plugin-apply.spec.ts 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381
  1. import { createServer } from 'node:http'
  2. import type { IncomingMessage, Server, ServerResponse } from 'node:http'
  3. import { mkdtemp, rm } from 'node:fs/promises'
  4. import { join } from 'node:path'
  5. import { tmpdir } from 'node:os'
  6. import { PassThrough, Writable } from 'node:stream'
  7. import { afterEach, describe, expect, it, vi } from 'vitest'
  8. import { Context } from '@deepseek-ai/cordis'
  9. import Loader from '@deepseek-ai/cordis-plugin-loader'
  10. import AgentLoop from '@deepseek-ai/dsh-agent-loop'
  11. import { mountAgentLoopTestDependencies } from '@deepseek-ai/dsh-agent-loop-testkit'
  12. import { LlmAdapter } from '@deepseek-ai/dsh-llm'
  13. import type { GenerateOptions, StreamChunk } from '@deepseek-ai/dsh-llm'
  14. import JsonlSessionPersistence from '@deepseek-ai/dsh-session-persistence-jsonl'
  15. import * as jsonrpc from '../src/index.ts'
  16. /**
  17. * Mount the real namespace plugin with in-memory stdio and exit hooks. Covers
  18. * the full transport/server path, response-before-exit shutdown exactly once,
  19. * and bare-fiber disposal without process exit.
  20. */
  21. /** One ordered frame, write completion, or exit observation. */
  22. type WireEvent =
  23. | { kind: 'frame'; frame: Record<string, unknown> }
  24. | { kind: 'write-complete'; ids: (string | number)[] }
  25. | { kind: 'root-disposed' }
  26. | { kind: 'exit'; code: number }
  27. interface ApplyHarness {
  28. ctx: Context
  29. /** The plugin fiber used by the bare-dispose case. */
  30. fiber: Awaited<ReturnType<Context['plugin']>>
  31. /** Frames, write completions, and exits in observation order. */
  32. events: WireEvent[]
  33. outputErrors: Error[]
  34. send(frame: Record<string, unknown>): void
  35. sendRaw(text: string): void
  36. frames(): Record<string, unknown>[]
  37. exits(): number[]
  38. waitForFrame(predicate: (frame: Record<string, unknown>) => boolean, description: string): Promise<Record<string, unknown>>
  39. dispose(): Promise<void>
  40. }
  41. /** Adapter whose route registration is the delayed Loader entry's readiness fact. */
  42. class DelayedAdapter extends LlmAdapter {
  43. async * stream(_options: GenerateOptions): AsyncIterable<StreamChunk> {
  44. throw new Error('not exercised')
  45. }
  46. }
  47. /** Poll asynchronous output for up to five seconds. */
  48. async function waitFor<T>(get: () => T | undefined, description: string): Promise<T> {
  49. const deadline = Date.now() + 5000
  50. for (;;) {
  51. const value = get()
  52. if (value !== undefined) return value
  53. if (Date.now() > deadline) throw new Error(`timed out waiting for ${description}`)
  54. await new Promise(resolve => setTimeout(resolve, 5))
  55. }
  56. }
  57. /** Drain asynchronous work before a negative assertion. */
  58. async function settle(): Promise<void> {
  59. await new Promise(resolve => setTimeout(resolve, 25))
  60. }
  61. /** Mount the real plugin on a minimal harness with in-memory stdio and exit. */
  62. async function mountPlugin(
  63. storageDir: string,
  64. options: {
  65. writeDelayMs?: number
  66. failFlush?: boolean
  67. beforeServer?: (ctx: Context) => Promise<void> | void
  68. } = {},
  69. ): Promise<ApplyHarness> {
  70. const ctx = new Context()
  71. await mountAgentLoopTestDependencies(ctx)
  72. await ctx.plugin(AgentLoop, { agents: [] })
  73. await ctx.plugin(JsonlSessionPersistence, { root: storageDir })
  74. await new Promise(resolve => setTimeout(resolve, 50))
  75. await options.beforeServer?.(ctx)
  76. const input = new PassThrough()
  77. const events: WireEvent[] = []
  78. const outputErrors: Error[] = []
  79. let pendingOutput = ''
  80. // Record frame admission separately from write completion so delayed output
  81. // tests the flush barrier.
  82. const output = new Writable({
  83. write(chunk: Buffer, _encoding, callback) {
  84. const ids: (string | number)[] = []
  85. pendingOutput += chunk.toString('utf8')
  86. for (;;) {
  87. const newline = pendingOutput.indexOf('\n')
  88. if (newline < 0) break
  89. const line = pendingOutput.slice(0, newline).trim()
  90. pendingOutput = pendingOutput.slice(newline + 1)
  91. if (line) {
  92. const frame = JSON.parse(line) as Record<string, unknown>
  93. events.push({ kind: 'frame', frame })
  94. if (typeof frame.id === 'string' || typeof frame.id === 'number') ids.push(frame.id)
  95. }
  96. }
  97. const complete = (): void => {
  98. if (options.failFlush === true && chunk.length === 0) {
  99. callback(new Error('flush callback failed'))
  100. return
  101. }
  102. events.push({ kind: 'write-complete', ids })
  103. callback()
  104. }
  105. if ((options.writeDelayMs ?? 0) > 0) setTimeout(complete, options.writeDelayMs)
  106. else complete()
  107. },
  108. })
  109. output.on('error', (error: Error) => { outputErrors.push(error) })
  110. const exit = (code: number): void => { events.push({ kind: 'exit', code }) }
  111. ctx.effect(() => () => { events.push({ kind: 'root-disposed' }) }, 'jsonrpc test root-disposal witness')
  112. const fiber = await ctx.plugin(jsonrpc, {
  113. input,
  114. output,
  115. exit,
  116. })
  117. const frames = (): Record<string, unknown>[] =>
  118. events.flatMap(event => event.kind === 'frame' ? [event.frame] : [])
  119. return {
  120. ctx,
  121. fiber,
  122. events,
  123. outputErrors,
  124. send: (frame) => { input.write(`${JSON.stringify(frame)}\n`) },
  125. sendRaw: (text) => { input.write(text) },
  126. frames,
  127. exits: () => events.flatMap(event => event.kind === 'exit' ? [event.code] : []),
  128. waitForFrame: (predicate, description) => waitFor(() => frames().find(predicate), description),
  129. dispose: async () => { await ctx.fiber.dispose() },
  130. }
  131. }
  132. const servers: Server[] = []
  133. afterEach(async () => {
  134. await Promise.all(servers.splice(0).map(server => new Promise(resolve => server.close(resolve))))
  135. vi.unstubAllEnvs()
  136. })
  137. /** Keyless SSE endpoint for completing a prompt turn. */
  138. async function mockCompletionServer(): Promise<{ url: string; requests: unknown[] }> {
  139. const requests: unknown[] = []
  140. const server = createServer((request: IncomingMessage, response: ServerResponse) => {
  141. let body = ''
  142. request.on('data', (chunk: Buffer) => { body += chunk.toString('utf8') })
  143. request.on('end', () => {
  144. requests.push(JSON.parse(body))
  145. response.writeHead(200, { 'content-type': 'text/event-stream' })
  146. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null,"reasoning_content":""}}]}\n\n')
  147. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  148. response.write('data: {"choices":[{"delta":{"content":""},"finish_reason":"stop"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  149. response.write('data: [DONE]\n\n')
  150. response.end()
  151. })
  152. })
  153. servers.push(server)
  154. await new Promise<void>(resolve => server.listen(0, '127.0.0.1', resolve))
  155. const address = server.address()
  156. if (address === null || typeof address === 'string') throw new Error('no port')
  157. return { url: `http://127.0.0.1:${address.port}`, requests }
  158. }
  159. describe('dsh-sdk-jsonrpc-server plugin apply', () => {
  160. it('serves initialize over the injected stdio pair', async () => {
  161. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-init-'))
  162. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  163. const harness = await mountPlugin(storageDir)
  164. try {
  165. harness.send({ jsonrpc: '2.0', id: 'init-1', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'apply-model' } })
  166. const response = await harness.waitForFrame(frame => frame.id === 'init-1', 'initialize response')
  167. expect(response).toEqual({
  168. jsonrpc: '2.0',
  169. id: 'init-1',
  170. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime', version: '0.0.1' } },
  171. })
  172. expect(harness.exits()).toEqual([])
  173. } finally {
  174. await harness.dispose()
  175. await rm(storageDir, { recursive: true, force: true })
  176. }
  177. })
  178. it('waits for Loader-owned adapter registration before initialize', async () => {
  179. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-readiness-'))
  180. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  181. let markStarted!: () => void
  182. let release!: () => void
  183. const started = new Promise<void>((resolve) => { markStarted = resolve })
  184. const ready = new Promise<void>((resolve) => { release = resolve })
  185. let delayedEntry: Promise<string> | undefined
  186. const harness = await mountPlugin(storageDir, {
  187. beforeServer: async (ctx) => {
  188. await ctx.plugin(Loader)
  189. ctx.loader.builtins['delayed-readiness'] = {
  190. inject: ['llm'],
  191. async apply(entryCtx: Context) {
  192. markStarted()
  193. await ready
  194. entryCtx.llm.registerAdapter(['delayed-private'], new DelayedAdapter())
  195. },
  196. }
  197. delayedEntry = ctx.loader.create({ name: 'cordis:delayed-readiness' })
  198. await started
  199. },
  200. })
  201. try {
  202. const initialize = {
  203. jsonrpc: '2.0',
  204. id: 'init-delayed',
  205. method: 'initialize',
  206. params: { cwd: storageDir, provider: 'delayed-private', model: 'apply-model' },
  207. }
  208. const probe = { jsonrpc: '2.0', id: 'probe-during-delay', method: 'nope/unknown' }
  209. harness.sendRaw(`${JSON.stringify(initialize)}\n${JSON.stringify(probe)}\n`)
  210. // The transport processes independent requests concurrently. Receiving
  211. // this later probe proves the preceding initialize handler has reached
  212. // its Loader wait, without relying on a scheduler delay.
  213. await harness.waitForFrame(frame => frame.id === 'probe-during-delay', 'probe while initialize waits')
  214. expect(harness.frames().some(frame => frame.id === 'init-delayed')).toBe(false)
  215. release()
  216. await delayedEntry
  217. const response = await harness.waitForFrame(frame => frame.id === 'init-delayed', 'initialize response after Loader settlement')
  218. expect(response).toMatchObject({
  219. id: 'init-delayed',
  220. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime' } },
  221. })
  222. expect(harness.ctx.llm.listProviders()).toContainEqual({ id: 'delayed-private', name: 'delayed-private' })
  223. } finally {
  224. release()
  225. await Promise.allSettled(delayedEntry === undefined ? [] : [delayedEntry])
  226. await harness.dispose()
  227. await rm(storageDir, { recursive: true, force: true })
  228. }
  229. })
  230. it('drives a session/prompt turn end-to-end and forwards session notifications as output frames', async () => {
  231. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-prompt-'))
  232. const llmServer = await mockCompletionServer()
  233. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  234. vi.stubEnv('DEEPSEEK_BASE_URL', llmServer.url)
  235. const harness = await mountPlugin(storageDir)
  236. try {
  237. harness.send({ jsonrpc: '2.0', id: 1, method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'dsagent-model' } })
  238. await harness.waitForFrame(frame => frame.id === 1, 'initialize response')
  239. harness.send({
  240. jsonrpc: '2.0',
  241. id: 2,
  242. method: 'session/prompt',
  243. params: { sessionId: 'main', contentBlocks: [{ type: 'text', text: 'fix it' }] },
  244. })
  245. const response = await harness.waitForFrame(frame => frame.id === 2, 'prompt response')
  246. expect((response.result as { messageId?: unknown }).messageId).toBeTypeOf('string')
  247. await harness.waitForFrame(
  248. frame => frame.method === 'session.status'
  249. && (frame.params as { status?: string } | undefined)?.status === 'idle',
  250. 'idle session status',
  251. )
  252. expect(llmServer.requests).toHaveLength(1)
  253. const body = llmServer.requests[0] as { model: string; messages: { role: string }[] }
  254. expect(body.model).toBe('dsagent-model')
  255. expect(body.messages.at(-1)?.role).toBe('user')
  256. // Notifications use the same transport and arrive as id-less frames.
  257. const notifications = harness.frames().filter(frame => frame.id === undefined)
  258. expect(notifications.some(frame => frame.method === 'session.event')).toBe(true)
  259. expect(notifications.findLast(frame => frame.method === 'session.status')).toMatchObject({
  260. jsonrpc: '2.0',
  261. params: { sessionId: 'main', status: 'idle' },
  262. })
  263. } finally {
  264. await harness.dispose()
  265. await rm(storageDir, { recursive: true, force: true })
  266. }
  267. })
  268. it('answers shutdown before exiting 0 exactly once, even against a racing second shutdown', async () => {
  269. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-shutdown-'))
  270. const harness = await mountPlugin(storageDir, { writeDelayMs: 10 })
  271. try {
  272. // One chunk makes the two deferred exit callbacks race.
  273. const first = { jsonrpc: '2.0', id: 'sd-1', method: 'shutdown' }
  274. const second = { jsonrpc: '2.0', id: 'sd-2', method: 'shutdown' }
  275. harness.sendRaw(`${JSON.stringify(first)}\n${JSON.stringify(second)}\n`)
  276. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit recorder call')
  277. expect(harness.exits()).toEqual([0])
  278. // Both response writes and the flush barrier complete before exit.
  279. const exitIndex = harness.events.findIndex(event => event.kind === 'exit')
  280. const firstResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-1')
  281. const secondResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-2')
  282. const firstComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-1'))
  283. const secondComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-2'))
  284. const flushComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.length === 0)
  285. const rootDisposed = harness.events.findIndex(event => event.kind === 'root-disposed')
  286. expect(firstResponse).toBeGreaterThanOrEqual(0)
  287. expect(secondResponse).toBeGreaterThanOrEqual(0)
  288. expect(firstComplete).toBeGreaterThan(firstResponse)
  289. expect(secondComplete).toBeGreaterThan(secondResponse)
  290. expect(flushComplete).toBeGreaterThan(firstComplete)
  291. expect(flushComplete).toBeGreaterThan(secondComplete)
  292. expect(rootDisposed).toBeGreaterThan(flushComplete)
  293. expect(exitIndex).toBeGreaterThan(rootDisposed)
  294. await settle()
  295. expect(harness.exits()).toEqual([0])
  296. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  297. const before = harness.frames().length
  298. harness.send({ jsonrpc: '2.0', id: 'after-exit', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  299. await settle()
  300. expect(harness.frames().length).toBe(before)
  301. } finally {
  302. await harness.dispose()
  303. await rm(storageDir, { recursive: true, force: true })
  304. }
  305. })
  306. it('still disposes and exits once when the flush callback fails', async () => {
  307. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-flush-failure-'))
  308. const harness = await mountPlugin(storageDir, { failFlush: true })
  309. try {
  310. harness.send({ jsonrpc: '2.0', id: 'sd-fail', method: 'shutdown' })
  311. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit after flush failure')
  312. await settle()
  313. expect(harness.exits()).toEqual([0])
  314. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  315. expect(harness.outputErrors.map(error => error.message)).toEqual(['flush callback failed'])
  316. const before = harness.frames().length
  317. harness.send({ jsonrpc: '2.0', id: 'after-flush-failure', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  318. await settle()
  319. expect(harness.frames().length).toBe(before)
  320. } finally {
  321. await harness.dispose()
  322. await rm(storageDir, { recursive: true, force: true })
  323. }
  324. })
  325. it('stops serving on a bare fiber dispose (HMR-style unload) without calling exit', async () => {
  326. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-dispose-'))
  327. const harness = await mountPlugin(storageDir)
  328. try {
  329. // Prove the handler-rejection path is live before disposal.
  330. harness.send({ jsonrpc: '2.0', id: 'probe-1', method: 'nope/unknown' })
  331. const error = await harness.waitForFrame(frame => frame.id === 'probe-1', 'error response for unknown method')
  332. expect(error.error).toMatchObject({
  333. code: -32603,
  334. message: 'unknown DeepSeek Harness SDK runtime method: nope/unknown',
  335. })
  336. await harness.fiber.dispose()
  337. expect(harness.events.some(event => event.kind === 'root-disposed')).toBe(false)
  338. const before = harness.frames().length
  339. harness.send({ jsonrpc: '2.0', id: 'probe-2', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  340. await settle()
  341. expect(harness.frames().length).toBe(before)
  342. expect(harness.exits()).toEqual([])
  343. } finally {
  344. await harness.dispose()
  345. await rm(storageDir, { recursive: true, force: true })
  346. }
  347. })
  348. })