155 lines
6.1 KiB
Python
155 lines
6.1 KiB
Python
"""
|
||
ai_review 黑盒用户模拟测试 — 内容审核 + 人工反馈。
|
||
|
||
用户旅程:
|
||
1. 提交明显违规内容 → 应返回 invalid: true
|
||
2. 提交正常无害内容 → 应返回 invalid: false
|
||
3. 提交空内容 → 应返回 400 错误
|
||
4. 缺少 content 字段 → 应返回 400 错误
|
||
5. 人工反馈纠正误判 → RAG 知识库更新
|
||
6. 查看 RAG 列表
|
||
7. 健康检查
|
||
|
||
注意: ai_review 依赖 DeepSeek API + ChromaDB，API 调用可能因网络/配额失败。
|
||
测试使用了通用违规关键词，实际审核结果取决于 DeepSeek 模型判断。
|
||
"""
|
||
|
||
import json
|
||
import random
|
||
import string
|
||
|
||
from helpers import (
|
||
http_get, http_post, try_parse_json,
|
||
assert_eq, assert_status, assert_true, assert_json_key,
|
||
skip, Counter, check_service,
|
||
)
|
||
from config import get_url
|
||
|
||
AI_REVIEW_URL = get_url("ai_review")
|
||
|
||
|
||
def run(url: str = None):
|
||
"""ai_review 用户模拟测试入口。返回 (passed, failed, skipped)。"""
|
||
au = url or AI_REVIEW_URL
|
||
c = Counter()
|
||
|
||
print(f"\n--- ai_review 内容审核流程 ---\n")
|
||
|
||
# ------------------------------------------------------------------
|
||
# 0. 检查服务可用性
|
||
# ------------------------------------------------------------------
|
||
if not check_service(au)[0]:
|
||
skip("全部测试", "ai_review 服务不可达", c)
|
||
return c.passed, c.failed, c.skipped
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 1: 提交明显违规内容 — 应标记为违规
|
||
# ------------------------------------------------------------------
|
||
violation_texts = [
|
||
"出售毒品冰毒海洛因，价格优惠，联系QQ123456",
|
||
"提供赌博网站搭建服务，包月维护，日入过万",
|
||
]
|
||
violation_text = random.choice(violation_texts)
|
||
|
||
status, body, _ = http_post(f"{au}/review", {"content": violation_text})
|
||
assert_eq(status, 200, "审核-违规内容-响应 (200)", c)
|
||
data = try_parse_json(body)
|
||
if data:
|
||
# 注意: invalid 可能是字符串 "true"/"false" 而非布尔值
|
||
invalid = data.get("invalid")
|
||
is_invalid = invalid in (True, "true", "True")
|
||
assert_true(
|
||
is_invalid,
|
||
"审核-违规内容-标记为invalid", c,
|
||
f"invalid={invalid}, body={body[:200]}",
|
||
)
|
||
else:
|
||
assert_true(False, "审核-违规内容-解析失败", c, f"body={body[:200]}")
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 2: 提交正常无害内容 — 应通过
|
||
# ------------------------------------------------------------------
|
||
safe_texts = [
|
||
"今天天气真好，适合出去散步。",
|
||
"Python 是一门优秀的编程语言，广泛应用于数据科学和人工智能领域。",
|
||
"微服务架构是一种将应用拆分为多个小型独立服务的软件设计方法。",
|
||
]
|
||
safe_text = random.choice(safe_texts)
|
||
|
||
status, body, _ = http_post(f"{au}/review", {"content": safe_text})
|
||
assert_eq(status, 200, "审核-正常内容-响应 (200)", c)
|
||
data = try_parse_json(body)
|
||
if data:
|
||
invalid = data.get("invalid")
|
||
is_invalid = invalid in (True, "true", "True")
|
||
# 正常内容应该不违规
|
||
assert_true(
|
||
not is_invalid,
|
||
"审核-正常内容-标记为safe", c,
|
||
f"invalid={invalid}, body={body[:200]}",
|
||
)
|
||
else:
|
||
assert_true(False, "审核-正常内容-解析失败", c, f"body={body[:200]}")
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 3: 提交空内容 — 应返回 400
|
||
# ------------------------------------------------------------------
|
||
status, body, _ = http_post(f"{au}/review", {"content": ""})
|
||
assert_eq(status, 400, "审核-空内容-拒绝 (400)", c)
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 4: 缺少 content 字段 — 应返回 400
|
||
# ------------------------------------------------------------------
|
||
status, body, _ = http_post(f"{au}/review", {})
|
||
assert_eq(status, 400, "审核-缺少content-拒绝 (400)", c)
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 5: RAG 人工反馈 — 纠正误判（标注一条违规内容）
|
||
# ------------------------------------------------------------------
|
||
feedback_text = f"test_feedback_{random.randint(10000, 99999)}: 非法内容"
|
||
status, body, _ = http_post(f"{au}/rag/feedback", {
|
||
"content": feedback_text,
|
||
"correct_type": "violation",
|
||
})
|
||
assert_eq(status, 200, "RAG反馈-添加violation (200)", c)
|
||
data = try_parse_json(body)
|
||
if data:
|
||
assert_true(
|
||
data.get("added") in (True, False),
|
||
"RAG反馈-返回added字段", c,
|
||
)
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 6: RAG 反馈 — 参数错误
|
||
# ------------------------------------------------------------------
|
||
status, body, _ = http_post(f"{au}/rag/feedback", {
|
||
"content": "test",
|
||
"correct_type": "invalid_type",
|
||
})
|
||
assert_eq(status, 400, "RAG反馈-错误类型-拒绝 (400)", c)
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 7: RAG 列表查询
|
||
# ------------------------------------------------------------------
|
||
status, body, _ = http_get(f"{au}/rag/list")
|
||
assert_eq(status, 200, "RAG列表-查询 (200)", c)
|
||
data = try_parse_json(body)
|
||
if data:
|
||
assert_json_key(data, "total", "RAG列表-有total字段", c)
|
||
|
||
# ------------------------------------------------------------------
|
||
# 场景 8: 健康检查
|
||
# ------------------------------------------------------------------
|
||
status, body, _ = http_get(f"{au}/health")
|
||
assert_eq(status, 200, "健康检查 (200)", c)
|
||
|
||
# ------------------------------------------------------------------
|
||
# 汇总
|
||
# ------------------------------------------------------------------
|
||
print(f"\n ai_review 测试完成: {c.summary()}")
|
||
return c.passed, c.failed, c.skipped
|
||
|
||
|
||
if __name__ == "__main__":
|
||
run()
|