fix(host): classify a raced cold-resume ID collision as agent-busy

When a generic `agentFor` cold resume loses the identity to a parent's
concurrent `enter()` — the collision rejection arrives from
`ctx.agents.resume` publication after the pre-resume re-check — the error
fell through to the `internal` mapping. Clients retrying then see a
transient-looking internal failure instead of the stable ownership error
that `ensureSession`'s `.catch` already produces for the exact same
published-winner case.

Mirror that re-classification in `agentFor`'s resume error path: after the
typed errors, re-check the registry and attached store and answer
`agent-busy` when the raced winner is subagent-owned. Adds a regression
test whose resume mock publishes the subagent winner before throwing the
ID-collision error.
This commit is contained in:
Tianyi Cui
2026-08-02 12:22:28 +08:00
parent 56e252bed3
commit 468fd29e51
2 changed files with 44 additions and 3 deletions

View File

@@ -2253,9 +2253,11 @@ export function createApiProxy(ctx: Context, defaults: ApiProxyDefaults): ApiPro
},
commands: {
// Both methods address one session's agent (agentFor keeps its
// resume-on-miss: clients only send a sessionId for a published
// session, and resume restores an existing entity).
// Both methods address one session's agent. agentFor resumes on miss
// and fences every subagent-owned identity with `agent-busy`; the
// api/commands.ts module contract owns that fence's wording, so this
// comment only notes the routing shape: clients send a sessionId for a
// published session, and resume restores an existing entity.
async list(request) {
// Missing service = the deployment omitted dsh-commands from its
// composition, not an empty catalog: fail loud instead of serving [].

View File

@@ -338,4 +338,43 @@ describe('sessions.prompt synchronous rejection', () => {
}
}
})
it('classifies a raced cold-resume ID collision as agent-busy', async () => {
const ctx = new Context()
await ctx.plugin(SessionStore)
await ctx.plugin(AgentRegistry)
await ctx.plugin(UserInteractionService)
const sessionId = sid('race-resume')
const meta: SessionHeader = header('race-resume', 1000)
ctx.provide('sessionPersistence', {
list: () => Promise.resolve([meta]),
inspect: () => Promise.resolve({ meta, events: [] as SessionEvent[] }),
locate: () => undefined,
} as never)
// The raced winner: a live parent-owned subagent publishes the identity
// while the generic cold resume is in flight, so the resume collides.
const parentSession = ctx.sessions.create(sid('race-parent'), { meta: { cwd: '/proj' } })
const parent = { id: parentSession.id, session: parentSession, status: 'idle', ctx } as Agent
ctx.agents.register(parent)
const childSession = ctx.sessions.create(sessionId, {
meta: { cwd: '/proj', parentSession: parent.id, origin: 'subagent' },
})
const child = { id: sessionId, session: childSession, status: 'idle', ctx } as unknown as Agent
vi.spyOn(ctx.agents, 'resume').mockImplementationOnce(async () => {
// The parent's `enter()` wins the identity between the pre-resume
// re-check and publication; the generic resume then collides.
ctx.agents.register(child)
throw new Error('session id already published')
})
const api = createApiProxy(ctx, { provider: 'p', model: 'm', cwd: '/tmp', workspaceRoot: '/tmp' })
const models = await api.sessions.models(request({ sessionId }))
expect(models.result.ok).toBe(false)
if (!models.result.ok) {
expect(models.result.error).toMatchObject({
code: 'agent-busy',
details: { reason: 'use subagent delivery for this child session' },
})
}
})
})