manual-compaction.spec.ts 36 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888
  1. import { describe, expect, it, vi } from 'vitest'
  2. import { Context } from '@deepseek-ai/cordis'
  3. import AgentLoop from '@deepseek-ai/dsh-agent-loop'
  4. import { mountAgentLoopTestDependencies } from '@deepseek-ai/dsh-agent-loop-testkit'
  5. import InvariantRegistry from '@deepseek-ai/dsh-invariants'
  6. import { CommandId } from '@deepseek-ai/dsh-commands/brand'
  7. import * as SessionInvariant from '@deepseek-ai/dsh-session/invariant'
  8. import * as AgentInvariant from '@deepseek-ai/dsh-agent/invariant'
  9. import * as AgentLoopInvariant from '@deepseek-ai/dsh-agent-loop/invariant'
  10. import * as CompactionInvariant from '@deepseek-ai/dsh-compaction/invariant'
  11. import * as CompactionBasicInvariant from '@deepseek-ai/dsh-compaction-basic/invariant'
  12. import { BasicCompactionEngine } from '@deepseek-ai/dsh-compaction-basic'
  13. import { CompactionId, isCompactCheckpointSource, ManualCompactionError } from '@deepseek-ai/dsh-compaction'
  14. import type { CompactionResult } from '@deepseek-ai/dsh-compaction'
  15. import {
  16. createAssistantMessage,
  17. createUserMessage,
  18. LlmAdapter,
  19. } from '@deepseek-ai/dsh-llm'
  20. import type {
  21. ContentBlock,
  22. LlmResolvedModelInfo,
  23. Message,
  24. StreamChunk,
  25. TokenUsage,
  26. } from '@deepseek-ai/dsh-llm'
  27. import SessionStore, { Session, SessionId, type SessionEvent } from '@deepseek-ai/dsh-session'
  28. import SessionProjectionRegistry from '@deepseek-ai/dsh-session-projection'
  29. import LlmRuntime from '@deepseek-ai/dsh-llm'
  30. import TokenMeter from '@deepseek-ai/dsh-token-meter'
  31. import type { Agent } from '@deepseek-ai/dsh-agent'
  32. import type {
  33. SummarizationInput,
  34. SummaryResult,
  35. } from '@deepseek-ai/dsh-compaction-basic/src/summarizer.ts'
  36. const MODEL = 'mock'
  37. const SIGNAL = new AbortController().signal
  38. const PROMPT = 'older conversation history '.repeat(60)
  39. /** A summarizer under test control: it can block, fail, or mutate mid-call. */
  40. class GatedCompactionEngine extends BasicCompactionEngine {
  41. summary: ContentBlock[] = [{ type: 'text', text: 'checkpoint' }]
  42. rawOutput: ContentBlock[] | undefined
  43. usage: TokenUsage | undefined
  44. error: unknown
  45. gate: Promise<undefined> | undefined
  46. duringSummary: (() => void) | undefined
  47. calls: SummarizationInput[] = []
  48. override async summarize(
  49. input: SummarizationInput,
  50. _agent: Agent,
  51. _signal?: AbortSignal,
  52. ): Promise<SummaryResult> {
  53. this.calls.push(input)
  54. this.duringSummary?.()
  55. if (this.gate !== undefined) await this.gate
  56. if (this.error !== undefined) throw this.error
  57. return {
  58. summary: this.summary,
  59. ...this.rawOutput === undefined ? {} : { rawOutput: this.rawOutput },
  60. provider: 'summary-provider',
  61. model: 'summary-model',
  62. ...this.usage === undefined ? {} : { usage: this.usage },
  63. }
  64. }
  65. }
  66. /** One text answer per request, with a context window large enough to avoid pressure. */
  67. class TextAdapter extends LlmAdapter {
  68. readonly requests: Message[][] = []
  69. override resolveModel(provider: string, model: string): Promise<LlmResolvedModelInfo> {
  70. return Promise.resolve({
  71. provider,
  72. id: model,
  73. name: model,
  74. context: { contextWindow: 100_000 },
  75. })
  76. }
  77. override async * stream(options: { messages: readonly Message[] }): AsyncIterable<StreamChunk> {
  78. this.requests.push([...options.messages])
  79. yield { type: 'block-start', index: 0, blockType: 'text' }
  80. yield { type: 'block-end', index: 0, block: { type: 'text', text: 'answer' } }
  81. yield { type: 'finish', reason: { kind: 'stop' } }
  82. }
  83. }
  84. interface LoopHarness {
  85. readonly ctx: Context
  86. readonly agent: Agent
  87. readonly compact: GatedCompactionEngine
  88. readonly adapter: TextAdapter
  89. readonly log: string[]
  90. }
  91. /** Real loop, session store, and invariant companions around manual compaction. */
  92. async function loopHarness(): Promise<LoopHarness> {
  93. const ctx = new Context()
  94. await mountAgentLoopTestDependencies(ctx)
  95. await ctx.plugin(InvariantRegistry)
  96. await ctx.plugin(SessionInvariant)
  97. await ctx.plugin(AgentInvariant)
  98. await ctx.plugin(AgentLoopInvariant)
  99. await ctx.plugin(CompactionInvariant)
  100. await ctx.plugin(CompactionBasicInvariant)
  101. await ctx.plugin(SessionProjectionRegistry)
  102. await ctx.plugin(AgentLoop, { agents: [] })
  103. await ctx.plugin(TokenMeter)
  104. const adapter = new TextAdapter()
  105. ctx.llm.registerAdapter([MODEL], adapter)
  106. const compact = new GatedCompactionEngine(ctx, { auto: false })
  107. const agent = ctx.agentLoop.create(SessionId('manual-compact'), { provider: MODEL, model: MODEL })
  108. const log: string[] = []
  109. ctx.on('session/event', (_session, event) => {
  110. if (event.type === 'turn/start') log.push('turn/start')
  111. if (event.type === 'turn/end') log.push('turn/end')
  112. if (event.type === 'compaction/start') log.push(`compaction/start:${String(event.data.turn)}`)
  113. if (event.type === 'compaction/summary') log.push('compaction/summary')
  114. if (event.type === 'compaction/end') log.push(`compaction/end:${String(event.data.turn)}`)
  115. if (event.type === 'user/message') log.push('user/message')
  116. })
  117. ctx.on('session/flush', () => { log.push('flush') })
  118. return { ctx, agent, compact, adapter, log }
  119. }
  120. /** Drive one real turn so the closed history holds a compactable older span. */
  121. async function seedHistory(harness: LoopHarness): Promise<void> {
  122. harness.agent.followup(createUserMessage({
  123. content: [{ type: 'text', text: PROMPT }],
  124. source: { kind: 'user' },
  125. }))
  126. await harness.agent.whenIdle()
  127. harness.log.length = 0
  128. }
  129. /** Text of every derived model-visible message, in request order. */
  130. function derivedText(session: Session): string[] {
  131. return session.deriveMessages().map((message: Message) => message.content
  132. .map(block => block.type === 'text' ? block.text : '')
  133. .join(''))
  134. }
  135. /** Await one classified manual-compaction rejection. */
  136. async function rejection(operation: Promise<unknown> | (() => Promise<unknown>)): Promise<ManualCompactionError> {
  137. let caught: unknown
  138. try {
  139. const value = await (typeof operation === 'function' ? operation() : operation)
  140. throw new Error(`expected a rejection, resolved with ${String(value)}`)
  141. } catch (error: unknown) {
  142. caught = error
  143. }
  144. if (!(caught instanceof ManualCompactionError)) {
  145. throw new Error(`expected a ManualCompactionError, got ${String(caught)}`)
  146. }
  147. return caught
  148. }
  149. /** The Error a classified failure wraps. */
  150. function causeOf(error: ManualCompactionError): Error {
  151. const { cause } = error
  152. if (!(cause instanceof Error)) throw new Error(`expected an Error cause, got ${String(cause)}`)
  153. return cause
  154. }
  155. function deferred(): { promise: Promise<undefined>; resolve: () => void } {
  156. const { promise, resolve } = Promise.withResolvers<undefined>()
  157. return { promise, resolve: () => { resolve(undefined) } }
  158. }
  159. /** A closed-tail session with compactable exchanges and no live agent. */
  160. function closedConversation(turns = 2, lastTurnNumber = turns): Session {
  161. const session = Session.create(SessionId(`closed-${turns}-${lastTurnNumber}`))
  162. for (let index = 1; index <= turns; index += 1) {
  163. const turn = index === turns ? lastTurnNumber : index
  164. session.append('turn/start', { turn })
  165. session.append('user/message', createUserMessage({
  166. content: [{ type: 'text', text: `${PROMPT} ${turn}` }],
  167. source: { kind: 'user' },
  168. }), { surfaceOp: 'append' })
  169. session.append('step/start', { turn, step: 1 })
  170. if (index === 1) {
  171. session.append('request/header', {
  172. header: { config: { provider: MODEL, model: MODEL } },
  173. reason: 'initial',
  174. })
  175. }
  176. session.append('assistant/message', {
  177. turn,
  178. step: 1,
  179. message: createAssistantMessage({
  180. content: [{ type: 'text', text: `answer ${turn}` }],
  181. source: { provider: MODEL, model: MODEL },
  182. }),
  183. }, { surfaceOp: 'append' })
  184. session.append('step/end', { turn, step: 1 })
  185. session.append('turn/end', { turn, reason: { kind: 'completed' } })
  186. }
  187. return session
  188. }
  189. /** A fake idle agent whose maintenance claim is scripted per test. */
  190. function fakeAgent(
  191. session: Session,
  192. reserve: () => (() => void) | undefined,
  193. maintenanceSignal = new AbortController().signal,
  194. ): Agent {
  195. return {
  196. session,
  197. options: { provider: MODEL, model: MODEL },
  198. runMaintenance<T>(task: (signal: AbortSignal) => Promise<T>): Promise<T> {
  199. const release = reserve()
  200. if (release === undefined) throw new Error('agent already has active work')
  201. return task(maintenanceSignal).finally(release)
  202. },
  203. } as unknown as Agent
  204. }
  205. /** Service over a store-detached session for failure classification. */
  206. function detachedService(): { ctx: Context; compact: GatedCompactionEngine; flushes: () => number } {
  207. const ctx = new Context()
  208. void new LlmRuntime(ctx)
  209. void new SessionStore(ctx)
  210. new SessionProjectionRegistry(ctx)
  211. void new TokenMeter(ctx)
  212. ctx.llm.registerAdapter([MODEL], new TextAdapter())
  213. let flushes = 0
  214. vi.spyOn(ctx.sessions, 'flush').mockImplementation(() => {
  215. flushes += 1
  216. return Promise.resolve(false)
  217. })
  218. return { ctx, compact: new GatedCompactionEngine(ctx, { auto: false }), flushes: () => flushes }
  219. }
  220. function compactEvents(session: Session): Array<Session['events'][number]> {
  221. return session.events.filter(event => event.type.startsWith('compaction/'))
  222. }
  223. describe('compactNow through the real loop', () => {
  224. it('holds a prompt accepted during summarization until the standalone bracket is flushed', async () => {
  225. const harness = await loopHarness()
  226. const { agent, compact, adapter, log } = harness
  227. await seedHistory(harness)
  228. const gate = deferred()
  229. compact.gate = gate.promise
  230. const running = compact.compactNow(agent, SIGNAL)
  231. await Promise.resolve()
  232. expect(log).toEqual(['compaction/start:null'])
  233. agent.followup(createUserMessage({
  234. content: [{ type: 'text', text: 'after compaction' }],
  235. source: { kind: 'user' },
  236. }))
  237. await new Promise<void>((resolve) => { setTimeout(resolve, 5) })
  238. expect(agent.status).toBe('idle')
  239. expect(adapter.requests).toHaveLength(1)
  240. expect(log).toEqual(['compaction/start:null'])
  241. gate.resolve()
  242. const result = await running
  243. expect(result).not.toBeNull()
  244. await agent.whenIdle()
  245. const start = log.indexOf('compaction/start:null')
  246. const summary = log.indexOf('compaction/summary')
  247. const end = log.indexOf('compaction/end:null')
  248. const flush = log.indexOf('flush')
  249. const nextTurn = log.indexOf('turn/start')
  250. expect(start).toBeLessThan(summary)
  251. expect(summary).toBeLessThan(end)
  252. expect(end).toBeLessThan(flush)
  253. expect(flush).toBeLessThan(nextTurn)
  254. expect(adapter.requests).toHaveLength(2)
  255. const second = (adapter.requests[1] ?? []).map(message => message.content
  256. .map(block => block.type === 'text' ? block.text : '')
  257. .join(''))
  258. expect(second[0]).toContain('checkpoint')
  259. expect(second.at(-1)).toBe('after compaction')
  260. expect(second.some(text => text.includes(PROMPT))).toBe(false)
  261. })
  262. it('keeps context injected during summarization pending for the next step', async () => {
  263. const harness = await loopHarness()
  264. const { agent, compact } = harness
  265. await seedHistory(harness)
  266. compact.duringSummary = () => {
  267. agent.inject(createUserMessage({
  268. content: [{ type: 'text', text: 'INJECTED CONTEXT' }],
  269. source: { kind: 'plugin', plugin: 'test' },
  270. }))
  271. }
  272. const result = await compact.compactNow(agent, SIGNAL)
  273. expect(result).not.toBeNull()
  274. const start = agent.session.events.findLast(event => event.type === 'compaction/start')
  275. const injected = agent.inbox.nextStep.find(message =>
  276. message.source.kind === 'plugin' && message.source.plugin === 'test')
  277. const end = agent.session.events.findLast(event => event.type === 'compaction/end')
  278. expect(start).toBeDefined()
  279. expect(injected).toBeDefined()
  280. expect(end).toBeDefined()
  281. expect(agent.session.events.some(event => event.type === 'user/message'
  282. && event.data.id === injected?.id)).toBe(false)
  283. agent.followup(createUserMessage({
  284. content: [{ type: 'text', text: 'after compaction' }],
  285. source: { kind: 'user' },
  286. }))
  287. await agent.whenIdle()
  288. const messages = derivedText(agent.session)
  289. expect(messages[0]).toContain('checkpoint')
  290. expect(messages.filter(text => text.includes('INJECTED CONTEXT'))).toHaveLength(1)
  291. })
  292. it('keeps the marker order when listeners attempt a re-entrant injection', async () => {
  293. const harness = await loopHarness()
  294. const { ctx, agent, compact } = harness
  295. await seedHistory(harness)
  296. const attempts: string[] = []
  297. ctx.on('session/event', (_session, event) => {
  298. if (event.type !== 'compaction/start' && event.type !== 'compaction/summary') return
  299. attempts.push(event.type)
  300. agent.inject(createUserMessage({
  301. content: [{ type: 'text', text: `from ${event.type}` }],
  302. source: { kind: 'plugin', plugin: 'listener' },
  303. }))
  304. })
  305. const result = await compact.compactNow(agent, SIGNAL)
  306. expect(attempts).toEqual(['compaction/start', 'compaction/summary'])
  307. expect(result).not.toBeNull()
  308. expect(derivedText(agent.session)[0]).toContain('checkpoint')
  309. expect(agent.session.events.filter(event => event.type === 'user/message'
  310. && event.data.source.kind === 'plugin' && event.data.source.plugin === 'listener')).toHaveLength(0)
  311. const types = compactEvents(agent.session).map(event => event.type)
  312. expect(types).toEqual(['compaction/start', 'compaction/summary', 'compaction/end'])
  313. })
  314. it('reports busy without summarizing when a prompt already owns the next turn', async () => {
  315. const harness = await loopHarness()
  316. const { agent, compact, adapter } = harness
  317. await seedHistory(harness)
  318. agent.followup(createUserMessage({
  319. content: [{ type: 'text', text: 'first in line' }],
  320. source: { kind: 'user' },
  321. }))
  322. expect((await rejection(() => compact.compactNow(agent, SIGNAL))).code).toBe('busy')
  323. expect(compact.calls).toHaveLength(0)
  324. await agent.whenIdle()
  325. expect(adapter.requests).toHaveLength(2)
  326. expect(agent.session.events.some(event => event.type === 'compaction/start')).toBe(false)
  327. })
  328. it('releases turn admission after a summarizer failure and records the failed attempt', async () => {
  329. const harness = await loopHarness()
  330. const { agent, compact, adapter } = harness
  331. await seedHistory(harness)
  332. compact.error = new Error('summarizer unavailable')
  333. const before = [...agent.session.surface.nodes]
  334. expect((await rejection(compact.compactNow(agent, SIGNAL))).code).toBe('summary')
  335. expect(agent.session.surface.nodes).toEqual(before)
  336. const markers = compactEvents(agent.session)
  337. expect(markers.map(event => event.type)).toEqual(['compaction/start', 'compaction/end'])
  338. expect(markers[1]?.type === 'compaction/end' && markers[1].data.error)
  339. .toContain('summarizer unavailable')
  340. agent.followup(createUserMessage({
  341. content: [{ type: 'text', text: 'runs after the failure' }],
  342. source: { kind: 'user' },
  343. }))
  344. await agent.whenIdle()
  345. expect(adapter.requests).toHaveLength(2)
  346. })
  347. })
  348. describe('compactNow transaction and failure classification', () => {
  349. it('returns null without writing a bracket for history that cannot be compacted', async () => {
  350. const { compact } = detachedService()
  351. const session = Session.create(SessionId('empty'))
  352. let released = 0
  353. const agent = fakeAgent(session, () => () => { released += 1 })
  354. expect(await compact.compactNow(agent, SIGNAL)).toBeNull()
  355. expect(released).toBe(1)
  356. expect(compact.calls).toHaveLength(0)
  357. expect(compactEvents(session)).toEqual([])
  358. })
  359. it('commits a standalone bracket without consuming a turn number and checkpoints durability', async () => {
  360. const { compact, flushes } = detachedService()
  361. const session = closedConversation(2, 7)
  362. const agent = fakeAgent(session, () => () => undefined)
  363. const commandId = CommandId('manual-compact-command')
  364. const result = await compact.compactNow(agent, SIGNAL, commandId)
  365. expect(result).not.toBeNull()
  366. expect(result?.sourceCommandId).toBe(commandId)
  367. expect(flushes()).toBe(1)
  368. expect(session.events.filter(event => event.type === 'turn/start').at(-1)?.data.turn).toBe(7)
  369. const start = session.events.findLast(event => event.type === 'compaction/start')
  370. const summaryEvent = session.events.findLast(event => event.type === 'compaction/summary')
  371. const checkpoint = session.events.findLast(
  372. (event): event is SessionEvent<'user/message'> => event.type === 'user/message'
  373. && isCompactCheckpointSource(event.data.source),
  374. )
  375. const end = session.events.findLast(event => event.type === 'compaction/end')
  376. const correlated = { compactionId: result?.compactionId, sourceCommandId: commandId }
  377. expect(start?.data).toEqual({ ...correlated, turn: null })
  378. expect(summaryEvent?.data.sourceCommandId).toBe(commandId)
  379. expect(checkpoint?.data.source).toMatchObject(correlated)
  380. expect(end?.data).toEqual({ ...correlated, turn: null })
  381. })
  382. it('reports a live unmatched bracket as busy without summarizing', async () => {
  383. const { compact } = detachedService()
  384. const session = closedConversation(2)
  385. session.append('compaction/start', {
  386. compactionId: CompactionId('live-manual-compaction'),
  387. turn: null,
  388. })
  389. const agent = fakeAgent(session, () => () => undefined)
  390. const error = await rejection(() => compact.compactNow(agent, SIGNAL))
  391. expect(error.code).toBe('busy')
  392. expect(error.message).toContain('compaction lock is already active')
  393. expect(compact.calls).toHaveLength(0)
  394. })
  395. it('ignores an unmatched bracket inherited before a later end-seed marker', async () => {
  396. const { compact } = detachedService()
  397. const original = closedConversation(2)
  398. original.append('compaction/start', {
  399. compactionId: CompactionId('stale-manual-compaction'),
  400. turn: null,
  401. })
  402. const reloaded = Session.create(SessionId('stale-orphan'), [...original.events])
  403. const boundary = reloaded.events.findLast(event => event.type === 'session/end-seed')
  404. const orphan = reloaded.events.find(event => event.type === 'compaction/start')
  405. const agent = fakeAgent(reloaded, () => () => undefined)
  406. expect(boundary?.seq).toBeGreaterThan(orphan?.seq ?? Number.MAX_SAFE_INTEGER)
  407. await expect(compact.compactNow(agent, SIGNAL)).resolves.not.toBeNull()
  408. expect(compact.calls).toHaveLength(1)
  409. })
  410. it('scans a stale orphan independently of later repaired turn state', async () => {
  411. const { compact } = detachedService()
  412. const original = closedConversation(2)
  413. original.append('compaction/start', {
  414. compactionId: CompactionId('reloaded-manual-compaction'),
  415. turn: null,
  416. })
  417. original.append('turn/start', { turn: 3 })
  418. original.append('turn/end', { turn: 3, reason: { kind: 'interrupted' } })
  419. const reloaded = Session.create(SessionId('reloaded-orphan'), [...original.events])
  420. const agent = fakeAgent(reloaded, () => () => undefined)
  421. await expect(compact.compactNow(agent, SIGNAL)).resolves.not.toBeNull()
  422. expect(compact.calls).toHaveLength(1)
  423. })
  424. it('refuses an open turn in the log', async () => {
  425. const { compact } = detachedService()
  426. const session = closedConversation(2)
  427. session.append('turn/start', { turn: 3 })
  428. const agent = fakeAgent(session, () => () => undefined)
  429. const error = await rejection(compact.compactNow(agent, SIGNAL))
  430. expect(error.code).toBe('busy')
  431. expect(error.message).toContain('already has an open turn')
  432. })
  433. it('reports busy and skips summarization when admission is unavailable', async () => {
  434. const { compact } = detachedService()
  435. const agent = fakeAgent(closedConversation(2), () => undefined)
  436. expect((await rejection(() => compact.compactNow(agent, SIGNAL))).code).toBe('busy')
  437. expect(compact.calls).toHaveLength(0)
  438. })
  439. it('rejects a selected span replaced during summarization and records an error close', async () => {
  440. const { compact, flushes } = detachedService()
  441. const session = closedConversation(2)
  442. let released = 0
  443. const agent = fakeAgent(session, () => () => { released += 1 })
  444. compact.duringSummary = () => {
  445. const [head] = session.surface.nodes
  446. session.append('user/message', createUserMessage({
  447. content: [{ type: 'text', text: 'competing replacement' }],
  448. source: { kind: 'plugin', plugin: 'rival' },
  449. }), {
  450. surfaceOp: { op: 'replace', start: head!, end: head! },
  451. sourceEventSeqs: [head!],
  452. })
  453. }
  454. expect((await rejection(compact.compactNow(agent, SIGNAL))).code).toBe('changed')
  455. expect(released).toBe(1)
  456. expect(flushes()).toBe(1)
  457. expect(compactEvents(session).map(event => event.type)).toEqual(['compaction/start', 'compaction/end'])
  458. })
  459. it('rejects a selected span whose middle node was replaced during summarization', async () => {
  460. const { compact } = detachedService()
  461. const session = closedConversation(3)
  462. const agent = fakeAgent(session, () => () => undefined)
  463. compact.duringSummary = () => {
  464. const middle = session.surface.nodes[1]
  465. session.append('user/message', createUserMessage({
  466. content: [{ type: 'text', text: 'rewritten middle node' }],
  467. source: { kind: 'plugin', plugin: 'rival' },
  468. }), {
  469. surfaceOp: { op: 'replace', start: middle!, end: middle! },
  470. sourceEventSeqs: [middle!],
  471. })
  472. }
  473. const error = await rejection(compact.compactNow(agent, SIGNAL))
  474. expect(error.code).toBe('changed')
  475. expect(causeOf(error).message).toContain('span changed during summarization')
  476. })
  477. it('revalidates the selected span after the summarizer continuation settles', async () => {
  478. const { compact, flushes } = detachedService()
  479. const session = closedConversation(2)
  480. const gate = deferred()
  481. compact.gate = gate.promise
  482. let released = 0
  483. const agent = fakeAgent(session, () => () => { released += 1 })
  484. const head = session.surface.nodes[0]!
  485. const generation = session.surface.replaceGeneration
  486. const running = compact.compactNow(agent, SIGNAL)
  487. await Promise.resolve()
  488. expect(compact.calls).toHaveLength(1)
  489. gate.resolve()
  490. queueMicrotask(() => {
  491. queueMicrotask(() => {
  492. session.append('user/message', createUserMessage({
  493. content: [{ type: 'text', text: 'late competing replacement' }],
  494. source: { kind: 'plugin', plugin: 'rival' },
  495. }), {
  496. surfaceOp: { op: 'replace', start: head, end: head },
  497. sourceEventSeqs: [head],
  498. })
  499. })
  500. })
  501. const error = await rejection(running)
  502. expect(error.code).toBe('changed')
  503. expect(causeOf(error).message).toContain('selected span')
  504. expect(released).toBe(1)
  505. expect(flushes()).toBe(1)
  506. expect(session.surface.replaceGeneration).toBe(generation + 1)
  507. expect(session.surface.nodes).not.toContain(head)
  508. expect(compactEvents(session).map(event => event.type)).toEqual(['compaction/start', 'compaction/end'])
  509. expect(session.events.some(event => event.type === 'user/message'
  510. && isCompactCheckpointSource(event.data.source))).toBe(false)
  511. })
  512. it('classifies a failing compaction/end as commit failure and leaves one orphan', async () => {
  513. const { compact, flushes } = detachedService()
  514. const session = closedConversation(2)
  515. const agent = fakeAgent(session, () => () => undefined)
  516. const append = session.append.bind(session)
  517. vi.spyOn(session, 'append').mockImplementation(((type: string, ...rest: never[]) => {
  518. if (type === 'compaction/end') throw new Error('boundary rejected')
  519. return (append as (...args: never[]) => unknown)(type as never, ...rest)
  520. }) as never)
  521. const error = await rejection(compact.compactNow(agent, SIGNAL))
  522. expect(error.code).toBe('commit')
  523. expect(causeOf(error).message).toBe('boundary rejected')
  524. vi.restoreAllMocks()
  525. expect(flushes()).toBe(0)
  526. expect(session.events.findLast(event => event.type.startsWith('compaction/'))?.type)
  527. .toBe('compaction/summary')
  528. expect(compactEvents(session).filter(event => event.type === 'compaction/start')).toHaveLength(1)
  529. const calls = compact.calls.length
  530. expect((await rejection(compact.compactNow(agent, SIGNAL))).code).toBe('busy')
  531. expect(compact.calls).toHaveLength(calls)
  532. })
  533. it('keeps a failed error-close as the commit failure and does not flush', async () => {
  534. const { compact, flushes } = detachedService()
  535. const session = closedConversation(2)
  536. let released = 0
  537. const agent = fakeAgent(session, () => () => { released += 1 })
  538. compact.error = new Error('summary rejected')
  539. const append = session.append.bind(session)
  540. vi.spyOn(session, 'append').mockImplementation(((type: string, ...rest: never[]) => {
  541. if (type === 'compaction/end') throw new Error('error boundary rejected')
  542. return (append as (...args: never[]) => unknown)(type as never, ...rest)
  543. }) as never)
  544. const error = await rejection(compact.compactNow(agent, SIGNAL))
  545. vi.restoreAllMocks()
  546. expect(error.code).toBe('commit')
  547. expect(causeOf(error).message).toBe('error boundary rejected')
  548. expect(released).toBe(1)
  549. expect(flushes()).toBe(0)
  550. expect(compactEvents(session).map(event => event.type)).toEqual(['compaction/start'])
  551. })
  552. it('rejects a selected span whose pricing changed during summarization', async () => {
  553. const { ctx, compact } = detachedService()
  554. const session = closedConversation(2)
  555. const agent = fakeAgent(session, () => () => undefined)
  556. const meter = ctx.tokenMeter
  557. const original = meter.measure.bind(meter)
  558. compact.duringSummary = () => {
  559. vi.spyOn(meter, 'measure').mockImplementationOnce((target) => {
  560. const measurement = original(target)
  561. return {
  562. ...measurement,
  563. nodes: measurement.nodes.map((node, index) =>
  564. index === 0 ? { ...node, tokens: node.tokens + 1 } : node),
  565. }
  566. })
  567. }
  568. expect((await rejection(compact.compactNow(agent, SIGNAL))).code).toBe('changed')
  569. vi.restoreAllMocks()
  570. })
  571. it('classifies a commit-body failure and still releases admission', async () => {
  572. const { compact } = detachedService()
  573. const session = closedConversation(2)
  574. let released = 0
  575. const agent = fakeAgent(session, () => () => { released += 1 })
  576. const append = session.append.bind(session)
  577. vi.spyOn(session, 'append').mockImplementation(((type: string, ...rest: never[]) => {
  578. if (type === 'compaction/summary') throw new Error('summary record rejected')
  579. return (append as (...args: never[]) => unknown)(type as never, ...rest)
  580. }) as never)
  581. const error = await rejection(compact.compactNow(agent, SIGNAL))
  582. vi.restoreAllMocks()
  583. expect(error.code).toBe('commit')
  584. expect(released).toBe(1)
  585. const end = session.events.findLast(event => event.type === 'compaction/end')
  586. expect(end?.type === 'compaction/end' && end.data.error).toContain('summary record rejected')
  587. expect(end?.type === 'compaction/end' && end.data.turn).toBeNull()
  588. })
  589. it('keeps a commit failure when the durability checkpoint also fails', async () => {
  590. const { ctx, compact } = detachedService()
  591. const session = closedConversation(2)
  592. const agent = fakeAgent(session, () => () => undefined)
  593. const append = session.append.bind(session)
  594. vi.spyOn(session, 'append').mockImplementation(((type: string, ...rest: never[]) => {
  595. if (type === 'compaction/summary') throw new Error('summary record rejected')
  596. return (append as (...args: never[]) => unknown)(type as never, ...rest)
  597. }) as never)
  598. vi.spyOn(ctx.sessions, 'flush').mockRejectedValueOnce(new Error('disk full'))
  599. const error = await rejection(compact.compactNow(agent, SIGNAL))
  600. expect(error.code).toBe('commit')
  601. expect(causeOf(error).message).toBe('summary record rejected')
  602. vi.restoreAllMocks()
  603. })
  604. it('compacts a session with no durable turn boundary without creating one', async () => {
  605. const { compact } = detachedService()
  606. const session = Session.create(SessionId('turnless'))
  607. for (const text of [PROMPT, 'recent tail']) {
  608. session.append('user/message', createUserMessage({
  609. content: [{ type: 'text', text }],
  610. source: { kind: 'user' },
  611. }), { surfaceOp: 'append' })
  612. }
  613. const agent = fakeAgent(session, () => () => undefined)
  614. const result = await compact.compactNow(agent, SIGNAL)
  615. expect(result).not.toBeNull()
  616. expect(session.events.some(event => event.type === 'turn/start')).toBe(false)
  617. expect(session.events.find(event => event.type === 'compaction/start')?.data)
  618. .toEqual({ compactionId: result?.compactionId, turn: null })
  619. })
  620. it('classifies a durability failure after the standalone bracket committed', async () => {
  621. const { ctx, compact } = detachedService()
  622. const session = closedConversation(2)
  623. const agent = fakeAgent(session, () => () => undefined)
  624. vi.spyOn(ctx.sessions, 'flush').mockRejectedValueOnce(new Error('disk full'))
  625. expect((await rejection(compact.compactNow(agent, SIGNAL))).code).toBe('persistence')
  626. vi.restoreAllMocks()
  627. expect(session.events.some(event => event.type === 'compaction/summary')).toBe(true)
  628. const start = session.events.findLast(event => event.type === 'compaction/start')
  629. const end = session.events.findLast(event => event.type === 'compaction/end')
  630. expect(end?.data).toEqual({ compactionId: start?.data.compactionId, turn: null })
  631. })
  632. it('lets a pre-aborted signal win before reservation, measurement, or summarization', async () => {
  633. const cases = [
  634. { name: 'busy', session: closedConversation(2), release: undefined },
  635. { name: 'empty', session: Session.create(SessionId('pre-aborted-empty')), release: () => undefined },
  636. { name: 'compactable', session: closedConversation(2, 9), release: () => undefined },
  637. ] as const
  638. for (const testCase of cases) {
  639. const { ctx, compact } = detachedService()
  640. const reserve = vi.fn(() => testCase.release)
  641. const measure = vi.spyOn(ctx.tokenMeter, 'measure')
  642. const agent = fakeAgent(testCase.session, reserve)
  643. const before = [...testCase.session.events]
  644. const reason = Object.freeze({ kind: 'cancelled', case: testCase.name })
  645. const controller = new AbortController()
  646. controller.abort(reason)
  647. let thrown: unknown
  648. try {
  649. void compact.compactNow(agent, controller.signal)
  650. } catch (error: unknown) {
  651. thrown = error
  652. }
  653. expect(thrown).toBe(reason)
  654. expect(reserve).not.toHaveBeenCalled()
  655. expect(measure).not.toHaveBeenCalled()
  656. expect(compact.calls).toHaveLength(0)
  657. expect(testCase.session.events).toEqual(before)
  658. vi.restoreAllMocks()
  659. }
  660. })
  661. it('preserves the exact cancellation reason when the summarizer also rejects', async () => {
  662. const { compact, flushes } = detachedService()
  663. const controller = new AbortController()
  664. const reason = new Error('cancelled by the caller')
  665. let released = 0
  666. const session = closedConversation(2)
  667. const agent = fakeAgent(session, () => () => { released += 1 })
  668. compact.duringSummary = () => { controller.abort(reason) }
  669. compact.error = new Error('summarizer aborted')
  670. await expect(compact.compactNow(agent, controller.signal)).rejects.toBe(reason)
  671. expect(released).toBe(1)
  672. expect(flushes()).toBe(1)
  673. const events = compactEvents(session)
  674. expect(events.map(event => event.type)).toEqual(['compaction/start', 'compaction/end'])
  675. expect(events[1]?.type === 'compaction/end' && events[1].data.error)
  676. .toContain('summarizer aborted')
  677. })
  678. it('classifies agent cancellation during maintenance as an expected cancellation', async () => {
  679. const { compact } = detachedService()
  680. const controller = new AbortController()
  681. const reason = new Error('agent cancelled maintenance')
  682. const session = closedConversation(2)
  683. const agent = fakeAgent(session, () => () => undefined, controller.signal)
  684. compact.duringSummary = () => { controller.abort(reason) }
  685. compact.error = new Error('summarizer observed cancellation')
  686. const error = await rejection(compact.compactNow(agent, SIGNAL))
  687. expect(error.code).toBe('cancelled')
  688. expect(error.cause).toBe(reason)
  689. })
  690. it('aborts before committing when cancellation lands after summarization', async () => {
  691. const { compact } = detachedService()
  692. const controller = new AbortController()
  693. const reason = new Error('cancelled by the caller')
  694. const session = closedConversation(2)
  695. const agent = fakeAgent(session, () => () => undefined)
  696. compact.duringSummary = () => { controller.abort(reason) }
  697. await expect(compact.compactNow(agent, controller.signal)).rejects.toBe(reason)
  698. expect(compactEvents(session).map(event => event.type)).toEqual(['compaction/start', 'compaction/end'])
  699. expect(session.events.some(event => event.type === 'compaction/summary')).toBe(false)
  700. })
  701. it('waits for the durability checkpoint before cancellation wins and admission releases', async () => {
  702. const { ctx, compact } = detachedService()
  703. const controller = new AbortController()
  704. const reason = new Error('cancelled during flush')
  705. const flushGate = Promise.withResolvers<boolean>()
  706. const flush = vi.spyOn(ctx.sessions, 'flush').mockReturnValueOnce(flushGate.promise)
  707. const session = closedConversation(2)
  708. let released = 0
  709. const agent = fakeAgent(session, () => () => { released += 1 })
  710. const running = compact.compactNow(agent, controller.signal)
  711. let settled = false
  712. void running.then(
  713. () => { settled = true },
  714. () => { settled = true },
  715. )
  716. await vi.waitFor(() => {
  717. expect(flush).toHaveBeenCalledWith(session)
  718. })
  719. controller.abort(reason)
  720. await Promise.resolve()
  721. expect(settled).toBe(false)
  722. expect(released).toBe(0)
  723. flushGate.resolve(false)
  724. await expect(running).rejects.toBe(reason)
  725. expect(released).toBe(1)
  726. })
  727. it('preserves raw output and usage in the manual summary event', async () => {
  728. const { compact } = detachedService()
  729. const session = closedConversation(2)
  730. const agent = fakeAgent(session, () => () => undefined)
  731. compact.rawOutput = [
  732. { type: 'text', text: 'checkpoint' },
  733. { type: 'reasoning', text: 'hidden reasoning' },
  734. ]
  735. compact.usage = { inputTokens: 40, outputTokens: 5 }
  736. await compact.compactNow(agent, SIGNAL)
  737. const summary = session.events.find(event => event.type === 'compaction/summary')
  738. expect(summary?.type === 'compaction/summary' && summary.data.rawOutput).toEqual(compact.rawOutput)
  739. expect(summary?.type === 'compaction/summary' && summary.data.usage).toEqual(compact.usage)
  740. })
  741. it('makes duration derivable from the opening and closing marker times', async () => {
  742. const { compact } = detachedService()
  743. const session = closedConversation(2)
  744. const agent = fakeAgent(session, () => () => undefined)
  745. compact.gate = new Promise<undefined>((resolve) => {
  746. setTimeout(() => { resolve(undefined) }, 5)
  747. })
  748. await compact.compactNow(agent, SIGNAL)
  749. const start = session.events.findLast(event => event.type === 'compaction/start')
  750. const end = session.events.findLast(event => event.type === 'compaction/end')
  751. expect(start).toBeDefined()
  752. expect(end).toBeDefined()
  753. expect(end!.time - start!.time).toBeGreaterThan(0)
  754. })
  755. it('excludes concurrent automatic and manual compaction of one session', async () => {
  756. const { compact } = detachedService()
  757. const session = closedConversation(3)
  758. const agent = fakeAgent(session, () => () => undefined)
  759. const gate = deferred()
  760. compact.gate = gate.promise
  761. const manual = compact.compactNow(agent, SIGNAL)
  762. await Promise.resolve()
  763. const nodes = session.surface.nodes
  764. await expect(compact.compactRegion(
  765. nodes[0]!,
  766. nodes[1]!,
  767. agent,
  768. )).rejects.toThrow('compaction lock is already active')
  769. gate.resolve()
  770. compact.gate = undefined
  771. const result: CompactionResult | null = await manual
  772. expect(result).not.toBeNull()
  773. })
  774. it('excludes a manual request while an explicit region compaction runs', async () => {
  775. const { compact } = detachedService()
  776. const session = closedConversation(3)
  777. session.append('turn/start', { turn: 4 })
  778. const agent = fakeAgent(session, () => () => undefined)
  779. const gate = deferred()
  780. compact.gate = gate.promise
  781. const nodes = session.surface.nodes
  782. const region = compact.compactRegion(nodes[0]!, nodes[1]!, agent)
  783. await Promise.resolve()
  784. expect((await rejection(compact.compactNow(agent, SIGNAL))).code).toBe('busy')
  785. gate.resolve()
  786. compact.gate = undefined
  787. await expect(region).resolves.toMatchObject({ shadowedSeqs: nodes.slice(0, 2) })
  788. })
  789. })