Merge pull request #3364 from microsoft/danielmeppiel-intel-release-budget

fix(ci): budget focused Intel release qualification

Author: danielmeppiel

.github/workflows/build-release.yml

@@ -253,7 +253,10 @@ jobs:
         run: |
           chmod +x scripts/test-integration.sh
           uv run ./scripts/test-integration.sh
-        timeout-minutes: 30
+        # Run 39834866538 timed out at 30m with workers still passing at 91%.
+        # 30 / 0.91 projects roughly 33m; 60m adds setup/tail headroom,
+        # not a completion guarantee. Keep selection and failure gates intact.
+        timeout-minutes: 60
 
       # PHASE 3: RELEASE VALIDATION (tag/schedule/dispatch only)
       - name: Prepare isolated release-validation environment

docs/src/content/docs/contributing/integration-testing.md

@@ -521,9 +521,13 @@ The Linux x64 and macOS Apple Silicon full integration steps have 120-minute
 limits and keep `not live` selection with `-n 4 --dist loadgroup` for HOME-mutating
 test grouping. Previous 60-minute limits stopped progressing workers at 78% and
 77%, respectively; the budgets allow setup and tail headroom without excluding
-tests. Linux ARM keeps 60 minutes, Windows 20, and the focused Intel step 30.
-Failures and timeouts still block release publication; a larger budget is not
-evidence that a failed assertion is fixed.
+tests. Linux ARM keeps 60 minutes and Windows 20. The focused Intel step has
+60 minutes: its prior 30-minute limit stopped passing workers at 91%.
+The rough `30 / 0.91` projection is 33 minutes; 60 adds setup/tail headroom,
+not a completion guarantee.
+Its `lifecycle_smoke and not live` selection and four `loadgroup` workers are
+unchanged. Failures and timeouts still block release publication; a larger
+budget is not evidence that a failed assertion is fixed.
 
 The existing unprivileged `PR Binary Smoke (Linux)` job also runs
 `tests/integration/test_audit_workspace_identity_lifecycle.py` against its built

tests/unit/test_platform_contract_workflow.py

@@ -160,7 +160,7 @@ def _assert_intel_focused_integration(workflow: dict) -> None:
     )
     assert step["env"].get("PYTEST_MARK_EXPR") == INTEL_FOCUSED_MARK_EXPRESSION
     assert step["env"].get("PYTEST_EXTRA_ARGS") == NON_LIVE_UNIX_PYTEST_ARGS
-    assert step.get("timeout-minutes") == 30
+    assert step.get("timeout-minutes") == 60
     assert_exact_command(
         shell_commands(step),
         ["uv", "run", "./scripts/test-integration.sh"],
@@ -293,6 +293,24 @@ def test_intel_non_live_corpus_mutation_is_rejected() -> None:
         _assert_intel_focused_integration(workflow)
 
 
+@pytest.mark.parametrize("timeout_minutes", (None, 30, 61))
+def test_intel_missing_old_or_oversized_timeout_mutation_is_rejected(
+    timeout_minutes: int | None,
+) -> None:
+    """Intel cannot lose its bound or regress to the exhausted 30-minute limit."""
+    workflow = deepcopy(_workflow())
+    step = workflow_step(
+        workflow_job(workflow, "build-and-validate-macos-intel"),
+        INTEL_FOCUSED_INTEGRATION_STEP,
+    )
+    del step["timeout-minutes"]
+    if timeout_minutes is not None:
+        step["timeout-minutes"] = timeout_minutes
+
+    with pytest.raises(AssertionError):
+        _assert_intel_focused_integration(workflow)
+
+
 def test_arm_focused_subset_mutation_is_rejected() -> None:
     """ARM remains the broad non-live macOS release authority."""
     workflow = deepcopy(_workflow())