plugin-apply.spec.ts 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383
  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 SessionProjectionRegistry from '@deepseek-ai/dsh-session-projection'
  15. import JsonlSessionPersistence from '@deepseek-ai/dsh-session-persistence-jsonl'
  16. import * as jsonrpc from '../src/index.ts'
  17. /**
  18. * Mount the real namespace plugin with in-memory stdio and exit hooks. Covers
  19. * the full transport/server path, response-before-exit shutdown exactly once,
  20. * and bare-fiber disposal without process exit.
  21. */
  22. /** One ordered frame, write completion, or exit observation. */
  23. type WireEvent =
  24. | { kind: 'frame'; frame: Record<string, unknown> }
  25. | { kind: 'write-complete'; ids: (string | number)[] }
  26. | { kind: 'root-disposed' }
  27. | { kind: 'exit'; code: number }
  28. interface ApplyHarness {
  29. ctx: Context
  30. /** The plugin fiber used by the bare-dispose case. */
  31. fiber: Awaited<ReturnType<Context['plugin']>>
  32. /** Frames, write completions, and exits in observation order. */
  33. events: WireEvent[]
  34. outputErrors: Error[]
  35. send(frame: Record<string, unknown>): void
  36. sendRaw(text: string): void
  37. frames(): Record<string, unknown>[]
  38. exits(): number[]
  39. waitForFrame(predicate: (frame: Record<string, unknown>) => boolean, description: string): Promise<Record<string, unknown>>
  40. dispose(): Promise<void>
  41. }
  42. /** Adapter whose route registration is the delayed Loader entry's readiness fact. */
  43. class DelayedAdapter extends LlmAdapter {
  44. async * stream(_options: GenerateOptions): AsyncIterable<StreamChunk> {
  45. throw new Error('not exercised')
  46. }
  47. }
  48. /** Poll asynchronous output for up to five seconds. */
  49. async function waitFor<T>(get: () => T | undefined, description: string): Promise<T> {
  50. const deadline = Date.now() + 5000
  51. for (;;) {
  52. const value = get()
  53. if (value !== undefined) return value
  54. if (Date.now() > deadline) throw new Error(`timed out waiting for ${description}`)
  55. await new Promise(resolve => setTimeout(resolve, 5))
  56. }
  57. }
  58. /** Drain asynchronous work before a negative assertion. */
  59. async function settle(): Promise<void> {
  60. await new Promise(resolve => setTimeout(resolve, 25))
  61. }
  62. /** Mount the real plugin on a minimal harness with in-memory stdio and exit. */
  63. async function mountPlugin(
  64. storageDir: string,
  65. options: {
  66. writeDelayMs?: number
  67. failFlush?: boolean
  68. beforeServer?: (ctx: Context) => Promise<void> | void
  69. } = {},
  70. ): Promise<ApplyHarness> {
  71. const ctx = new Context()
  72. await mountAgentLoopTestDependencies(ctx)
  73. await ctx.plugin(SessionProjectionRegistry)
  74. await ctx.plugin(AgentLoop, { agents: [] })
  75. await ctx.plugin(JsonlSessionPersistence, { root: storageDir })
  76. await new Promise(resolve => setTimeout(resolve, 50))
  77. await options.beforeServer?.(ctx)
  78. const input = new PassThrough()
  79. const events: WireEvent[] = []
  80. const outputErrors: Error[] = []
  81. let pendingOutput = ''
  82. // Record frame admission separately from write completion so delayed output
  83. // tests the flush barrier.
  84. const output = new Writable({
  85. write(chunk: Buffer, _encoding, callback) {
  86. const ids: (string | number)[] = []
  87. pendingOutput += chunk.toString('utf8')
  88. for (;;) {
  89. const newline = pendingOutput.indexOf('\n')
  90. if (newline < 0) break
  91. const line = pendingOutput.slice(0, newline).trim()
  92. pendingOutput = pendingOutput.slice(newline + 1)
  93. if (line) {
  94. const frame = JSON.parse(line) as Record<string, unknown>
  95. events.push({ kind: 'frame', frame })
  96. if (typeof frame.id === 'string' || typeof frame.id === 'number') ids.push(frame.id)
  97. }
  98. }
  99. const complete = (): void => {
  100. if (options.failFlush === true && chunk.length === 0) {
  101. callback(new Error('flush callback failed'))
  102. return
  103. }
  104. events.push({ kind: 'write-complete', ids })
  105. callback()
  106. }
  107. if ((options.writeDelayMs ?? 0) > 0) setTimeout(complete, options.writeDelayMs)
  108. else complete()
  109. },
  110. })
  111. output.on('error', (error: Error) => { outputErrors.push(error) })
  112. const exit = (code: number): void => { events.push({ kind: 'exit', code }) }
  113. ctx.effect(() => () => { events.push({ kind: 'root-disposed' }) }, 'jsonrpc test root-disposal witness')
  114. const fiber = await ctx.plugin(jsonrpc, {
  115. input,
  116. output,
  117. exit,
  118. })
  119. const frames = (): Record<string, unknown>[] =>
  120. events.flatMap(event => event.kind === 'frame' ? [event.frame] : [])
  121. return {
  122. ctx,
  123. fiber,
  124. events,
  125. outputErrors,
  126. send: (frame) => { input.write(`${JSON.stringify(frame)}\n`) },
  127. sendRaw: (text) => { input.write(text) },
  128. frames,
  129. exits: () => events.flatMap(event => event.kind === 'exit' ? [event.code] : []),
  130. waitForFrame: (predicate, description) => waitFor(() => frames().find(predicate), description),
  131. dispose: async () => { await ctx.fiber.dispose() },
  132. }
  133. }
  134. const servers: Server[] = []
  135. afterEach(async () => {
  136. await Promise.all(servers.splice(0).map(server => new Promise(resolve => server.close(resolve))))
  137. vi.unstubAllEnvs()
  138. })
  139. /** Keyless SSE endpoint for completing a prompt turn. */
  140. async function mockCompletionServer(): Promise<{ url: string; requests: unknown[] }> {
  141. const requests: unknown[] = []
  142. const server = createServer((request: IncomingMessage, response: ServerResponse) => {
  143. let body = ''
  144. request.on('data', (chunk: Buffer) => { body += chunk.toString('utf8') })
  145. request.on('end', () => {
  146. requests.push(JSON.parse(body))
  147. response.writeHead(200, { 'content-type': 'text/event-stream' })
  148. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null,"reasoning_content":""}}]}\n\n')
  149. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  150. response.write('data: {"choices":[{"delta":{"content":""},"finish_reason":"stop"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  151. response.write('data: [DONE]\n\n')
  152. response.end()
  153. })
  154. })
  155. servers.push(server)
  156. await new Promise<void>(resolve => server.listen(0, '127.0.0.1', resolve))
  157. const address = server.address()
  158. if (address === null || typeof address === 'string') throw new Error('no port')
  159. return { url: `http://127.0.0.1:${address.port}`, requests }
  160. }
  161. describe('dsh-sdk-jsonrpc-server plugin apply', () => {
  162. it('serves initialize over the injected stdio pair', async () => {
  163. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-init-'))
  164. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  165. const harness = await mountPlugin(storageDir)
  166. try {
  167. harness.send({ jsonrpc: '2.0', id: 'init-1', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'apply-model' } })
  168. const response = await harness.waitForFrame(frame => frame.id === 'init-1', 'initialize response')
  169. expect(response).toEqual({
  170. jsonrpc: '2.0',
  171. id: 'init-1',
  172. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime', version: '0.0.1' } },
  173. })
  174. expect(harness.exits()).toEqual([])
  175. } finally {
  176. await harness.dispose()
  177. await rm(storageDir, { recursive: true, force: true })
  178. }
  179. })
  180. it('waits for Loader-owned adapter registration before initialize', async () => {
  181. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-readiness-'))
  182. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  183. let markStarted!: () => void
  184. let release!: () => void
  185. const started = new Promise<void>((resolve) => { markStarted = resolve })
  186. const ready = new Promise<void>((resolve) => { release = resolve })
  187. let delayedEntry: Promise<string> | undefined
  188. const harness = await mountPlugin(storageDir, {
  189. beforeServer: async (ctx) => {
  190. await ctx.plugin(Loader)
  191. ctx.loader.builtins['delayed-readiness'] = {
  192. inject: ['llm'],
  193. async apply(entryCtx: Context) {
  194. markStarted()
  195. await ready
  196. entryCtx.llm.registerAdapter(['delayed-private'], new DelayedAdapter())
  197. },
  198. }
  199. delayedEntry = ctx.loader.create({ name: 'cordis:delayed-readiness' })
  200. await started
  201. },
  202. })
  203. try {
  204. const initialize = {
  205. jsonrpc: '2.0',
  206. id: 'init-delayed',
  207. method: 'initialize',
  208. params: { cwd: storageDir, provider: 'delayed-private', model: 'apply-model' },
  209. }
  210. const probe = { jsonrpc: '2.0', id: 'probe-during-delay', method: 'nope/unknown' }
  211. harness.sendRaw(`${JSON.stringify(initialize)}\n${JSON.stringify(probe)}\n`)
  212. // The transport processes independent requests concurrently. Receiving
  213. // this later probe proves the preceding initialize handler has reached
  214. // its Loader wait, without relying on a scheduler delay.
  215. await harness.waitForFrame(frame => frame.id === 'probe-during-delay', 'probe while initialize waits')
  216. expect(harness.frames().some(frame => frame.id === 'init-delayed')).toBe(false)
  217. release()
  218. await delayedEntry
  219. const response = await harness.waitForFrame(frame => frame.id === 'init-delayed', 'initialize response after Loader settlement')
  220. expect(response).toMatchObject({
  221. id: 'init-delayed',
  222. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime' } },
  223. })
  224. expect(harness.ctx.llm.listProviders()).toContainEqual({ id: 'delayed-private', name: 'delayed-private' })
  225. } finally {
  226. release()
  227. await Promise.allSettled(delayedEntry === undefined ? [] : [delayedEntry])
  228. await harness.dispose()
  229. await rm(storageDir, { recursive: true, force: true })
  230. }
  231. })
  232. it('drives a session/prompt turn end-to-end and forwards session notifications as output frames', async () => {
  233. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-prompt-'))
  234. const llmServer = await mockCompletionServer()
  235. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  236. vi.stubEnv('DEEPSEEK_BASE_URL', llmServer.url)
  237. const harness = await mountPlugin(storageDir)
  238. try {
  239. harness.send({ jsonrpc: '2.0', id: 1, method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'dsagent-model' } })
  240. await harness.waitForFrame(frame => frame.id === 1, 'initialize response')
  241. harness.send({
  242. jsonrpc: '2.0',
  243. id: 2,
  244. method: 'session/prompt',
  245. params: { sessionId: 'main', contentBlocks: [{ type: 'text', text: 'fix it' }] },
  246. })
  247. const response = await harness.waitForFrame(frame => frame.id === 2, 'prompt response')
  248. expect((response.result as { messageId?: unknown }).messageId).toBeTypeOf('string')
  249. await harness.waitForFrame(
  250. frame => frame.method === 'session.status'
  251. && (frame.params as { status?: string } | undefined)?.status === 'idle',
  252. 'idle session status',
  253. )
  254. expect(llmServer.requests).toHaveLength(1)
  255. const body = llmServer.requests[0] as { model: string; messages: { role: string }[] }
  256. expect(body.model).toBe('dsagent-model')
  257. expect(body.messages.at(-1)?.role).toBe('user')
  258. // Notifications use the same transport and arrive as id-less frames.
  259. const notifications = harness.frames().filter(frame => frame.id === undefined)
  260. expect(notifications.some(frame => frame.method === 'session.event')).toBe(true)
  261. expect(notifications.findLast(frame => frame.method === 'session.status')).toMatchObject({
  262. jsonrpc: '2.0',
  263. params: { sessionId: 'main', status: 'idle' },
  264. })
  265. } finally {
  266. await harness.dispose()
  267. await rm(storageDir, { recursive: true, force: true })
  268. }
  269. })
  270. it('answers shutdown before exiting 0 exactly once, even against a racing second shutdown', async () => {
  271. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-shutdown-'))
  272. const harness = await mountPlugin(storageDir, { writeDelayMs: 10 })
  273. try {
  274. // One chunk makes the two deferred exit callbacks race.
  275. const first = { jsonrpc: '2.0', id: 'sd-1', method: 'shutdown' }
  276. const second = { jsonrpc: '2.0', id: 'sd-2', method: 'shutdown' }
  277. harness.sendRaw(`${JSON.stringify(first)}\n${JSON.stringify(second)}\n`)
  278. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit recorder call')
  279. expect(harness.exits()).toEqual([0])
  280. // Both response writes and the flush barrier complete before exit.
  281. const exitIndex = harness.events.findIndex(event => event.kind === 'exit')
  282. const firstResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-1')
  283. const secondResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-2')
  284. const firstComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-1'))
  285. const secondComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-2'))
  286. const flushComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.length === 0)
  287. const rootDisposed = harness.events.findIndex(event => event.kind === 'root-disposed')
  288. expect(firstResponse).toBeGreaterThanOrEqual(0)
  289. expect(secondResponse).toBeGreaterThanOrEqual(0)
  290. expect(firstComplete).toBeGreaterThan(firstResponse)
  291. expect(secondComplete).toBeGreaterThan(secondResponse)
  292. expect(flushComplete).toBeGreaterThan(firstComplete)
  293. expect(flushComplete).toBeGreaterThan(secondComplete)
  294. expect(rootDisposed).toBeGreaterThan(flushComplete)
  295. expect(exitIndex).toBeGreaterThan(rootDisposed)
  296. await settle()
  297. expect(harness.exits()).toEqual([0])
  298. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  299. const before = harness.frames().length
  300. harness.send({ jsonrpc: '2.0', id: 'after-exit', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  301. await settle()
  302. expect(harness.frames().length).toBe(before)
  303. } finally {
  304. await harness.dispose()
  305. await rm(storageDir, { recursive: true, force: true })
  306. }
  307. })
  308. it('still disposes and exits once when the flush callback fails', async () => {
  309. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-flush-failure-'))
  310. const harness = await mountPlugin(storageDir, { failFlush: true })
  311. try {
  312. harness.send({ jsonrpc: '2.0', id: 'sd-fail', method: 'shutdown' })
  313. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit after flush failure')
  314. await settle()
  315. expect(harness.exits()).toEqual([0])
  316. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  317. expect(harness.outputErrors.map(error => error.message)).toEqual(['flush callback failed'])
  318. const before = harness.frames().length
  319. harness.send({ jsonrpc: '2.0', id: 'after-flush-failure', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  320. await settle()
  321. expect(harness.frames().length).toBe(before)
  322. } finally {
  323. await harness.dispose()
  324. await rm(storageDir, { recursive: true, force: true })
  325. }
  326. })
  327. it('stops serving on a bare fiber dispose (HMR-style unload) without calling exit', async () => {
  328. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-dispose-'))
  329. const harness = await mountPlugin(storageDir)
  330. try {
  331. // Prove the handler-rejection path is live before disposal.
  332. harness.send({ jsonrpc: '2.0', id: 'probe-1', method: 'nope/unknown' })
  333. const error = await harness.waitForFrame(frame => frame.id === 'probe-1', 'error response for unknown method')
  334. expect(error.error).toMatchObject({
  335. code: -32603,
  336. message: 'unknown DeepSeek Harness SDK runtime method: nope/unknown',
  337. })
  338. await harness.fiber.dispose()
  339. expect(harness.events.some(event => event.kind === 'root-disposed')).toBe(false)
  340. const before = harness.frames().length
  341. harness.send({ jsonrpc: '2.0', id: 'probe-2', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  342. await settle()
  343. expect(harness.frames().length).toBe(before)
  344. expect(harness.exits()).toEqual([])
  345. } finally {
  346. await harness.dispose()
  347. await rm(storageDir, { recursive: true, force: true })
  348. }
  349. })
  350. })