test(pipeline): run llm_keyword_suggest against real LLM when keys set

Remove mock; load market_assistant/.env before checking credentials.
Live test skips without OPENAI_* or LLM_* key+base URL.

Made-with: Cursor

backend/pipeline/tests/test_llm_keyword_suggest.py

@@ -0,0 +1,110 @@
+"""llm_keyword_suggest：分块/解析烟测；有 API 配置时直连大模型做联调。"""
+from __future__ import annotations
+
+import os
+import unittest
+from pathlib import Path
+
+try:
+    from dotenv import load_dotenv
+
+    _ma_env = Path(__file__).resolve().parents[3] / ".env"
+    if _ma_env.is_file():
+        load_dotenv(_ma_env)
+except ImportError:
+    pass
+
+from django.test import SimpleTestCase
+
+from pipeline.llm_keyword_suggest import (
+    MAX_CHUNK_CHARS,
+    MAX_CHUNKS,
+    _chunk_comment_texts,
+    _parse_phrases_object,
+    suggest_focus_keywords_from_all_comments,
+)
+
+
+def _llm_configured() -> bool:
+    key = (os.environ.get("OPENAI_API_KEY") or os.environ.get("LLM_API_KEY") or "").strip()
+    base = (os.environ.get("OPENAI_BASE_URL") or os.environ.get("LLM_BASE_URL") or "").strip()
+    return bool(key and base)
+
+
+class ChunkCommentTextsTests(SimpleTestCase):
+    def test_empty(self) -> None:
+        self.assertEqual(_chunk_comment_texts([]), [])
+
+    def test_respects_max_chunk_chars(self) -> None:
+        a = "x" * (MAX_CHUNK_CHARS // 2)
+        b = "y" * (MAX_CHUNK_CHARS // 2)
+        c = "z" * (MAX_CHUNK_CHARS // 2)
+        parts = _chunk_comment_texts([a, b, c])
+        self.assertGreaterEqual(len(parts), 2)
+        for p in parts:
+            self.assertLessEqual(len(p) + p.count("\n"), MAX_CHUNK_CHARS + 50)
+
+    def test_max_chunks_trims(self) -> None:
+        texts = [f"段落{i} " + "字" * 800 for i in range(80)]
+        parts = _chunk_comment_texts(texts)
+        self.assertLessEqual(len(parts), MAX_CHUNKS)
+
+
+class ParsePhrasesObjectTests(SimpleTestCase):
+    def test_plain_json(self) -> None:
+        raw = '{"phrases": ["低糖", "口感好"]}'
+        self.assertEqual(_parse_phrases_object(raw), ["低糖", "口感好"])
+
+    def test_fenced_json(self) -> None:
+        raw = '```json\n{"phrases": ["A", "B"]}\n```'
+        self.assertEqual(_parse_phrases_object(raw), ["A", "B"])
+
+    def test_embedded_object(self) -> None:
+        raw = '前缀 {"phrases": ["x"]} 后缀'
+        self.assertEqual(_parse_phrases_object(raw), ["x"])
+
+    def test_invalid_returns_empty(self) -> None:
+        self.assertEqual(_parse_phrases_object("not json"), [])
+
+
+class SuggestFocusKeywordsTests(SimpleTestCase):
+    def test_no_comments_returns_empty(self) -> None:
+        out = suggest_focus_keywords_from_all_comments(
+            keyword="低GI",
+            brief_slice={"comment_focus_keywords": []},
+            all_comment_texts=[],
+        )
+        self.assertEqual(out["suggested_focus_keywords"], [])
+        self.assertEqual(out["chunks_processed"], 0)
+        self.assertIn("无评价", out["rationale"])
+
+
+@unittest.skipUnless(
+    _llm_configured(),
+    "需要环境变量 OPENAI_API_KEY+OPENAI_BASE_URL（或 LLM_API_KEY+LLM_BASE_URL），"
+    "与 AI_crawler 相同；可在 market_assistant/.env 配置后重跑。",
+)
+class SuggestFocusKeywordsLiveLLMTests(SimpleTestCase):
+    """直连网关调用 ``chat_completion_text``，会消耗少量 token。"""
+
+    def test_live_extracts_phrases_from_comments(self) -> None:
+        comments = [
+            "低GI饼干口感偏硬，甜度刚好，饱腹感不错。",
+            "物流有点慢，包装压扁了一角，但味道还行。",
+            "希望出小包装，一次吃不完容易受潮。",
+        ]
+        out = suggest_focus_keywords_from_all_comments(
+            keyword="低GI饼干",
+            brief_slice={"comment_focus_keywords": [{"word": "甜度"}]},
+            all_comment_texts=comments,
+        )
+        self.assertGreaterEqual(out["chunks_processed"], 1)
+        self.assertEqual(out["total_comment_texts"], 3)
+        kws = out["suggested_focus_keywords"]
+        self.assertIsInstance(kws, list)
+        self.assertGreater(len(kws), 0, "模型应返回至少 1 条短语")
+        for p in kws:
+            self.assertIsInstance(p, str)
+            self.assertGreaterEqual(len(p), 2)
+            self.assertLessEqual(len(p), 24)
+        self.assertNotIn("甜度", kws)