keyless-smoke.e2e.ts 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202
  1. import { createServer } from 'node:http'
  2. import { mkdtemp, readFile, readdir, rm } from 'node:fs/promises'
  3. import { tmpdir } from 'node:os'
  4. import { join } from 'node:path'
  5. import { fileURLToPath } from 'node:url'
  6. import { promisify } from 'node:util'
  7. import { zstdDecompress } from 'node:zlib'
  8. import { execa } from 'execa'
  9. import { describe, expect, it } from 'vitest'
  10. const binScript = fileURLToPath(new URL('../../../packages/sdk/python-runtime/src/packaged-bin.ts', import.meta.url))
  11. const configPath = fileURLToPath(new URL('../cordis.yml', import.meta.url))
  12. const repoRoot = fileURLToPath(new URL('../../..', import.meta.url))
  13. const decompress = promisify(zstdDecompress)
  14. function waitForLine(
  15. lines: string[],
  16. predicate: (value: Record<string, unknown>) => boolean,
  17. stderr: () => string,
  18. ): Promise<Record<string, unknown>> {
  19. return new Promise((resolve, reject) => {
  20. const deadline = Date.now() + 30_000
  21. const poll = (): void => {
  22. while (lines.length > 0) {
  23. const line = lines.shift()!
  24. if (!line.trim()) continue
  25. try {
  26. const value = JSON.parse(line) as Record<string, unknown>
  27. if (predicate(value)) {
  28. resolve(value)
  29. return
  30. }
  31. } catch {
  32. reject(new Error(`non-JSON stdout from JSON-RPC agent runtime: ${line}`))
  33. return
  34. }
  35. }
  36. if (Date.now() >= deadline) {
  37. reject(new Error(`timed out waiting for JSON-RPC response; stderr=${stderr()}`))
  38. return
  39. }
  40. setTimeout(poll, 10)
  41. }
  42. poll()
  43. })
  44. }
  45. describe('Python SDK runtime carrier keyless smoke', () => {
  46. it.each([
  47. { label: 'reports max-token turns with the default mapping config', envValue: undefined },
  48. { label: 'reports max-token turns with mapping enabled through env', envValue: 'true' },
  49. { label: 'reports max-token turns with mapping disabled through env', envValue: 'false' },
  50. ])('$label', async ({ envValue }) => {
  51. const root = await mkdtemp(join(tmpdir(), 'dsh-python-sdk-runtime-smoke-'))
  52. const modelRequests: Record<string, unknown>[] = []
  53. const modelServer = createServer((request, response) => {
  54. let body = ''
  55. request.setEncoding('utf8')
  56. request.on('data', (chunk: string) => { body += chunk })
  57. request.on('end', () => {
  58. modelRequests.push(JSON.parse(body) as Record<string, unknown>)
  59. response.writeHead(200, { 'content-type': 'text/event-stream' })
  60. response.write('data: {"choices":[{"delta":{"role":"assistant","content":null}}]}\n\n')
  61. response.write('data: {"choices":[{"delta":{"content":"done"}}]}\n\n')
  62. response.write('data: {"choices":[{"delta":{},"finish_reason":"length"}],"usage":{"prompt_tokens":3,"completion_tokens":1}}\n\n')
  63. response.end('data: [DONE]\n\n')
  64. })
  65. })
  66. await new Promise<void>(resolve => modelServer.listen(0, '127.0.0.1', resolve))
  67. const address = modelServer.address()
  68. if (address === null || typeof address === 'string') throw new Error('model server did not bind a TCP port')
  69. // The line-predicate protocol driving below is the genuinely custom part;
  70. // execa owns spawn, the deadline, and exit settlement around it.
  71. const child = execa(process.execPath, [
  72. '--import',
  73. 'tsx',
  74. binScript,
  75. configPath,
  76. ], {
  77. cwd: repoRoot,
  78. env: {
  79. DEEPSEEK_API_KEY: 'keyless-smoke-no-call',
  80. DEEPSEEK_BASE_URL: `http://127.0.0.1:${address.port}`,
  81. DSH_CWD: root,
  82. DSH_SESSION_ROOT: join(root, '.sessions'),
  83. ...(envValue === undefined ? {} : { DSH_MAX_TOKENS_AS_SUCCESS: envValue }),
  84. },
  85. timeout: 35_000,
  86. killSignal: 'SIGKILL',
  87. reject: false,
  88. })
  89. const lines: string[] = []
  90. let stdoutBuffer = ''
  91. let stderr = ''
  92. child.stdout.on('data', (chunk: Buffer) => {
  93. stdoutBuffer += chunk.toString('utf8')
  94. const parts = stdoutBuffer.split('\n')
  95. stdoutBuffer = parts.pop() ?? ''
  96. lines.push(...parts)
  97. })
  98. child.stderr.on('data', (chunk: Buffer) => { stderr += chunk.toString('utf8') })
  99. try {
  100. child.stdin.write(`${JSON.stringify({
  101. jsonrpc: '2.0',
  102. id: 1,
  103. method: 'initialize',
  104. params: { cwd: root, provider: 'deepseek-official', model: 'deepseek-v4-pro', maxTokens: 1234 },
  105. })}\n`)
  106. const initialized = await waitForLine(lines, value => value.id === 1, () => stderr)
  107. expect(initialized).toMatchObject({
  108. jsonrpc: '2.0',
  109. id: 1,
  110. result: { serverInfo: { name: 'deepseek-harness-sdk-runtime' } },
  111. })
  112. child.stdin.write(`${JSON.stringify({
  113. jsonrpc: '2.0',
  114. id: 2,
  115. method: 'session/prompt',
  116. params: { sessionId: 'main', contentBlocks: [{ type: 'text', text: 'inspect tools' }] },
  117. })}\n`)
  118. const prompt = await waitForLine(lines, value => value.id === 2, () => stderr)
  119. expect(prompt).toMatchObject({
  120. jsonrpc: '2.0',
  121. id: 2,
  122. result: { messageId: expect.any(String) as unknown },
  123. })
  124. const turnEnd = await waitForLine(lines, (value) => {
  125. if (value.method !== 'session.event') return false
  126. const params = value.params as Record<string, unknown> | undefined
  127. const event = params?.event as Record<string, unknown> | undefined
  128. return params?.sessionId === 'main' && event?.type === 'turn/end'
  129. }, () => stderr)
  130. expect(turnEnd).toMatchObject({
  131. jsonrpc: '2.0',
  132. method: 'session.event',
  133. params: {
  134. sessionId: 'main',
  135. event: {
  136. type: 'turn/end',
  137. data: { reason: { kind: 'max-tokens' } },
  138. },
  139. },
  140. })
  141. const tools = modelRequests[0]?.tools as { function?: { name?: string } }[]
  142. expect(modelRequests[0]?.max_tokens).toBe(1234)
  143. expect(tools.map(tool => tool.function?.name).sort()).toEqual([
  144. 'bash',
  145. 'edit',
  146. 'read',
  147. 'subagent',
  148. 'todo_write',
  149. 'write',
  150. ])
  151. child.stdin.write(`${JSON.stringify({ jsonrpc: '2.0', id: 3, method: 'shutdown' })}\n`)
  152. const shutdown = await waitForLine(lines, value => value.id === 3, () => stderr)
  153. expect(shutdown).toMatchObject({ jsonrpc: '2.0', id: 3, result: {} })
  154. const exit = await child
  155. expect(exit.exitCode, `signal=${String(exit.signal)}; stderr=${stderr}`).toBe(0)
  156. const sessionsRoot = join(root, '.sessions')
  157. const files = await readdir(sessionsRoot, { recursive: true })
  158. const log = files.find(file => file.endsWith('.jsonl.zstd'))
  159. expect(log).toBeDefined()
  160. const compressed = await readFile(join(sessionsRoot, log!))
  161. expect(compressed.subarray(0, 4).toString('hex')).toBe('28b52ffd')
  162. expect(JSON.parse((await decompress(compressed)).toString())).toMatchObject({ type: 'session', id: 'main' })
  163. } finally {
  164. // No-op after exit; reject: false settles on every outcome, so cleanup never races teardown.
  165. child.kill('SIGKILL')
  166. await child
  167. await new Promise<void>(resolve => modelServer.close(() => { resolve() }))
  168. await rm(root, { recursive: true, force: true })
  169. }
  170. }, 40_000)
  171. it('rejects an invalid max-token success env value', async () => {
  172. const { exitCode, stdout, stderr } = await execa(process.execPath, [
  173. '--import',
  174. 'tsx',
  175. binScript,
  176. configPath,
  177. ], {
  178. cwd: repoRoot,
  179. env: {
  180. DEEPSEEK_API_KEY: 'keyless-smoke-no-call',
  181. DSH_MAX_TOKENS_AS_SUCCESS: 'sometimes',
  182. },
  183. stdin: 'ignore',
  184. timeout: 25_000,
  185. killSignal: 'SIGKILL',
  186. reject: false,
  187. })
  188. expect(exitCode, stderr).toBe(1)
  189. expect(stdout).toBe('')
  190. expect(stderr).toContain('plugin tree failed to load')
  191. expect(stderr).toContain('failed to apply loader entry sdk-jsonrpc-server (@deepseek-ai/dsh-sdk-jsonrpc-server)')
  192. expect(stderr).toContain('sometimes')
  193. }, 30_000)
  194. })