fix(runners): dispatch run_after_run_callback in _run_node_async (#5282)

[surfai]

Summary

Fixes #5282 by wiring plugin_manager.run_after_run_callback into Runner._run_node_async, closing the runners.py:427 TODO for node runtime plugin lifecycle.

Companion to test-only PR #5301 (which established the xfail regression anchor on stock 2.0.0a3). This PR ships the source fix plus flips the anchor by replacing the WorkaroundRunner scaffolding with a direct positive assertion.

The fix (3 lines)

try:
  async for event in self._consume_event_queue(ic, done_sentinel):
    yield event
  # 5. Run the after_run callbacks. Mirrors _exec_with_plugin:1230.
  # This does NOT emit any event.
  await ic.plugin_manager.run_after_run_callback(invocation_context=ic)
finally:
  await self._cleanup_root_task(task, self.agent.name)

Placement mirrors the legacy dispatch at runners.py:1230 exactly: outside the finally block, so semantics match legacy —

Exit path	Legacy _exec_with_plugin	This fix
Natural drain	✓ fires	✓ fires
Error in loop	✗ skipped	✗ skipped (exception jumps to finally)
Caller break	✗ skipped	✗ skipped (GeneratorExit at yield jumps to finally)

Full semantic equivalence. Single insertion point covers both LlmAgent and BaseNode root types because run_async Branches A (line 807) and B (line 838) both funnel into _run_node_async. This also removes the last pre-condition to collapsing the two branches per the TODO at runners.py:836-837.

Also in this PR

• Narrowed TODO at runners.py:427 from "tracing and plugin lifecycle" to "tracing" — plugin lifecycle is now wired, tracing remains out of scope for this fix.
• Incidental pyink reformat at runners.py:1451 — pre-existing 82-char line on v2 HEAD, unrelated to Custom memory persistence silently no-ops on Workflow (BaseNode) roots in 2.x — no working post-run hook #5282, but required by the pyink CI gate since the workflow runs pyink --check on the whole changed file. Keeping this out of a separate PR to avoid churn.
• Test file tests/unittests/runners/test_issue_5282.py — replaces the xfail-anchored version from test: regression tests for #5282 (after_run_callback on BaseNode roots) #5301 with two positive assertions:
  • test_workflow_root_dispatches_pre_run_and_event_hooks (baseline)
  • test_workflow_root_dispatches_after_run_callback (regression anchor, no xfail)

Alternatives considered and rejected

1. Dispatch in finally block — would fire after_run_callback on error and caller-break, changing the legacy contract. Plugin authors wrote handlers expecting "successful completion" semantics; flipping to "always fires" is a breaking change.
2. Lift dispatch to run_async call sites (lines 825, 841) — ic is created inside _run_node_async at line 446, so this would require plumbing ic out of the generator. Expands blast radius.
3. Extract _with_plugin_lifecycle(ic) shared context manager — right long-term shape, but refactors the legacy _exec_with_plugin too (which has early-exit logic inside before_run_callback that's awkward to extract). Out of scope for a fix PR.

Test plan

• pytest tests/unittests/runners/test_issue_5282.py -v — 2 passed
• pytest tests/unittests/runners/ (full subdir) — 69 passed, 1 skipped, 5 xfailed (all pre-existing), 0 new failures
• isort --check — clean
• pyink --check --config pyproject.toml — clean
• CI: pyink.yml, isort.yml, python-unit-tests.yml, mypy.yml, mypy-new-errors.yml

Links

• Issue: Custom memory persistence silently no-ops on Workflow (BaseNode) roots in 2.x — no working post-run hook #5282
• Companion test PR: test: regression tests for #5282 (after_run_callback on BaseNode roots) #5301
• Related: before_model_callback and after_model_callback not invoked for live streaming sessions (run_live) #4181 (same structural pattern on the run_live path)
• Related: Add Lifecycle Error Callbacks (on_agent_error, on_run_error) to ADK Framework #4774 (motivates why after_run_callback is load-bearing for enterprise telemetry)

🤖 Generated with Claude Code

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.