|
| 1 | +import { createServer, type IncomingMessage, type ServerResponse } from "node:http"; |
| 2 | +import type { AddressInfo } from "node:net"; |
| 3 | +import type { MemoryEmbeddingProviderCreateOptions } from "openclaw/plugin-sdk/memory-core-host-engine-embeddings"; |
| 4 | +import { afterEach, describe, expect, it } from "vitest"; |
| 5 | +import { createOpenAICompatibleEmbeddingProvider } from "./embedding-provider.js"; |
| 6 | + |
| 7 | +type CapturedRequest = { |
| 8 | + method: string | undefined; |
| 9 | + url: string | undefined; |
| 10 | + headers: IncomingMessage["headers"]; |
| 11 | + body: Record<string, unknown>; |
| 12 | +}; |
| 13 | + |
| 14 | +type FixtureResponse = { |
| 15 | + object: "list"; |
| 16 | + data: Array<{ |
| 17 | + object?: "embedding"; |
| 18 | + embedding: number[]; |
| 19 | + index: number; |
| 20 | + }>; |
| 21 | + model?: string; |
| 22 | + usage?: { |
| 23 | + prompt_tokens?: number; |
| 24 | + total_tokens?: number; |
| 25 | + }; |
| 26 | +}; |
| 27 | + |
| 28 | +const servers: Array<{ close: () => Promise<void> }> = []; |
| 29 | + |
| 30 | +function createOptions( |
| 31 | + overrides: Partial<MemoryEmbeddingProviderCreateOptions> = {}, |
| 32 | +): MemoryEmbeddingProviderCreateOptions { |
| 33 | + return { |
| 34 | + config: {} as MemoryEmbeddingProviderCreateOptions["config"], |
| 35 | + provider: "openai-compatible", |
| 36 | + model: "text-embedding-bge-m3", |
| 37 | + fallback: "none", |
| 38 | + ...overrides, |
| 39 | + }; |
| 40 | +} |
| 41 | + |
| 42 | +async function readJsonBody(req: IncomingMessage): Promise<Record<string, unknown>> { |
| 43 | + const chunks: Buffer[] = []; |
| 44 | + for await (const chunk of req) { |
| 45 | + chunks.push(Buffer.isBuffer(chunk) ? chunk : Buffer.from(chunk)); |
| 46 | + } |
| 47 | + const text = Buffer.concat(chunks).toString("utf8"); |
| 48 | + return JSON.parse(text) as Record<string, unknown>; |
| 49 | +} |
| 50 | + |
| 51 | +async function startEmbeddingServer(params?: { |
| 52 | + token?: string; |
| 53 | + respond?: (request: CapturedRequest) => FixtureResponse | Record<string, unknown>; |
| 54 | + status?: number; |
| 55 | +}): Promise<{ baseUrl: string; requests: CapturedRequest[] }> { |
| 56 | + const requests: CapturedRequest[] = []; |
| 57 | + const server = createServer(async (req: IncomingMessage, res: ServerResponse) => { |
| 58 | + try { |
| 59 | + const body = await readJsonBody(req); |
| 60 | + const captured: CapturedRequest = { |
| 61 | + method: req.method, |
| 62 | + url: req.url, |
| 63 | + headers: req.headers, |
| 64 | + body, |
| 65 | + }; |
| 66 | + requests.push(captured); |
| 67 | + |
| 68 | + if (params?.token) { |
| 69 | + expect(req.headers.authorization).toBe(`Bearer ${params.token}`); |
| 70 | + } else { |
| 71 | + expect(req.headers.authorization).toBeUndefined(); |
| 72 | + } |
| 73 | + |
| 74 | + res.writeHead(params?.status ?? 200, { "content-type": "application/json" }); |
| 75 | + res.end( |
| 76 | + JSON.stringify( |
| 77 | + params?.respond?.(captured) ?? { |
| 78 | + object: "list", |
| 79 | + data: [{ object: "embedding", embedding: [0.1, 0.2, 0.3], index: 0 }], |
| 80 | + model: body.model, |
| 81 | + }, |
| 82 | + ), |
| 83 | + ); |
| 84 | + } catch (error) { |
| 85 | + res.writeHead(500, { "content-type": "application/json" }); |
| 86 | + res.end(JSON.stringify({ error: error instanceof Error ? error.message : String(error) })); |
| 87 | + } |
| 88 | + }); |
| 89 | + |
| 90 | + await new Promise<void>((resolve, reject) => { |
| 91 | + server.once("error", reject); |
| 92 | + server.listen(0, "127.0.0.1", () => { |
| 93 | + server.off("error", reject); |
| 94 | + resolve(); |
| 95 | + }); |
| 96 | + }); |
| 97 | + |
| 98 | + servers.push({ |
| 99 | + close: () => |
| 100 | + new Promise<void>((resolve, reject) => { |
| 101 | + server.close((error) => (error ? reject(error) : resolve())); |
| 102 | + }), |
| 103 | + }); |
| 104 | + |
| 105 | + const address = server.address() as AddressInfo; |
| 106 | + return { |
| 107 | + baseUrl: `http://127.0.0.1:${address.port}/v1`, |
| 108 | + requests, |
| 109 | + }; |
| 110 | +} |
| 111 | + |
| 112 | +afterEach(async () => { |
| 113 | + const pending = servers.splice(0); |
| 114 | + await Promise.all(pending.map((server) => server.close())); |
| 115 | +}); |
| 116 | + |
| 117 | +describe("openai-compatible embedding provider", () => { |
| 118 | + it("posts OpenAI-compatible embedding requests without warming up during create", async () => { |
| 119 | + const token = "local-test-token"; |
| 120 | + const server = await startEmbeddingServer({ |
| 121 | + token, |
| 122 | + respond: ({ body }) => { |
| 123 | + const input = body.input; |
| 124 | + const texts = Array.isArray(input) ? input : [input]; |
| 125 | + return { |
| 126 | + object: "list", |
| 127 | + data: texts.map((text, index) => ({ |
| 128 | + object: "embedding", |
| 129 | + embedding: [String(text).length, index + 0.25, 1], |
| 130 | + index, |
| 131 | + })), |
| 132 | + model: String(body.model), |
| 133 | + usage: { prompt_tokens: texts.length, total_tokens: texts.length }, |
| 134 | + }; |
| 135 | + }, |
| 136 | + }); |
| 137 | + |
| 138 | + const { provider, client } = await createOpenAICompatibleEmbeddingProvider( |
| 139 | + createOptions({ |
| 140 | + model: "text-embedding-bge-m3", |
| 141 | + outputDimensionality: 1024, |
| 142 | + remote: { |
| 143 | + baseUrl: ` ${server.baseUrl} `, |
| 144 | + apiKey: ` ${token} `, |
| 145 | + headers: { "x-local-runtime": "ollama" }, |
| 146 | + }, |
| 147 | + }), |
| 148 | + ); |
| 149 | + |
| 150 | + expect(provider.id).toBe("openai-compatible"); |
| 151 | + expect(provider.model).toBe("text-embedding-bge-m3"); |
| 152 | + expect(client.baseUrl).toBe(server.baseUrl); |
| 153 | + expect(client.headers.authorization).toBe(`Bearer ${token}`); |
| 154 | + expect(server.requests).toHaveLength(0); |
| 155 | + |
| 156 | + await expect(provider.embedQuery("hello")).resolves.toEqual([5, 0.25, 1]); |
| 157 | + await expect(provider.embedBatch(["a", "abcd"])).resolves.toEqual([ |
| 158 | + [1, 0.25, 1], |
| 159 | + [4, 1.25, 1], |
| 160 | + ]); |
| 161 | + |
| 162 | + expect(server.requests).toHaveLength(2); |
| 163 | + expect(server.requests[0]).toMatchObject({ |
| 164 | + method: "POST", |
| 165 | + url: "/v1/embeddings", |
| 166 | + body: { |
| 167 | + model: "text-embedding-bge-m3", |
| 168 | + input: ["hello"], |
| 169 | + dimensions: 1024, |
| 170 | + }, |
| 171 | + }); |
| 172 | + expect(server.requests[0]?.body).not.toHaveProperty("encoding_format"); |
| 173 | + expect(server.requests[0]?.headers["content-type"]).toContain("application/json"); |
| 174 | + expect(server.requests[0]?.headers.accept).toBe("application/json"); |
| 175 | + expect(server.requests[0]?.headers["x-local-runtime"]).toBe("ollama"); |
| 176 | + expect(server.requests[1]?.body).toEqual({ |
| 177 | + model: "text-embedding-bge-m3", |
| 178 | + input: ["a", "abcd"], |
| 179 | + dimensions: 1024, |
| 180 | + }); |
| 181 | + }); |
| 182 | + |
| 183 | + it("omits Authorization when no apiKey is configured", async () => { |
| 184 | + const server = await startEmbeddingServer(); |
| 185 | + const { provider, client } = await createOpenAICompatibleEmbeddingProvider( |
| 186 | + createOptions({ |
| 187 | + model: "nomic-embed-text", |
| 188 | + remote: { baseUrl: server.baseUrl }, |
| 189 | + }), |
| 190 | + ); |
| 191 | + |
| 192 | + expect(client.headers).not.toHaveProperty("authorization"); |
| 193 | + |
| 194 | + await expect(provider.embedQuery("hello")).resolves.toEqual([0.1, 0.2, 0.3]); |
| 195 | + expect(server.requests[0]?.headers.authorization).toBeUndefined(); |
| 196 | + }); |
| 197 | + |
| 198 | + it.each([ |
| 199 | + { |
| 200 | + runtime: "Ollama", |
| 201 | + response: { |
| 202 | + object: "list", |
| 203 | + data: [{ object: "embedding", embedding: [0.11, 0.12], index: 0 }], |
| 204 | + model: "nomic-embed-text", |
| 205 | + usage: { prompt_tokens: 1, total_tokens: 1 }, |
| 206 | + }, |
| 207 | + }, |
| 208 | + { |
| 209 | + runtime: "llama.cpp llama-server", |
| 210 | + response: { |
| 211 | + object: "list", |
| 212 | + data: [{ object: "embedding", embedding: [0.21, 0.22], index: 0 }], |
| 213 | + model: "bge-small-en-v1.5", |
| 214 | + }, |
| 215 | + }, |
| 216 | + { |
| 217 | + runtime: "vLLM", |
| 218 | + response: { |
| 219 | + object: "list", |
| 220 | + data: [{ object: "embedding", embedding: [0.31, 0.32], index: 0 }], |
| 221 | + model: "intfloat/e5-small-v2", |
| 222 | + }, |
| 223 | + }, |
| 224 | + { |
| 225 | + runtime: "LocalAI", |
| 226 | + response: { |
| 227 | + object: "list", |
| 228 | + data: [{ object: "embedding", embedding: [0.41, 0.42], index: 0 }], |
| 229 | + model: "text-embedding-ada-002", |
| 230 | + }, |
| 231 | + }, |
| 232 | + { |
| 233 | + runtime: "TGI-compatible server", |
| 234 | + response: { |
| 235 | + object: "list", |
| 236 | + data: [{ object: "embedding", embedding: [0.51, 0.52], index: 0 }], |
| 237 | + model: "tei-bge-small", |
| 238 | + }, |
| 239 | + }, |
| 240 | + { |
| 241 | + runtime: "llamafile", |
| 242 | + response: { |
| 243 | + object: "list", |
| 244 | + data: [{ object: "embedding", embedding: [0.61, 0.62], index: 0 }], |
| 245 | + model: "all-MiniLM-L6-v2", |
| 246 | + }, |
| 247 | + }, |
| 248 | + ] satisfies Array<{ runtime: string; response: FixtureResponse }>)( |
| 249 | + "parses $runtime OpenAI-compatible embedding responses through the same path", |
| 250 | + async ({ response }) => { |
| 251 | + const server = await startEmbeddingServer({ respond: () => response }); |
| 252 | + const { provider } = await createOpenAICompatibleEmbeddingProvider( |
| 253 | + createOptions({ |
| 254 | + model: response.model ?? "embedding-model", |
| 255 | + remote: { baseUrl: server.baseUrl }, |
| 256 | + }), |
| 257 | + ); |
| 258 | + |
| 259 | + await expect(provider.embedQuery("hello")).resolves.toEqual(response.data[0]?.embedding); |
| 260 | + expect(server.requests[0]?.url).toBe("/v1/embeddings"); |
| 261 | + expect(server.requests[0]?.body).toEqual({ |
| 262 | + model: response.model ?? "embedding-model", |
| 263 | + input: ["hello"], |
| 264 | + }); |
| 265 | + }, |
| 266 | + ); |
| 267 | + |
| 268 | + it("reports missing required config with actionable keys", async () => { |
| 269 | + await expect( |
| 270 | + createOpenAICompatibleEmbeddingProvider( |
| 271 | + createOptions({ remote: { baseUrl: " " }, model: "text-embedding-bge-m3" }), |
| 272 | + ), |
| 273 | + ).rejects.toThrow("embedding.baseUrl"); |
| 274 | + await expect( |
| 275 | + createOpenAICompatibleEmbeddingProvider( |
| 276 | + createOptions({ remote: { baseUrl: "http://127.0.0.1:11434/v1" }, model: " " }), |
| 277 | + ), |
| 278 | + ).rejects.toThrow("embedding.model"); |
| 279 | + }); |
| 280 | + |
| 281 | + it("keeps remote parser failures behind the provider-specific error prefix", async () => { |
| 282 | + const server = await startEmbeddingServer({ respond: () => ({ data: [] }) }); |
| 283 | + const { provider } = await createOpenAICompatibleEmbeddingProvider( |
| 284 | + createOptions({ |
| 285 | + model: "text-embedding-bge-m3", |
| 286 | + remote: { baseUrl: server.baseUrl }, |
| 287 | + }), |
| 288 | + ); |
| 289 | + |
| 290 | + await expect(provider.embedQuery("hello")).rejects.toThrow( |
| 291 | + "openai-compatible embeddings failed: malformed JSON response", |
| 292 | + ); |
| 293 | + }); |
| 294 | +}); |
0 commit comments