plugin-apply.spec.ts 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363
  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 * as agentCore from '@deepseek-ai/dsh-agent-spine-demo'
  11. import JsonlSessionPersistence from '@deepseek-ai/dsh-session-persistence-jsonl'
  12. import * as jsonrpc from '../src/index.ts'
  13. /**
  14. * Mount the real namespace plugin with in-memory stdio and exit hooks. Covers
  15. * the full transport/server path, response-before-exit shutdown exactly once,
  16. * and bare-fiber disposal without process exit.
  17. */
  18. /** One ordered frame, write completion, or exit observation. */
  19. type WireEvent =
  20. | { kind: 'frame'; frame: Record<string, unknown> }
  21. | { kind: 'write-complete'; ids: (string | number)[] }
  22. | { kind: 'root-disposed' }
  23. | { kind: 'exit'; code: number }
  24. interface ApplyHarness {
  25. ctx: Context
  26. /** The plugin fiber used by the bare-dispose case. */
  27. fiber: Awaited<ReturnType<Context['plugin']>>
  28. /** Frames, write completions, and exits in observation order. */
  29. events: WireEvent[]
  30. outputErrors: Error[]
  31. send(frame: Record<string, unknown>): void
  32. sendRaw(text: string): void
  33. frames(): Record<string, unknown>[]
  34. exits(): number[]
  35. waitForFrame(predicate: (frame: Record<string, unknown>) => boolean, description: string): Promise<Record<string, unknown>>
  36. dispose(): Promise<void>
  37. }
  38. /** Poll asynchronous output for up to five seconds. */
  39. async function waitFor<T>(get: () => T | undefined, description: string): Promise<T> {
  40. const deadline = Date.now() + 5000
  41. for (;;) {
  42. const value = get()
  43. if (value !== undefined) return value
  44. if (Date.now() > deadline) throw new Error(`timed out waiting for ${description}`)
  45. await new Promise(resolve => setTimeout(resolve, 5))
  46. }
  47. }
  48. /** Drain asynchronous work before a negative assertion. */
  49. async function settle(): Promise<void> {
  50. await new Promise(resolve => setTimeout(resolve, 25))
  51. }
  52. /** Mount the real plugin on a minimal harness with in-memory stdio and exit. */
  53. async function mountPlugin(
  54. storageDir: string,
  55. options: {
  56. writeDelayMs?: number
  57. failFlush?: boolean
  58. beforeServer?: (ctx: Context) => Promise<void> | void
  59. } = {},
  60. ): Promise<ApplyHarness> {
  61. const ctx = new Context()
  62. await ctx.plugin(agentCore, { workspaceContext: false })
  63. await ctx.plugin(JsonlSessionPersistence, { root: storageDir })
  64. await new Promise(resolve => setTimeout(resolve, 50))
  65. await options.beforeServer?.(ctx)
  66. const input = new PassThrough()
  67. const events: WireEvent[] = []
  68. const outputErrors: Error[] = []
  69. let pendingOutput = ''
  70. // Record frame admission separately from write completion so delayed output
  71. // tests the flush barrier.
  72. const output = new Writable({
  73. write(chunk: Buffer, _encoding, callback) {
  74. const ids: (string | number)[] = []
  75. pendingOutput += chunk.toString('utf8')
  76. for (;;) {
  77. const newline = pendingOutput.indexOf('\n')
  78. if (newline < 0) break
  79. const line = pendingOutput.slice(0, newline).trim()
  80. pendingOutput = pendingOutput.slice(newline + 1)
  81. if (line) {
  82. const frame = JSON.parse(line) as Record<string, unknown>
  83. events.push({ kind: 'frame', frame })
  84. if (typeof frame.id === 'string' || typeof frame.id === 'number') ids.push(frame.id)
  85. }
  86. }
  87. const complete = (): void => {
  88. if (options.failFlush === true && chunk.length === 0) {
  89. callback(new Error('flush callback failed'))
  90. return
  91. }
  92. events.push({ kind: 'write-complete', ids })
  93. callback()
  94. }
  95. if ((options.writeDelayMs ?? 0) > 0) setTimeout(complete, options.writeDelayMs)
  96. else complete()
  97. },
  98. })
  99. output.on('error', (error: Error) => { outputErrors.push(error) })
  100. const exit = (code: number): void => { events.push({ kind: 'exit', code }) }
  101. ctx.effect(() => () => { events.push({ kind: 'root-disposed' }) }, 'jsonrpc test root-disposal witness')
  102. const fiber = await ctx.plugin(jsonrpc, { input, output, exit })
  103. const frames = (): Record<string, unknown>[] =>
  104. events.flatMap(event => event.kind === 'frame' ? [event.frame] : [])
  105. return {
  106. ctx,
  107. fiber,
  108. events,
  109. outputErrors,
  110. send: (frame) => { input.write(`${JSON.stringify(frame)}\n`) },
  111. sendRaw: (text) => { input.write(text) },
  112. frames,
  113. exits: () => events.flatMap(event => event.kind === 'exit' ? [event.code] : []),
  114. waitForFrame: (predicate, description) => waitFor(() => frames().find(predicate), description),
  115. dispose: async () => { await ctx.fiber.dispose() },
  116. }
  117. }
  118. const servers: Server[] = []
  119. afterEach(async () => {
  120. await Promise.all(servers.splice(0).map(server => new Promise(resolve => server.close(resolve))))
  121. vi.unstubAllEnvs()
  122. })
  123. /** Keyless SSE endpoint for completing a prompt turn. */
  124. async function mockCompletionServer(): Promise<{ url: string; requests: unknown[] }> {
  125. const requests: unknown[] = []
  126. const server = createServer((request: IncomingMessage, response: ServerResponse) => {
  127. let body = ''
  128. request.on('data', (chunk: Buffer) => { body += chunk.toString('utf8') })
  129. request.on('end', () => {
  130. requests.push(JSON.parse(body))
  131. response.writeHead(200, { 'content-type': 'text/event-stream' })
  132. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null,"reasoning_content":""}}]}\n\n')
  133. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  134. response.write('data: {"choices":[{"delta":{"content":""},"finish_reason":"stop"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  135. response.write('data: [DONE]\n\n')
  136. response.end()
  137. })
  138. })
  139. servers.push(server)
  140. await new Promise<void>(resolve => server.listen(0, '127.0.0.1', resolve))
  141. const address = server.address()
  142. if (address === null || typeof address === 'string') throw new Error('no port')
  143. return { url: `http://127.0.0.1:${address.port}`, requests }
  144. }
  145. describe('dsh-sdk-jsonrpc-server plugin apply', () => {
  146. it('serves initialize over the injected stdio pair', async () => {
  147. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-init-'))
  148. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  149. const harness = await mountPlugin(storageDir)
  150. try {
  151. harness.send({ jsonrpc: '2.0', id: 'init-1', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'apply-model' } })
  152. const response = await harness.waitForFrame(frame => frame.id === 'init-1', 'initialize response')
  153. expect(response).toEqual({
  154. jsonrpc: '2.0',
  155. id: 'init-1',
  156. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime', version: '0.0.1' } },
  157. })
  158. expect(harness.exits()).toEqual([])
  159. } finally {
  160. await harness.dispose()
  161. await rm(storageDir, { recursive: true, force: true })
  162. }
  163. })
  164. it('does not answer initialize until async sibling Loader entries settle', async () => {
  165. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-readiness-'))
  166. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  167. let markStarted!: () => void
  168. let release!: () => void
  169. const started = new Promise<void>((resolve) => { markStarted = resolve })
  170. const ready = new Promise<void>((resolve) => { release = resolve })
  171. let delayedEntry: Promise<string> | undefined
  172. const harness = await mountPlugin(storageDir, {
  173. beforeServer: async (ctx) => {
  174. await ctx.plugin(Loader)
  175. ctx.loader.builtins['delayed-readiness'] = {
  176. async apply() {
  177. markStarted()
  178. await ready
  179. },
  180. }
  181. delayedEntry = ctx.loader.create({ name: 'cordis:delayed-readiness' })
  182. await started
  183. },
  184. })
  185. try {
  186. const initialize = {
  187. jsonrpc: '2.0',
  188. id: 'init-delayed',
  189. method: 'initialize',
  190. params: { cwd: storageDir, provider: 'deepseek-official', model: 'apply-model' },
  191. }
  192. const probe = { jsonrpc: '2.0', id: 'probe-during-delay', method: 'nope/unknown' }
  193. harness.sendRaw(`${JSON.stringify(initialize)}\n${JSON.stringify(probe)}\n`)
  194. // The transport processes independent requests concurrently. Receiving
  195. // this later probe proves the preceding initialize handler has reached
  196. // its Loader wait, without relying on a scheduler delay.
  197. await harness.waitForFrame(frame => frame.id === 'probe-during-delay', 'probe while initialize waits')
  198. expect(harness.frames().some(frame => frame.id === 'init-delayed')).toBe(false)
  199. release()
  200. await delayedEntry
  201. const response = await harness.waitForFrame(frame => frame.id === 'init-delayed', 'initialize response after Loader settlement')
  202. expect(response).toMatchObject({
  203. id: 'init-delayed',
  204. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime' } },
  205. })
  206. } finally {
  207. release()
  208. await Promise.allSettled(delayedEntry === undefined ? [] : [delayedEntry])
  209. await harness.dispose()
  210. await rm(storageDir, { recursive: true, force: true })
  211. }
  212. })
  213. it('drives a session/prompt turn end-to-end and forwards session notifications as output frames', async () => {
  214. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-prompt-'))
  215. const llmServer = await mockCompletionServer()
  216. vi.stubEnv('DEEPSEEK_API_KEY', 'test-key')
  217. vi.stubEnv('DEEPSEEK_BASE_URL', llmServer.url)
  218. const harness = await mountPlugin(storageDir)
  219. try {
  220. harness.send({ jsonrpc: '2.0', id: 1, method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'dsagent-model' } })
  221. await harness.waitForFrame(frame => frame.id === 1, 'initialize response')
  222. harness.send({
  223. jsonrpc: '2.0',
  224. id: 2,
  225. method: 'session/prompt',
  226. params: { sessionId: 'main', contentBlocks: [{ type: 'text', text: 'fix it' }] },
  227. })
  228. const response = await harness.waitForFrame(frame => frame.id === 2, 'prompt response')
  229. expect((response.result as { messageId?: unknown }).messageId).toBeTypeOf('string')
  230. await harness.waitForFrame(
  231. frame => frame.method === 'session.status'
  232. && (frame.params as { status?: string } | undefined)?.status === 'idle',
  233. 'idle session status',
  234. )
  235. expect(llmServer.requests).toHaveLength(1)
  236. const body = llmServer.requests[0] as { model: string; messages: { role: string }[] }
  237. expect(body.model).toBe('dsagent-model')
  238. expect(body.messages.at(-1)?.role).toBe('user')
  239. // Notifications use the same transport and arrive as id-less frames.
  240. const notifications = harness.frames().filter(frame => frame.id === undefined)
  241. expect(notifications.some(frame => frame.method === 'session.event')).toBe(true)
  242. expect(notifications.findLast(frame => frame.method === 'session.status')).toMatchObject({
  243. jsonrpc: '2.0',
  244. params: { sessionId: 'main', status: 'idle' },
  245. })
  246. } finally {
  247. await harness.dispose()
  248. await rm(storageDir, { recursive: true, force: true })
  249. }
  250. })
  251. it('answers shutdown before exiting 0 exactly once, even against a racing second shutdown', async () => {
  252. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-shutdown-'))
  253. const harness = await mountPlugin(storageDir, { writeDelayMs: 10 })
  254. try {
  255. // One chunk makes the two deferred exit callbacks race.
  256. const first = { jsonrpc: '2.0', id: 'sd-1', method: 'shutdown' }
  257. const second = { jsonrpc: '2.0', id: 'sd-2', method: 'shutdown' }
  258. harness.sendRaw(`${JSON.stringify(first)}\n${JSON.stringify(second)}\n`)
  259. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit recorder call')
  260. expect(harness.exits()).toEqual([0])
  261. // Both response writes and the flush barrier complete before exit.
  262. const exitIndex = harness.events.findIndex(event => event.kind === 'exit')
  263. const firstResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-1')
  264. const secondResponse = harness.events.findIndex(event => event.kind === 'frame' && event.frame.id === 'sd-2')
  265. const firstComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-1'))
  266. const secondComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.includes('sd-2'))
  267. const flushComplete = harness.events.findIndex(event => event.kind === 'write-complete' && event.ids.length === 0)
  268. const rootDisposed = harness.events.findIndex(event => event.kind === 'root-disposed')
  269. expect(firstResponse).toBeGreaterThanOrEqual(0)
  270. expect(secondResponse).toBeGreaterThanOrEqual(0)
  271. expect(firstComplete).toBeGreaterThan(firstResponse)
  272. expect(secondComplete).toBeGreaterThan(secondResponse)
  273. expect(flushComplete).toBeGreaterThan(firstComplete)
  274. expect(flushComplete).toBeGreaterThan(secondComplete)
  275. expect(rootDisposed).toBeGreaterThan(flushComplete)
  276. expect(exitIndex).toBeGreaterThan(rootDisposed)
  277. await settle()
  278. expect(harness.exits()).toEqual([0])
  279. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  280. const before = harness.frames().length
  281. harness.send({ jsonrpc: '2.0', id: 'after-exit', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  282. await settle()
  283. expect(harness.frames().length).toBe(before)
  284. } finally {
  285. await harness.dispose()
  286. await rm(storageDir, { recursive: true, force: true })
  287. }
  288. })
  289. it('still disposes and exits once when the flush callback fails', async () => {
  290. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-flush-failure-'))
  291. const harness = await mountPlugin(storageDir, { failFlush: true })
  292. try {
  293. harness.send({ jsonrpc: '2.0', id: 'sd-fail', method: 'shutdown' })
  294. await waitFor(() => harness.exits().length > 0 ? true : undefined, 'exit after flush failure')
  295. await settle()
  296. expect(harness.exits()).toEqual([0])
  297. expect(harness.events.filter(event => event.kind === 'root-disposed')).toHaveLength(1)
  298. expect(harness.outputErrors.map(error => error.message)).toEqual(['flush callback failed'])
  299. const before = harness.frames().length
  300. harness.send({ jsonrpc: '2.0', id: 'after-flush-failure', 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('stops serving on a bare fiber dispose (HMR-style unload) without calling exit', async () => {
  309. const storageDir = await mkdtemp(join(tmpdir(), 'dsh-jsonrpc-apply-dispose-'))
  310. const harness = await mountPlugin(storageDir)
  311. try {
  312. // Prove the handler-rejection path is live before disposal.
  313. harness.send({ jsonrpc: '2.0', id: 'probe-1', method: 'nope/unknown' })
  314. const error = await harness.waitForFrame(frame => frame.id === 'probe-1', 'error response for unknown method')
  315. expect(error.error).toMatchObject({
  316. code: -32603,
  317. message: 'unknown DeepSeek Harness SDK runtime method: nope/unknown',
  318. })
  319. await harness.fiber.dispose()
  320. expect(harness.events.some(event => event.kind === 'root-disposed')).toBe(false)
  321. const before = harness.frames().length
  322. harness.send({ jsonrpc: '2.0', id: 'probe-2', method: 'initialize', params: { cwd: storageDir, provider: 'deepseek-official', model: 'x' } })
  323. await settle()
  324. expect(harness.frames().length).toBe(before)
  325. expect(harness.exits()).toEqual([])
  326. } finally {
  327. await harness.dispose()
  328. await rm(storageDir, { recursive: true, force: true })
  329. }
  330. })
  331. })