Python: preserve tool span context for parallel calls

[ByteWise (2830500285)]

Motivation and Context

Fixes #6357.

Parallel function calls are scheduled together during tool invocation. Each scheduled tool execution should inherit the active agent invocation context so its execute_tool span stays correctly parented and trace-linked when multiple tool calls run in the same turn.

Description

This updates the parallel tool invocation path to create each task from an explicit copy of the current contextvars context before gathering the results.

It also adds an observability regression test that drives two parallel tool calls through an agent and asserts both execute_tool spans are present, share the agent trace, and use the agent span as parent.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the Contribution Guidelines
• All unit tests pass, and I have added new tests where possible (targeted observability tests pass; the full core unit suite currently has unrelated local failures for missing agent_framework_tools shell imports and one hosted-environment import-mock test)
• Is this a breaking change? No

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds a regression test and updates parallel tool invocation to preserve active tracing context across concurrent executions.

Changes:

• Add a new observability test validating parallel tool execution spans are nested under the agent invoke span.
• Update parallel tool execution to create asyncio tasks under a copied contextvars context to preserve tracing context propagation.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
python/packages/core/tests/core/test_observability.py	Adds a regression test covering parent/trace relationships for parallel tool spans.
python/packages/core/agent_framework/_tools.py	Wraps task creation with contextvars.copy_context() to keep active span context in parallel tool execution.

[github-actions]

Python Test Coverage Report •

File	Stmts	Miss	Cover	Missing
packages/core/agent_framework
_tools.py	1162	83	92%	227–228, 405, 407, 420, 445–447, 455, 473, 487, 494, 501, 524, 526, 533, 541, 670, 704–706, 709–711, 713, 719, 770–772, 797, 823, 827, 865–867, 871, 893, 1044, 1056, 1063–1066, 1087, 1095, 1109–1111, 1461, 1553, 1581, 1603, 1611, 1669, 1716, 1723–1724, 1783, 1787, 1840, 1901–1902, 1913, 1985, 1999, 2002, 2015, 2018, 2041, 2048, 2058, 2062, 2139, 2192, 2214, 2270, 2349, 2546, 2612–2613, 2779–2780, 2871
TOTAL	44095	5274	88%

Python Unit Test Overview

Tests	Skipped	Failures	Errors	Time
8826	33 💤	0 ❌	0 🔥	2m 20s ⏱️

[Eduard van Valkenburg (eavanvalkenburg)]

ByteWise (@2830500285) there have been some updates in the underlying codebase, so some checks are failing because of that, could you check them?

[ByteWise (2830500285)]

Pushed 805f419 to address the failing checks from the new test:\n\n- Match the mock chat client's _inner_get_response signature with BaseChatClient (Sequence/Mapping).\n- Assert span contexts before accessing span IDs/trace IDs.\n- Align the test-only type suppressions with the adjacent tool invocation telemetry test.\n\nLocal validation:\n- uv run pytest packages/core/tests/core/test_observability.py -k parallel_function_call_spans_nested_under_agent_span\n- uv run poe lint -P core\n- uv run poe fmt -P core\n- uv run poe test-typing -P core now has no diagnostics from this new test; my local run still reports the existing environment-only ty resolution issue for graphviz.backend.execute in tests/workflow/test_viz.py.\n\nThe new GitHub Actions runs for this commit are currently marked action_required with no jobs, so they likely need maintainer approval before CI can execute.