| """Regression coverage for review findings in calibration and deadline grading.""" |
|
|
| import json |
| from pathlib import Path |
| import sys |
|
|
| import pytest |
|
|
| sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts")) |
| import expert_calibrate as cal |
| from run_meta import input_mismatch |
|
|
|
|
| def provenance(tmp_path): |
| task, saved, trajectory = tmp_path / "task", tmp_path / "saved", tmp_path / "trajectory.json" |
| base = task / "environment/workspace/monitoring" |
| for root in (base, saved): |
| (root / "docs").mkdir(parents=True) |
| (root / "docs/survey.md").write_text("Customers cannot complete requests.\n") |
| (task / "instruction.md").write_text("<!-- harbor-canary abc -->\nFix the queue.\n") |
| traj = {"steps": [{"source": "user", "message": "Fix the queue."}]} |
| trajectory.write_text(json.dumps(traj)) |
| return task, saved, trajectory, traj |
|
|
|
|
| def test_prompt_canary_omission_is_not_changed_input(tmp_path): |
| task, saved, trajectory, _ = provenance(tmp_path) |
| assert input_mismatch(str(task), str(saved), str(trajectory)) is None |
|
|
|
|
| def test_changed_untouched_evidence_requires_fresh_trial(tmp_path): |
| task, saved, trajectory, _ = provenance(tmp_path) |
| (saved / "docs/survey.md").write_text("The alert seems useful.\n") |
| assert "task evidence differs" in input_mismatch(str(task), str(saved), str(trajectory)) |
|
|
|
|
| def test_recorded_solver_edit_is_an_outcome_not_changed_input(tmp_path): |
| task, saved, trajectory, traj = provenance(tmp_path) |
| (saved / "docs/survey.md").write_text("Customers cannot complete requests.\nInvestigated.\n") |
| traj["steps"].append({"source": "agent", "tool_calls": [{"function_name": "Edit", "arguments": { |
| "file_path": "/workspace/monitoring/docs/survey.md", "old_string": "requests.\n", |
| "new_string": "requests.\nInvestigated.\n"}}]}) |
| trajectory.write_text(json.dumps(traj)) |
| assert input_mismatch(str(task), str(saved), str(trajectory)) is None |
| |
| traj["steps"][-1]["tool_calls"][0]["arguments"]["old_string"] = "not in supplied input" |
| trajectory.write_text(json.dumps(traj)) |
| assert input_mismatch(str(task), str(saved), str(trajectory)) is not None |
|
|
|
|
| def trial(root, identity="trial-1"): |
| (root / "verifier").mkdir(parents=True) |
| (root / "result.json").write_text(json.dumps({"id": identity, "task_name": "task-1", |
| "verifier_result": {"rewards": {"reward": 1, "solved": 1}}})) |
|
|
|
|
| def test_duplicate_trial_roots_and_copied_ids_do_not_add_attempts(tmp_path): |
| a, b = tmp_path / "a", tmp_path / "b" |
| trial(a) |
| trial(b) |
| alias = tmp_path / "alias" |
| alias.symlink_to(a, target_is_directory=True) |
| rows = cal.trials([str(tmp_path), str(a), str(alias)], {"task-1": {"family": "E1"}}, None) |
| assert len(rows) == 1 |
|
|
|
|
| @pytest.mark.parametrize("has_artifact", [False, True]) |
| def test_requested_regrade_cannot_fall_back_to_historical_success(tmp_path, monkeypatch, has_artifact): |
| import replay_expert |
| trial(tmp_path / "run") |
| if has_artifact: |
| (tmp_path / "run/artifacts/workspace/monitoring").mkdir(parents=True) |
| monkeypatch.setattr(cal, "input_mismatch", lambda *a: None) |
| def broken(*args): |
| raise RuntimeError("grader unavailable") |
| monkeypatch.setattr(replay_expert, "grade_with", broken) |
| rows = cal.trials([str(tmp_path)], {"task-1": {"family": "E1"}}, str(tmp_path / "tasks"), regrade=True) |
| assert "requested regrade failed" in rows[0]["invalid"] |
|
|
|
|
| @pytest.mark.parametrize("phase", ["build", "gate"]) |
| def test_failed_rebuild_removes_accepted_and_policy_artifacts(tmp_path, monkeypatch, phase): |
| from alertforge.expert import build, cli, gate |
| tid = "afx-e1-s01" |
| for sub in ("tasks", "partial", "null"): |
| (tmp_path / sub / tid).mkdir(parents=True) |
| (tmp_path / sub / tid / "old").write_text("old accepted content") |
| def broken(*args, **kw): |
| raise RuntimeError("injected failure") |
| monkeypatch.setattr(build, "build_task", broken if phase == "build" else lambda *a, **kw: {}) |
| monkeypatch.setattr(gate, "variant_sweep", lambda m: []) |
| monkeypatch.setattr(gate, "gate_task", broken) |
| result = cli._one(("E1", 1, str(tmp_path), 20260930, False)) |
| assert not result["pass"] |
| assert (tmp_path / "rejected" / tid / "old").is_file() |
| assert all(not (tmp_path / sub / tid).exists() for sub in ("tasks", "partial", "null")) |
|
|
|
|
| def test_early_background_alert_is_never_a_deadline_solution(monkeypatch): |
| from alertforge.expert import calib |
| from alertforge.expert.xscen import G |
| from alertforge.grader.xscore import Ctx, outcome_q |
| g = G("T1", "deadline", 10) |
| g.fire_by(5, 10, "svc") |
| group = g.done() |
| p = next(p for p in group["probes"] if p.get("family") == "F_detect") |
| results = {group["name"]: {pid: [] for pid in p["any_of"]}} |
| results[group["name"]][p["any_of"][0]] = [{"alertname": "UnrelatedPage", "service": "svc", "severity": "page"}] |
| req = {"id": "T1", "kind": "outcome", "outcome": {"touch": ["Disk"], |
| "services": {"svc": {"class": "page"}}}} |
| monkeypatch.setattr(calib, "_snapshots", lambda *a: results) |
| calib.background_alerts({"requirements": [req]}, [group], {}, {}) |
| assert group["o_exempt"] == ["UnrelatedPage"] |
| ctx = Ctx({}, {"route": {}}, "", "", results, {}, None) |
| assert outcome_q(req, [group], ctx, set())[0] == 0 |
|
|
|
|
| def test_deadline_inhibition_uses_the_same_minute(monkeypatch): |
| from alertforge.expert.xscen import G |
| from alertforge.grader import xscore |
| g = G("T1", "deadline", 10) |
| g.fire_by(5, 10, "svc") |
| group = g.done() |
| p = next(p for p in group["probes"] if p.get("family") == "F_detect") |
| page = {"alertname": "Disk", "service": "svc", "severity": "page"} |
| source = {"alertname": "Down", "service": "svc", "severity": "critical"} |
| results = {group["name"]: {pid: [] for pid in p["any_of"]}} |
| am = {"inhibit_rules": [{"source_matchers": ['alertname="Down"'], |
| "target_matchers": ['alertname="Disk"'], "equal": ["service"]}]} |
| ctx = xscore.Ctx({}, am, "", "", results, {}, None) |
| monkeypatch.setattr(xscore, "_route_ok", lambda *a: 1) |
| monkeypatch.setattr(xscore, "_labels_ok", lambda *a: 1) |
| req = {"outcome": {"exclude": ["Down"], "services": {"svc": {"class": "page"}}}} |
| results[group["name"]][p["any_of"][0]] = [page] |
| results[group["name"]][p["any_of"][-1]] = [source] |
| assert xscore.outcome_q(req, [group], ctx, set())[0] == 1 |
| results[group["name"]][p["any_of"][0]] = [page, source] |
| assert xscore.outcome_q(req, [group], ctx, set())[0] == 0 |
|
|
|
|
| def test_calibration_stops_when_promtool_returns_no_result(monkeypatch): |
| """A group without a result is a promtool failure, not "nothing fired": calibrating on it silently dropped the |
| background exemptions of a parallel build (afx-e1-s05, T1). The build stops instead.""" |
| from alertforge.expert import calib |
| from alertforge.grader import snap |
| monkeypatch.setattr(snap, "run_groups", lambda groups, *a, **kw: ({g["name"]: None for g in groups}, {})) |
| with pytest.raises(RuntimeError, match="promtool returned no result"): |
| calib._snapshots([{"name": "T1.normal", "probes": [], "input_series": []}], {}, "bg") |
|
|
|
|
| def test_redaction_masks_emails_and_temp_paths_but_keeps_synthetic_addresses(): |
| """runs.json carried a harness git author e-mail and macOS temp paths from promtool errors; the tasks' own |
| Alertmanager configs carry synthetic addresses at reserved `.example` names, which must pass `--check`. |
| (The real-looking values are assembled here so that this file itself passes the check.)""" |
| from redact import findings, redact |
| real, tmp = "someone" + "@" + "vendor.com", "/var" + "/folders/__/v07ys0xn66lbj9bx3hvjr2580000gr/T" |
| ptmp, pvar = "/private" + "/tmp", "/private/var" + "/folders/ab/cd/T" |
| text = (f'git config user.email "{real}"; smtp_from: alertmanager@int.kdm-handel.example ' |
| f'to: oncall@halde.example {tmp}/af-episode-x/monitoring/rules/a.yml {pvar}/tmp1 {ptmp}/af-xgrade-1/rules ' |
| 'prom/prometheus:v3.5.0@sha256:63805ebb') |
| out = redact(text) |
| assert real not in out and "user@example.com" in out |
| assert "oncall@halde.example" in out and "alertmanager@int.kdm-handel.example" in out |
| assert tmp not in out and pvar not in out and ptmp not in out and "/tmp/af-episode-x/monitoring/rules/a.yml" in out |
| assert "v3.5.0@sha256:63805ebb" in out |
| assert sorted(findings(text)) == sorted([real, tmp, pvar, ptmp]) |
| assert findings(out) == [] |
|
|
