refactor(firestore-vector-search): use Firebase Genkit wherever possi…

…ble. (#603)

* refactor(firestore-vector-search): use genkit where possible

* chore(firestore-vector-search): update CHANGELOG and bump ext version

* test(firestore-vector-search): fix tests and add more coverage

firestore-vector-search/CHANGELOG.md

@@ -1,3 +1,7 @@
+## Version 0.0.6
+
+refactor - use Firebase Genkit where possible
+
 ## Version 0.0.5
 
 fix - fix backfill and fix npm audit

firestore-vector-search/extension.yaml

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 name: firestore-vector-search
-version: 0.0.5
+version: 0.0.6
 specVersion: v1beta
 
 tags:

firestore-vector-search/functions/__tests__/embeddings/gemini.test.ts

@@ -1,61 +1,84 @@
-import {GeminiAITextEmbedClient} from '../../src/embeddings/client/text/gemini';
-// import { config } from "../../src/config";
+jest.resetModules();
+
+// Mock GoogleGenerativeAI and its methods
+const mockGetGenerativeModel = jest.fn();
+const mockBatchEmbedContents = jest.fn();
 
-// mock config
-// jest.mock("../../src/config", () => ({
-//   ...jest.requireActual("../../src/config"),
-//   geminiApiKey: "test-api-key",
-// }));
+jest.mock('@google/generative-ai', () => ({
+  GoogleGenerativeAI: jest.fn().mockImplementation(() => ({
+    getGenerativeModel: mockGetGenerativeModel,
+  })),
+}));
+
+jest.mock('../../src/config', () => ({
+  config: {
+    geminiApiKey: 'test-api-key',
+  },
+}));
+
+import {GeminiAITextEmbedClient} from '../../src/embeddings/client/text/gemini';
+import {GoogleGenerativeAI} from '@google/generative-ai';
 
 describe('Gemini Embeddings', () => {
-  let embedClient;
+  let embedClient: GeminiAITextEmbedClient;
+
+  beforeEach(async () => {
+    // Reset mocks
+    jest.clearAllMocks();
 
-  beforeEach(() => {
+    // Mock return value for getGenerativeModel
+    mockGetGenerativeModel.mockReturnValue({
+      batchEmbedContents: mockBatchEmbedContents,
+    });
+
+    // Instantiate and initialize the client
     embedClient = new GeminiAITextEmbedClient();
+    await embedClient.initialize();
   });
 
   describe('initialize', () => {
     test('should properly initialize the client', async () => {
-      await embedClient.initialize();
-
       expect(embedClient.client).toBeDefined();
-      //   expect(GoogleGenerativeAI).toHaveBeenCalledWith(config.geminiApiKey);
+      expect(GoogleGenerativeAI).toHaveBeenCalledWith('test-api-key');
     });
   });
 
   describe('getEmbeddings', () => {
     test('should return embeddings for a batch of text', async () => {
-      const mockEmbedContent = jest
-        .fn()
-        .mockResolvedValue({embedding: [1, 2, 3]});
-      embedClient.client = {
-        getGenerativeModel: jest.fn(() => ({
-          embedContent: mockEmbedContent,
-        })),
-      };
+      // Mock batchEmbedContents to resolve with embeddings
+      mockBatchEmbedContents.mockResolvedValueOnce({
+        embeddings: [{values: [1, 2, 3]}, {values: [4, 5, 6]}],
+      });
 
       const batch = ['text1', 'text2'];
       const results = await embedClient.getEmbeddings(batch);
 
-      expect(mockEmbedContent).toHaveBeenCalledTimes(batch.length);
+      expect(mockBatchEmbedContents).toHaveBeenCalledWith({
+        requests: [
+          {content: {parts: [{text: 'text1'}], role: 'user'}},
+          {content: {parts: [{text: 'text2'}], role: 'user'}},
+        ],
+      });
+
       expect(results).toEqual([
         [1, 2, 3],
-        [1, 2, 3],
+        [4, 5, 6],
       ]);
     });
 
     test('should throw an error if the embedding process fails', async () => {
-      embedClient.client = {
-        getGenerativeModel: jest.fn(() => ({
-          embedContent: jest
-            .fn()
-            .mockRejectedValue(new Error('Embedding failed')),
-        })),
-      };
+      // Mock batchEmbedContents to throw an error
+      mockBatchEmbedContents.mockRejectedValueOnce(
+        new Error('Embedding failed')
+      );
 
       await expect(embedClient.getEmbeddings(['text'])).rejects.toThrow(
         'Error with embedding'
       );
+
+      expect(mockBatchEmbedContents).toHaveBeenCalledWith({
+        requests: [{content: {parts: [{text: 'text'}], role: 'user'}}],
+      });
     });
   });
 });

firestore-vector-search/functions/__tests__/embeddings/genkit.test.ts

@@ -0,0 +1,160 @@
+jest.resetModules();
+
+// Mocking `@genkit-ai/googleai` and `@genkit-ai/vertexai`
+jest.mock('@genkit-ai/googleai', () => ({
+  googleAI: jest.fn(),
+  textEmbeddingGecko001: 'gecko-001-model',
+}));
+
+jest.mock('@genkit-ai/vertexai', () => ({
+  vertexAI: jest.fn(),
+  textEmbedding004: 'text-embedding-004-model',
+}));
+
+jest.mock('../../src/config', () => ({
+  config: {
+    geminiApiKey: 'test-api-key',
+    location: 'us-central1',
+  },
+}));
+
+import {GenkitEmbedClient} from '../../src/embeddings/client/genkit';
+import {genkit} from 'genkit';
+import {vertexAI} from '@genkit-ai/vertexai';
+import {googleAI} from '@genkit-ai/googleai';
+
+// Mock the genkit client with properly structured responses
+const mockEmbedMany = jest.fn();
+const mockEmbed = jest.fn();
+jest.mock('genkit', () => ({
+  genkit: jest.fn().mockImplementation(() => ({
+    embedMany: mockEmbedMany,
+    embed: mockEmbed,
+  })),
+}));
+
+describe('GenkitEmbedClient', () => {
+  let embedClient: GenkitEmbedClient;
+  let mockVertexAI: jest.Mock;
+  let mockGoogleAI: jest.Mock;
+
+  beforeEach(() => {
+    jest.clearAllMocks();
+    mockVertexAI = vertexAI as jest.Mock;
+    mockGoogleAI = googleAI as jest.Mock;
+  });
+
+  describe('constructor', () => {
+    test('should initialize with Vertex AI provider', () => {
+      embedClient = new GenkitEmbedClient({
+        provider: 'vertexai',
+        batchSize: 100,
+        dimension: 768,
+      });
+
+      expect(embedClient.provider).toBe('vertexai');
+      expect(embedClient.embedder).toBe('text-embedding-004-model');
+      expect(mockVertexAI).toHaveBeenCalledWith({
+        location: 'us-central1',
+      });
+      expect(genkit).toHaveBeenCalledWith({
+        plugins: [undefined], // because the mock returns undefined
+      });
+    });
+
+    test('should initialize with Google AI provider', () => {
+      embedClient = new GenkitEmbedClient({
+        provider: 'googleai',
+        batchSize: 100,
+        dimension: 768,
+      });
+
+      expect(embedClient.provider).toBe('googleai');
+      expect(embedClient.embedder).toBe('gecko-001-model');
+      expect(mockGoogleAI).toHaveBeenCalledWith({
+        apiKey: 'test-api-key',
+      });
+      expect(genkit).toHaveBeenCalledWith({
+        plugins: [undefined], // because the mock returns undefined
+      });
+    });
+  });
+
+  describe('getEmbeddings', () => {
+    beforeEach(() => {
+      embedClient = new GenkitEmbedClient({
+        provider: 'vertexai',
+        batchSize: 100,
+        dimension: 768,
+      });
+    });
+
+    test('should return embeddings for a batch of inputs', async () => {
+      const mockResults = [{embedding: [1, 2, 3]}, {embedding: [4, 5, 6]}];
+      mockEmbedMany.mockResolvedValueOnce(mockResults);
+
+      const inputs = ['input1', 'input2'];
+      const embeddings = await embedClient.getEmbeddings(inputs);
+
+      expect(mockEmbedMany).toHaveBeenCalledWith({
+        embedder: embedClient.embedder,
+        content: inputs,
+      });
+
+      expect(embeddings).toEqual([
+        [1, 2, 3],
+        [4, 5, 6],
+      ]);
+    });
+
+    test('should throw an error if embedding fails', async () => {
+      mockEmbedMany.mockRejectedValueOnce(new Error('Embedding failed'));
+
+      await expect(embedClient.getEmbeddings(['input'])).rejects.toThrow(
+        'Embedding failed'
+      );
+
+      expect(mockEmbedMany).toHaveBeenCalledWith({
+        embedder: embedClient.embedder,
+        content: ['input'],
+      });
+    });
+  });
+
+  describe('getSingleEmbedding', () => {
+    beforeEach(() => {
+      embedClient = new GenkitEmbedClient({
+        provider: 'googleai',
+        batchSize: 100,
+        dimension: 768,
+      });
+    });
+
+    test('should return a single embedding for an input', async () => {
+      mockEmbed.mockResolvedValueOnce([7, 8, 9]); // Changed to return array directly
+
+      const input = 'input1';
+      const embedding = await embedClient.getSingleEmbedding(input);
+
+      expect(mockEmbed).toHaveBeenCalledWith({
+        embedder: embedClient.embedder,
+        content: input,
+      });
+
+      expect(embedding).toEqual([7, 8, 9]);
+    });
+
+    test('should throw an error if embedding fails', async () => {
+      mockEmbed.mockRejectedValueOnce(new Error('Embedding failed'));
+
+      await expect(embedClient.getSingleEmbedding('input')).rejects.toThrow(
+        'Embedding failed'
+      );
+
+      expect(mockEmbed).toHaveBeenCalledWith({
+        embedder: embedClient.embedder,
+        content: 'input',
+      });
+    });
+  });
+});

firestore-vector-search/functions/jest.config.js

@@ -7,7 +7,7 @@ module.exports = {
   rootDir: './',
   globals: {
     'ts-jest': {
-      tsConfig: '<rootDir>/__tests__/tsconfig.json',
+      tsconfig: '<rootDir>/tsconfig.test.json', // Correct reference to test-specific config
     },
     fetch: global.fetch,
   },