[zzylol]

Why

The test suite had grown slow, padded and stale in places:

• Three tests took most of the debug CI time. Example 4's monthly() variant (126 s, 67,228 candidates) and the exhaustive-limit devtool test (about 30 s) planned far more than they checked. The metrics-observability corpus test (41 s) only asserted that its own counters add up.
• Twelve tests repeated a stronger test elsewhere, and two assertions could not fail.
• Four ignored tests could not pass by design, or were replaced by newer tests.
• Several ignore reasons cited merged PRs or "no PR yet".
• Test helpers still used the retired post-ASAP names.
• The DAG viewer's Python and Node tests did not run in CI at all.

What

• Slow tests

  • The metrics-observability test keeps its POST_ASAP_MISS report and now ratchets each corpus at today's counts: at least N queries with an ASAP candidate, and at most M Stage 1 errors. A regression in coverage now fails the test. It is renamed benchmark_corpora_are_total_and_keep_their_coverage.
  • Example 4's two constraint tests run once, at_rest and Pattern B in CI. The monthly() variant moves into two #[ignore = "slow: run in release by hand"] tests: …monthly_ingestion_time_upstream_is_ingestion_time and …monthly_materialized_output_covers_its_consumers.
  • exhaustive_default_candidate_limit_is_ten_thousand becomes a unit test in the stage_pipeline binary that asserts on MAX_COMBINATIONS. workload_above_the_candidate_limit_is_rejected stays.

• Duplicate tests deleted. Each one was compared line by line with the test that stays.

  Deleted	Kept (equal or stronger)
  promql_conformance::rate_range_lives_in_time_range_node	promql_lowering::rate_has_time_range_child_not_window (identical), time_range::q13_rate (full DAG equality)
  promql_conformance::increase_range_lives_in_time_range_node	time_range::q14_increase (full DAG equality)
  promql_conformance::sum_of_rate_is_two_levels	promql_lowering::sum_over_rate_keeps_both_levels (also checks the inner TimeRange)
  promql_conformance::sum_by_of_rate_groups_outer_level	promql_lowering::sum_by_over_rate_groups_the_outer_sum
  promql_conformance::histogram_quantile_over_sum_by_le_preserves_le_grouping	promql_lowering::histogram_quantile_over_sum_by_le_preserves_grouping (also checks q)
  promql_conformance::quantile_over_time_is_aggregate_over_time_range	promql_lowering::quantile_over_time_is_time_range_aggregate, time_range::q17_quantile_over_time
  promql_conformance::topk_over_count_is_heavy_hitter	promql_lowering::topk_over_count_is_heavy_hitter_topk (grouped, full chain), fractional_topk_k_truncates_toward_zero (ungrouped)
  promql_conformance::outer_group_key_absent_from_nested_aggregate_is_dropped	nested::q53_outer_group_key_absent_from_nested_aggregate (full DAG equality)
  promql_conformance::range_vector_selector_is_time_range	promql_lowering::explicit_range_selector_is_a_range_time_range, which now also asserts the bare selector's 5-min range (the one check it lacked)
  awesome_prometheus_alerts::vector_literal_lowers_to_a_labelless_vector	promql_conformance::vector_promotes_a_scalar_to_a_vector (also checks the value field)
  promql_lowering::count_over_time_is_count_intent	time_range::q15_count_over_time (full DAG equality)
  promql_lowering::stddev_and_stdvar_over_time	time_range::q_stddev_over_time, q_stdvar_over_time (full DAG equality)

• Assertions that cannot fail, deleted (each test stays)

  • operator_design_examples.rs: !json.contains("KeepPreAsap"/"ScalarBridge"). Neither name exists anywhere in the workspace.
  • frontend-promql/src/error.rs: !msg.contains("PromQL"), right after assert_eq!(msg, "unsupported feature: subquery").

• Dead helper deleted: asap_types::ir::cse::dag_node_count and its 3 tests. It had no non-test caller here, in ASAPQuery or in ASAPCollector. Its one other test use is now OperatorNode::reachable(..).len(). The CSE cost-model decision doc names that expression instead.

• Stale ignored tests deleted. I ran each one first and each still fails.

  • operator_sharing::project_above_and_scan_below_a_summary_are_both_non_asap_nodes: the IR now guarantees what it checks (Flatten ASAP DAG Operators for Both Pre-ASAP and Post-ASAP IR #468). It fails because the plan holds no summary evaluation. Its two helpers went with it.
  • planner_layering_example1::stage2_physical_asap_optimization_every_candidate_compiles_in_the_physical_planner: it cites the merged fix(ir): top-k sketch readouts return selected rows; Example 1 expectations follow planner output #579, and unconfirmed candidates can't compile by design.
  • planner_layering_example4::stage3_plan_selection_b_rebuilding_every_window_costs_most and …b_prefers_ingestion_time_tumbling_windows: replaced by …b_cost_models_choose_differently and …b_pareto_front_over_memory_disk_and_latency. The acceptance doc's rows are removed.

• Ignore reasons retargeted. Each test was rerun to confirm its current failure.

  • Sliding windows in examples 3 and 4: "no PR yet" → Pass 2: sliding-window and Exponential Histogram composition — after #511 #522.
  • count_planning::grouped_count_offers_hydra_but_selection_keeps_per_group_state and example1::stage1_logical_asap_optimization_q2_offers_hydra: feat(planner): Pass 1 offers HydraCms for grouped approximate counts #605 → Hydra follow-ups: group totals in the executor, Pass 1 HydraCms candidates, Hydra top-k #703.
  • Example 2:
    • the specialized entropy/L2 families → [Refactor] Model UnivMon multi-metric support as part of the sketch strategy #444;
    • both UnivMon accuracy tests → UnivMon accuracy rule: use the paper guarantee (1±ε, failure probability 0.3) #788. The shared-UnivMon test now fails because its candidate is unpriced, not on materialization (Stage 2 materialization: ingestion vs query time per summary #604 is merged).
  • operator_sharing::each_side_of_union_all_holds_a_summary_estimate: Stage 2 materialization: ingestion vs query time per summary #604 → Pass 2 sharing and Stage 1 coverage parity with the retired MajorPass #580. The exact plan is still selected.
  • weighted_topk_binding::planner_weighted_topk_binds_at_either_deployment_phase: fix(ir): top-k sketch readouts return selected rows; Example 1 expectations follow planner output #579 → Stage 1 whole-expression top-k heap keeps outer partition keys indexed against the absorbed aggregate #639. That open issue describes this exact test's whole-expression heap key indexing, and today the CountSketch variant fails with "invalid keyed evaluation shape". The reason also notes that the unconfirmed Count-Min heap has no non-negative weight proof. The other two ignored tests in this file already cite the open Top-k results drop the item columns (port/job); heavy-hitter candidates resolve columns by name #672.

• Mechanical renames to current terms:

  Old	New
  asap_integration_tests::post_asap	asap_integration_tests::physical_export
  post_asap_dag (integration tests, frontend-promql test support)	physical_asap_dag
  at_ingestion_post_asap_dag	at_ingestion_physical_asap_dag
  bind_post_asap_before_execution	bind_physical_asap_dag_before_execution
  post_asap_summary_projection_survives_recovery	summary_projection_survives_recovery
  physical_dag.rs comment "A literal Fallback"	"A literal Values source"

• CI: the test job now runs the DAG viewer's Python tests (python3 -m unittest test_viewer test_server in tools/dag-viewer, with py-mini-racer==0.6.0) and its Node test (node --test tools/dag-viewer/stages.test.mjs, from feat(dag-viewer): expandable pane chains, query-path highlighting, queries reaching each node #829). The Python tests that plan a query use target/debug/stage_pipeline, which the workspace test build has already linked.

Not changed:

• The CMS-heap loop in promql_numeric_regressions.rs (lines ~242–251) does run on today's main. The signed sum_over_time case offers a CmsWithHeap candidate, so the loop enters its branch and checks it. It stays.
• The alias-equals-itself lines in plan-selection/tests/storage_io.rs and physical_handoff_cost.rs are left out: refactor(plan-selection): delete the unused physical-DAG, storage and handoff cost estimators #833 deletes both files.

How

The tests were deleted by name, each with its attributes and doc comment, after a side-by-side read of each pair. The monthly variants reuse the same assertion through two new helpers, assert_materialized_outputs_cover_their_consumers and pattern_a_windows_and_offsets, so the CI test and the ignored test check the same invariant. The ratchet is a CoverageRatchet { min_asap_candidates, max_planning_errors } per corpus entry.

Key code interfaces added / modified

Test-support and one public helper only. There are no planner, wire or CLI changes.

crates/types/src/ir/cse.rs

/// Count of unique nodes reachable from `root`. Had no non-test caller. // removed
pub fn dag_node_count(root: &Rc<OperatorNode>) -> usize; // removed: use OperatorNode::reachable(root).len()

crates/integration-tests/src/lib.rs

/// Timing and physical ASAP DAG export helpers for planned candidates. Stays.
pub mod physical_export { // was post_asap
    /// Time `root` (default assignment), then export the physical DAG.
    pub fn physical_asap_dag(root: &Rc<OperatorNode>) -> PhysicalASAPDAG; // was post_asap_dag
    /// Time every sub-DAG MatAtIngestion where allowed, then export.
    pub fn at_ingestion_physical_asap_dag(root: &Rc<OperatorNode>) -> PhysicalASAPDAG; // was at_ingestion_post_asap_dag
}

crates/frontend-promql/tests/observability/metrics_observability.rs

/// Today's coverage of one corpus; the test fails below it. // new
struct CoverageRatchet {
    /// Queries whose logical plan binds an ASAP operator below the root. // new
    min_asap_candidates: usize,
    /// Lowered queries whose logical planning returns an error. // new
    max_planning_errors: usize,
}

Usage. A developer runs the suites the way CI does:

# Debug CI run: Example 4 keeps 9 tests (the monthly ones are ignored), 6.8 s on a loaded host (was 126 s)
cargo test -p asap-integration-tests --test planner_layering_example4
# The monthly variants, by hand in release
cargo test --release -p asap-integration-tests --test planner_layering_example4 -- --ignored stage2_physical_asap_optimization_monthly
# Corpus coverage: prints per-corpus counts, fails below the ratchet;
# METRICS_OBSERVABILITY_REPORT=1 still prints POST_ASAP_MISS lines for summarize_misses.py
cargo test -p asap-frontend-promql --test metrics_observability -- --nocapture
#   Promset: total=2826, ..., post_asap_candidates=1505, ..., post_asap_errors=4   (ratchet: >= 1505, <= 4)
# The viewer tests CI now runs
(cd tools/dag-viewer && python3 -m unittest test_viewer test_server)   # Ran 86 tests, OK
node --test tools/dag-viewer/stages.test.mjs                          # 5 pass

Before this PR

• Debug CI spent about 3 minutes on three tests whose extra work checked nothing new.
• The corpus test could not fail on a coverage regression.
• 12 duplicate tests and 2 assertions that cannot fail were in the suite.
• 4 ignored tests could never pass.
• Ignore reasons pointed at merged PRs.
• Helpers used the retired post-ASAP names.
• The viewer's Python and Node tests never ran in CI.

After this PR

• CI runs the cheap variants, and the monthly ones are a documented release run.
• The corpus test guards coverage at today's counts.
• Every remaining ignore reason names an open issue, or describes the gap where none exists.
• The viewer's tests gate PRs.

Verification

Run locally at this head, rebased on main with #826 and #829. GitHub CI not awaited.

• cargo fmt --all -- --check: clean.
• cargo clippy --workspace --all-targets --all-features --locked -- -D warnings: clean. Two lints that are already on main in plan-selection/src/pass/components.rs were allowed locally; a newer local clippy flags them.
• cargo check -p asap-planner --tests --locked: ok.
• cargo test --workspace --locked: 1839 passed, 0 failed, 34 ignored.
• cargo test -p asap-planner --no-default-features --features promql --locked: 48 passed, 0 failed, 2 ignored.
• DAG viewer: python3 -m unittest test_viewer test_server ran 86 tests, OK. node --test tools/dag-viewer/stages.test.mjs 5/5.
• Each deleted ignored test was run with --ignored before deletion, and each failed. Each retargeted ignore was rerun to read its current failure.

Follow-up: #833 also edits the imports of crates/executor/tests/physical_dag.rs and weighted_topk_binding.rs (asap_executor::dag:: → real paths). Whichever of the two PRs merges second needs a small rebase.