task.py	Implements deferred-failure behavior in WhenAllTask by staging the first exception until all children complete.
dapr/ext/workflow/dapr_workflow_context.py	Updates the when_all wrapper docstring to document deferred failure surfacing.
tests/ext/workflow/durabletask/test_task.py	Adds/updates task-level tests for staged failures, multiple failures, and pre-failed children.
tests/ext/workflow/durabletask/test_orchestration_executor.py	Adjusts orchestration executor tests to validate no early completion/actions and correct failure surfacing timing.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.