sessions: 添加多后端回放一致性测试框架

.github/workflows/ci.yml

@@ -77,6 +77,12 @@ jobs:
         # run: pytest --cov=trpc_agent_sdk --cov-report=xml --cov-report=term tests/
         run: pytest --cov=trpc_agent_sdk --cov-report=xml --cov-report=term --cov-fail-under=80 tests/
 
+      - name: Run InMemory-only replay smoke test (not full matrix)
+        env:
+          TRPC_REPLAY_BACKENDS: in_memory
+        # The full cross-backend replay matrix runs in the coverage step above.
+        run: pytest -q tests/sessions/test_replay_consistency.py tests/sessions/test_replay_real_agent.py
+
       - name: Upload coverage reports to Codecov
         uses: codecov/codecov-action@v4
         env:

requirements-test.txt

@@ -1,7 +1,7 @@
 # Test framework
 pytest
 pytest-mock
-pytest-asyncio
+pytest-asyncio>=0.24.0
 pyparsing>=2.4.2
 pytest-cov
 unittest-xml-reporting

tests/sessions/DESIGN.md

@@ -0,0 +1,3 @@
+# Session / Memory / Summary Replay Harness 设计说明
+
+同一组轨迹驱动 InMemory、SQLite 和可选 Redis，快照覆盖事件、state、memory、summary。归一化只处理自动 ID、相对时间、Unicode 空白和无序容器；Summary 文本可规范化，但 session 归属、版本和覆盖关系必须一致。差异按后端对和字段路径精确匹配。SQLite 重开验证持久化；`TRPC_REPLAY_BACKENDS=in_memory` 为轻量模式。两份 JSON 均为测试生成产物，可删除后重跑再生，不是手写契约。

tests/sessions/conftest.py

@@ -0,0 +1,34 @@
+#
+# Tencent is pleased to support the open source community by making trpc-agent-python available.
+#
+# Copyright (C) 2026 Tencent. All rights reserved.
+#
+# trpc-agent-python is licensed under the Apache License Version 2.0.
+#
+"""Collection policy for the replay acceptance tests."""
+
+import os
+
+import pytest
+
+BACKEND_MODE_ENVIRONMENT_VARIABLE = "TRPC_REPLAY_BACKENDS"
+IN_MEMORY_BACKEND_MODE = "in_memory"
+IN_MEMORY_TEST_NAME = "test_in_memory_only_lightweight_mode"
+IN_MEMORY_REAL_AGENT_SAFE_TEST_NAME = "test_subtle_realistic_drift_is_detected"
+REPLAY_TEST_FILES = frozenset({
+    "test_replay_consistency.py",
+    "test_replay_real_agent.py",
+})
+IN_MEMORY_SKIP_REASON = "disabled by TRPC_REPLAY_BACKENDS=in_memory"
+
+
+def pytest_collection_modifyitems(items):
+    """Keep only the lightweight replay when InMemory-only mode is selected."""
+    if os.getenv(BACKEND_MODE_ENVIRONMENT_VARIABLE) != IN_MEMORY_BACKEND_MODE:
+        return
+    skip = pytest.mark.skip(reason=IN_MEMORY_SKIP_REASON)
+    for item in items:
+        if (item.path.name == "test_replay_consistency.py" and item.name != IN_MEMORY_TEST_NAME) or (
+                item.path.name == "test_replay_real_agent.py"
+                and item.name != IN_MEMORY_REAL_AGENT_SAFE_TEST_NAME):
+            item.add_marker(skip)

tests/sessions/real_agent_replay_validation_report.json

@@ -0,0 +1,56 @@
+{
+  "case_id": "real_agent_tool_memory",
+  "clean_comparison": {
+    "diffs": [],
+    "unexpected_diff_count": 0
+  },
+  "injected_anomalies": {
+    "memory_single_character_drift": {
+      "detected": true,
+      "diffs": [
+        {
+          "allowed": false,
+          "backend_pair": [
+            "in_memory",
+            "sqlite"
+          ],
+          "case_id": "real_agent_tool_memory",
+          "category": "memory",
+          "event_index": null,
+          "field_path": "/memory/Chinese/memories/0/content/parts/0/text",
+          "left": "Here are the preferences for **replay-user**:\n\n- **Preferred Language:** Chinese\n- **Notification Channel:** email",
+          "reason": null,
+          "right": "Here are the preferences for **replay-user**:\n\n- **Preferred Language:** Chinese\n- **Notification Channel:** emaiX",
+          "session_id": "replay-session",
+          "summary_id": null
+        }
+      ],
+      "unexpected_diff_count": 1
+    },
+    "tool_response_single_character_drift": {
+      "detected": true,
+      "diffs": [
+        {
+          "allowed": false,
+          "backend_pair": [
+            "in_memory",
+            "sqlite"
+          ],
+          "case_id": "real_agent_tool_memory",
+          "category": "events",
+          "event_index": 2,
+          "field_path": "/events/2/content/parts/0/function_response/response/preferred_language",
+          "left": "Chinese",
+          "reason": null,
+          "right": "ChinesX",
+          "session_id": "replay-session",
+          "summary_id": null
+        }
+      ],
+      "unexpected_diff_count": 1
+    }
+  },
+  "model": "deepseek-v4-flash",
+  "schema_version": 1,
+  "tool_round_trip": true
+}