"""HarnessLog + RunLogImpl 单元测试。 HarnessLog: SQLite 薄包装的线程安全、幂等、context manager 语义验证。 RunLogImpl: 只读 RunLog Protocol 实现，独立连接不污染 _runs 运行状态。 """ from __future__ import annotations import sqlite3 import threading from typing import TYPE_CHECKING import pytest from app.harness.log import HarnessLog, RunLogImpl if TYPE_CHECKING: from pathlib import Path # --------------------------------------------------------------------------- # Fixtures # --------------------------------------------------------------------------- @pytest.fixture() def db_path(tmp_path: Path) -> str: """返回临时 SQLite 数据库路径。""" return str(tmp_path / "test.db") @pytest.fixture() def run_id() -> str: return "test-run-001" # =========================================================================== # HarnessLog 测试 # =========================================================================== class TestHarnessLog: """HarnessLog 核心功能测试。""" def test_create_table_and_insert(self, db_path: str, run_id: str) -> None: """create_table 建表 + insert 写入 + query 读回。""" with HarnessLog(db_path, run_id) as log: log.create_table("metrics", {"epoch": "INTEGER", "loss": "REAL"}) log.insert("metrics", {"epoch": 1, "loss": 0.5}) rows = log.query("SELECT * FROM metrics WHERE run_id = ?", (run_id,)) assert len(rows) == 1 assert rows[0]["epoch"] == 1 assert rows[0]["loss"] == 0.5 assert rows[0]["run_id"] == run_id # insert 自动填充 timestamp assert rows[0]["timestamp"] is not None def test_query_thread_safety(self, db_path: str, run_id: str) -> None: """并发读写不损坏游标状态。""" errors: list[Exception] = [] with HarnessLog(db_path, run_id) as log: log.create_table("counter", {"value": "INTEGER"}) def writer() -> None: try: for i in range(50): log.insert("counter", {"value": i}) except Exception as exc: errors.append(exc) def reader() -> None: try: for _ in range(50): log.query("SELECT COUNT(*) as cnt FROM counter") except Exception as exc: errors.append(exc) threads = [ threading.Thread(target=writer), threading.Thread(target=reader), threading.Thread(target=reader), ] for t in threads: t.start() for t in threads: t.join() assert errors == [], f"并发读写产生错误: {errors}" def test_context_manager_completed(self, db_path: str, run_id: str) -> None: """正常退出时 status = completed。""" with HarnessLog(db_path, run_id): pass conn = sqlite3.connect(db_path) conn.row_factory = sqlite3.Row row = conn.execute( "SELECT status, finished_at FROM _runs WHERE run_id = ?", (run_id,) ).fetchone() conn.close() assert row["status"] == "completed" assert row["finished_at"] is not None def test_context_manager_failed(self, db_path: str, run_id: str) -> None: """异常退出时 status = failed。""" with pytest.raises(ValueError, match="boom"), HarnessLog(db_path, run_id): raise ValueError("boom") conn = sqlite3.connect(db_path) conn.row_factory = sqlite3.Row row = conn.execute( "SELECT status FROM _runs WHERE run_id = ?", (run_id,) ).fetchone() conn.close() assert row["status"] == "failed" def test_insert_or_ignore_idempotent(self, db_path: str, run_id: str) -> None: """同一 run_id 多次创建 HarnessLog 不报错（INSERT OR IGNORE 幂等）。""" with HarnessLog(db_path, run_id): pass # 再次用同一 run_id 打开——不应抛异常 with HarnessLog(db_path, run_id) as log: log.create_table("t", {"x": "INTEGER"}) log.insert("t", {"x": 42}) conn = sqlite3.connect(db_path) count = conn.execute( "SELECT COUNT(*) FROM _runs WHERE run_id = ?", (run_id,) ).fetchone()[0] conn.close() assert count == 1, "INSERT OR IGNORE 应保证 _runs 只有一行" def test_wal_mode(self, db_path: str, run_id: str) -> None: """连接初始化后 journal_mode 应为 WAL。""" with HarnessLog(db_path, run_id) as log: rows = log.query("PRAGMA journal_mode") assert rows[0]["journal_mode"].lower() == "wal" def test_insert_many(self, db_path: str, run_id: str) -> None: """insert_many 批量插入多条记录。""" records = [{"epoch": i, "loss": float(i) * 0.1} for i in range(5)] with HarnessLog(db_path, run_id) as log: log.create_table("batch", {"epoch": "INTEGER", "loss": "REAL"}) log.insert_many("batch", records) rows = log.query( "SELECT * FROM batch WHERE run_id = ? ORDER BY epoch", (run_id,) ) assert len(rows) == 5 assert [r["epoch"] for r in rows] == [0, 1, 2, 3, 4] def test_log_event(self, db_path: str, run_id: str) -> None: """log_event 向 _events 表写入事件。""" with HarnessLog(db_path, run_id) as log: log.log_event("train_start", {"epoch": 1, "lr": 0.001}) log.log_event("train_end", {"epoch": 1, "loss": 0.42}) rows = log.query( "SELECT * FROM _events WHERE run_id = ? ORDER BY id", (run_id,) ) assert len(rows) == 2 assert rows[0]["event_type"] == "train_start" assert rows[1]["event_type"] == "train_end" def test_execute_raw_sql(self, db_path: str, run_id: str) -> None: """execute 执行原生 SQL 写操作。""" with HarnessLog(db_path, run_id) as log: log.create_table("raw", {"val": "INTEGER"}) log.execute( "INSERT INTO raw (run_id, timestamp, val) VALUES (?, ?, ?)", (run_id, "2026-01-01T00:00:00", 99), ) rows = log.query("SELECT val FROM raw WHERE run_id = ?", (run_id,)) assert len(rows) == 1 assert rows[0]["val"] == 99 def test_upsert_mode(self, db_path: str, run_id: str) -> None: """insert mode='upsert' 使用 INSERT OR REPLACE。""" with HarnessLog(db_path, run_id) as log: log.create_table("kv", {"key": "TEXT", "val": "TEXT"}, primary_key="key") log.insert("kv", {"key": "a", "val": "1"}, mode="upsert") log.insert("kv", {"key": "a", "val": "2"}, mode="upsert") rows = log.query("SELECT val FROM kv WHERE key = 'a'") assert len(rows) == 1 assert rows[0]["val"] == "2" # =========================================================================== # RunLogImpl 测试 # =========================================================================== def _setup_predictions_and_traces(db_path: str, run_id: str) -> None: """向测试数据库写入 predictions 和 traces 数据，模拟 inference 阶段产物。""" with HarnessLog(db_path, run_id) as log: log.create_table( "predictions", { "question_id": "TEXT", "predicted_answer": "TEXT", "correct": "INTEGER", }, ) log.create_table( "traces", { "question_id": "TEXT", "step_idx": "INTEGER", "action": "TEXT", "observation": "TEXT", }, ) log.insert_many( "predictions", [ {"question_id": "q1", "predicted_answer": "A", "correct": 1}, {"question_id": "q2", "predicted_answer": "B", "correct": 0}, {"question_id": "q3", "predicted_answer": "C", "correct": 1}, ], ) log.insert_many( "traces", [ { "question_id": "q1", "step_idx": 0, "action": "search", "observation": "found", }, { "question_id": "q1", "step_idx": 1, "action": "verify", "observation": "ok", }, { "question_id": "q2", "step_idx": 0, "action": "search", "observation": "not found", }, ], ) class TestRunLogImpl: """RunLogImpl 只读查询端口测试。""" @pytest.mark.asyncio async def test_get_predictions(self, db_path: str, run_id: str) -> None: """get_predictions 返回指定 run 的全部预测记录。""" _setup_predictions_and_traces(db_path, run_id) impl = RunLogImpl(db_path) preds = await impl.get_predictions(run_id) assert len(preds) == 3 q_ids = {p["question_id"] for p in preds} assert q_ids == {"q1", "q2", "q3"} @pytest.mark.asyncio async def test_get_predictions_filtered(self, db_path: str, run_id: str) -> None: """get_predictions 按 question_ids 过滤。""" _setup_predictions_and_traces(db_path, run_id) impl = RunLogImpl(db_path) preds = await impl.get_predictions(run_id, question_ids=["q1", "q3"]) assert len(preds) == 2 q_ids = {p["question_id"] for p in preds} assert q_ids == {"q1", "q3"} @pytest.mark.asyncio async def test_get_traces(self, db_path: str, run_id: str) -> None: """get_traces 返回指定 run 的全部轨迹记录。""" _setup_predictions_and_traces(db_path, run_id) impl = RunLogImpl(db_path) traces = await impl.get_traces(run_id) assert len(traces) == 3 # q1 有 2 步，q2 有 1 步 q1_traces = [t for t in traces if t["question_id"] == "q1"] assert len(q1_traces) == 2 @pytest.mark.asyncio async def test_get_traces_filtered(self, db_path: str, run_id: str) -> None: """get_traces 按 question_ids 过滤。""" _setup_predictions_and_traces(db_path, run_id) impl = RunLogImpl(db_path) traces = await impl.get_traces(run_id, question_ids=["q2"]) assert len(traces) == 1 assert traces[0]["question_id"] == "q2" @pytest.mark.asyncio async def test_readonly_no_runs_insert(self, db_path: str, run_id: str) -> None: """RunLogImpl 查询不触发 _runs INSERT（不污染运行状态）。""" _setup_predictions_and_traces(db_path, run_id) # 用不同 run_id 查询，确保不会创建新的 _runs 行 other_run = "nonexistent-run" impl = RunLogImpl(db_path) preds = await impl.get_predictions(other_run) assert preds == [] # 验证 _runs 表没有 other_run 的记录 conn = sqlite3.connect(db_path) count = conn.execute( "SELECT COUNT(*) FROM _runs WHERE run_id = ?", (other_run,) ).fetchone()[0] conn.close() assert count == 0, "RunLogImpl 不应向 _runs 插入记录" @pytest.mark.asyncio async def test_protocol_compliance(self, db_path: str, run_id: str) -> None: """RunLogImpl 满足 RunLog Protocol（runtime_checkable isinstance 检查）。""" from core.evolution.protocols import RunLog impl = RunLogImpl(db_path) assert isinstance(impl, RunLog) @pytest.mark.asyncio async def test_missing_table_returns_empty(self, db_path: str, run_id: str) -> None: """查询不存在的表（predictions/traces 未建）返回空列表。""" # 仅创建 _runs，不建 predictions/traces 表 with HarnessLog(db_path, run_id): pass impl = RunLogImpl(db_path) preds = await impl.get_predictions(run_id) traces = await impl.get_traces(run_id) assert preds == [] assert traces == []