[zzylol]

Why

Tests whose result depends on a wall-clock budget fail on a slow or loaded machine. For example, structural_seed_timeout_preserves_validated_incumbent sleeps 1.1 s against a 1 s search budget; it passed 3 of 3 runs alone and failed when two other builds shared the CPU. #1166 ignored the nine such tests that had flaked on CI. Stage 3 is being replaced (#1095), and #1067 moves the search to optimal-only stopping, so until then every test of this kind is ignored, not only those seen flaking.

What

36 tests get #[ignore = "flaky: depends on wall-clock search budgets (#1067)"] or #[ignore = "flaky: depends on a wall-clock deadline (#1067)"]. They run with cargo test -- --ignored.

• Search budgets, a real search under wall_time / time_limit_ms whose result can change when the budget is hit:
  • plan-selection refinement/planner.rs: 6 tests;
  • refinement/search.rs: the 2 tests that sleep past their budget;
  • integration-tests: finer_aggregate_reuse, identical_exact_queries (2), independent_plan_seed (3), joint_search_equivalence (7), memo_enumeration (1), production_rollup_workloads (4), sharing_issue_regressions (1);
  • planner runtime_install.rs: 2 tests.
• Deadlines, Instant::now() plus a few ms or s, expecting expiry or completion:
  • executor physical_planner (2);
  • logical-asap-optimization shared_configurations.rs (1);
  • types flat.rs (1), tests/pane_chain.rs (2), tests/physical_export.rs (1).

Not ignored:

• tests whose deadline is already expired (Some(Instant::now()), Duration::ZERO, a zero budget), which are deterministic;
• search.rs / materialization.rs / objectives.rs unit tests whose fake evaluators or tiny graphs return far inside their 5–30 s budgets;
• the budget_exhausted flag test.

How

The tests were selected by a scan of every #[test] / #[tokio::test] whose body, or a helper in the same file that it calls (transitively), sets a real budget (wall_time: Duration::from_*(n>0), time_limit_ms: Some(..), thread::sleep, Instant::now() + Duration). The list was then triaged by hand into the groups above.

Key code interfaces added / modified

None (test attributes only).

Major algorithm changes

None.

Before this PR

These tests could fail on a loaded CI runner or developer machine, on branches unrelated to them.

After this PR

They are skipped by default and still run with --ignored.

Verification

• cargo fmt --all --check: clean.
• cargo check --tests for asap-types, asap-executor, asap-logical-asap-optimization, asap-plan-selection, asap-planner, asap-integration-tests: clean.
• cargo test -p asap-plan-selection --lib -- --ignored --list lists the 8 newly ignored plan-selection tests.