webapp/tipitaka-api/tests/test_rag_service.py

"""
Unit tests for RAGService — embedding cache, Qdrant client.

Uses mocked dependencies to test in isolation.
"""

import pytest
from unittest.mock import MagicMock, patch


class TestEmbeddingCache:
    @pytest.fixture
    def rag_service(self):
        from app.services.rag_service import RAGService
        with patch.object(RAGService, '_load_model'), \
             patch.object(RAGService, '_load_reranker'), \
             patch.object(RAGService, '_init_turbovec'):
            service = RAGService()
            service.tv_index = MagicMock()
            return service

    def test_cache_initialized(self, rag_service):
        assert rag_service._embed_cache is not None
        assert rag_service._embed_cache_max == 256

    def test_cache_stores_and_returns(self, rag_service):
        """Cache should store and return the same value for identical input."""
        mock_response = MagicMock()
        mock_response.json.return_value = {"embeddings": [[0.1, 0.2, 0.3]]}
        mock_response.raise_for_status.return_value = None

        with patch("app.services.rag_service.httpx.post", return_value=mock_response):
            emb1 = rag_service._get_embedding("test query")

        # Second call should use cache (no httpx call)
        emb2 = rag_service._get_embedding("test query")
        assert emb1 == emb2

    def test_cache_max_respected(self, rag_service):
        """When cache exceeds max, oldest entries should be evicted."""
        old_entry = "a" * 10
        for i in range(rag_service._embed_cache_max + 5):
            rag_service._embed_cache[f"key_{i}"] = [0.1] * 1024
            if len(rag_service._embed_cache) > rag_service._embed_cache_max:
                rag_service._embed_cache.popitem(last=False)

        assert len(rag_service._embed_cache) <= rag_service._embed_cache_max


class TestRAGServiceInit:
    def test_init_graceful_on_turbovec_failure(self):
        """RAGService constructor should not crash when turbovec is unavailable."""
        from app.services.rag_service import RAGService
        with patch.object(RAGService, '_load_model'), \
             patch.object(RAGService, '_load_reranker'), \
             patch.object(RAGService, '_init_turbovec') as mock_init:
            service = RAGService()
            mock_init.assert_called_once()
            assert service.tv_index is None  # _init_turbovec didn't set it