fix(team-mode): close peer message delivery races

This commit is contained in:
YeonGyu-Kim
2026-05-19 18:23:57 +09:00
parent bcea4a9d28
commit b2918fd4db
13 changed files with 492 additions and 47 deletions
@@ -711,6 +711,46 @@ describe("dispatchInternalPrompt shared gate behavior", () => {
expect(promptCalls).toBe(0)
})
test("#given latest assistant turn has unknown finish #when an internal promptAsync is requested #then no prompt is sent", async () => {
// given
let promptCalls = 0
const client = {
session: {
status: async () => ({ data: { ses_unknown_finish: { type: "idle" } } }),
messages: async () => ({
data: [
{
info: { id: "msg_user", role: "user" },
parts: [{ type: "text", text: "run work" }],
},
{
info: { id: "msg_assistant", role: "assistant", finish: "unknown" },
parts: [{ type: "reasoning", text: "still resolving" }],
},
],
}),
promptAsync: async () => {
promptCalls += 1
},
},
}
// when
const result = await dispatchInternalPrompt({
mode: "async",
client,
sessionID: "ses_unknown_finish",
input: { path: { id: "ses_unknown_finish" }, body: { parts: [] } },
source: "test:unknown-finish",
settleMs: 0,
postDispatchHoldMs: 0,
})
// then
expect(result.status).toBe("queued")
expect(promptCalls).toBe(0)
})
test("#given internal user tail follows an assistant waiting on tools #when an internal promptAsync is requested #then no prompt is sent", async () => {
// given
let promptCalls = 0
@@ -567,6 +567,48 @@ describe("createTeamIdleWakeHint", () => {
expect(processedEntries).toContain(`${messageId}.json`)
})
test("#given stale idle while pending live delivery is still busy #when idle wake runs #then it keeps the reservation pending", async () => {
// given
const baseDir = await createTemporaryBaseDir()
const config = createConfig(baseDir)
const teamRunId = randomUUID()
const messageId = randomUUID()
await seedRuntimeState(createRuntimeState(teamRunId, [messageId]), config)
await seedReservedUnreadMessage(teamRunId, config, messageId, "live delivery body", 100)
const ackSpy = spyOn(ackModule, "ackMessages")
const promptAsyncSpy = mock(async (_input: WakeHintPromptInput) => ({}))
const handler = createTeamIdleWakeHint({
directory: "/tmp/project",
client: {
session: {
promptAsync: promptAsyncSpy,
status: async () => ({ data: { "member-session": { type: "busy" } } }),
},
},
}, config, { idleSettleMs: 0 })
// when
await handler({
event: {
type: "session.idle",
properties: { sessionID: "member-session" },
},
})
// then
expect(ackSpy).not.toHaveBeenCalled()
expect(promptAsyncSpy).not.toHaveBeenCalled()
const runtimeState = await loadRuntimeState(teamRunId, config)
expect(runtimeState.members[0]?.pendingInjectedMessageIds).toEqual([messageId])
const inboxDir = getInboxDir(resolveBaseDir(config), teamRunId, "worker")
const inboxEntries = await readdir(inboxDir)
expect(inboxEntries).toContain(`.delivering-${messageId}.json`)
expect(inboxEntries).not.toContain(`${messageId}.json`)
})
test("#given a pending live-delivery ack and later unread message #when member idles after the live reply #then it wakes the member for the unread message", async () => {
// given
const baseDir = await createTemporaryBaseDir()
@@ -10,6 +10,7 @@ import { loadRuntimeState, transitionRuntimeState } from "../../features/team-mo
import { resolveSessionEventID } from "../../shared/event-session-id"
import { isAmbiguousPromptDispatchFailure } from "../../shared/prompt-failure-classifier"
import { log } from "../../shared/logger"
import { isSessionActive, settleAfterSessionIdle } from "../../shared/session-idle-settle"
import { dispatchInternalPrompt, isInternalPromptDispatchAccepted } from "../shared/prompt-async-gate"
type PromptAsyncInput = {
@@ -73,6 +74,20 @@ export function createTeamIdleWakeHint(ctx: TeamIdleWakeHintContext, config: Tea
const pendingInjectedMessageIds = [...memberEntry.pendingInjectedMessageIds]
if (pendingInjectedMessageIds.length > 0) {
if (typeof ctx.client.session.status === "function") {
await settleAfterSessionIdle(options?.idleSettleMs ?? 0)
if (await isSessionActive(ctx.client, sessionID)) {
log("team idle pending ack skipped while session remains active", {
event: "team-mode-idle-pending-ack-active",
teamRunId: runtimeState.teamRunId,
memberName: memberEntry.name,
sessionID,
pendingCount: pendingInjectedMessageIds.length,
})
return
}
}
await ackMessages(runtimeState.teamRunId, memberEntry.name, pendingInjectedMessageIds, config)
await transitionRuntimeState(runtimeState.teamRunId, (currentRuntimeState) => ({
...currentRuntimeState,
@@ -226,4 +226,89 @@ describe("createTeamMemberErrorHandler", () => {
expect(inboxEntries).not.toContain(`.delivering-${messageId}.json`)
expect(inboxEntries).not.toContain("processed")
})
test("#given session.error arrives while OpenCode still reports busy #when pending live delivery exists #then it does not requeue a duplicate peer message", async () => {
// given
const baseDir = await createTemporaryBaseDir()
const config = createConfig(baseDir)
const teamRunId = randomUUID()
const messageId = randomUUID()
await seedRuntimeState(createRuntimeStateWithPendingMessage(teamRunId, messageId), config)
await seedReservedMessage(teamRunId, config, messageId)
const handler = createTeamMemberErrorHandler(config, {
settleMs: 0,
client: {
session: {
status: async () => ({ data: { "member-session": { type: "busy" } } }),
},
},
})
// when
await handler({
event: {
type: "session.error",
properties: { sessionID: "member-session", error: new Error("transient provider error") },
},
})
// then
const runtimeState = await loadRuntimeState(teamRunId, config)
expect(runtimeState.members[0]?.status).toBe("running")
expect(runtimeState.members[0]?.pendingInjectedMessageIds).toEqual([messageId])
const inboxEntries = await readdir(getInboxDir(resolveBaseDir(config), teamRunId, "worker"))
expect(inboxEntries).toContain(`.delivering-${messageId}.json`)
expect(inboxEntries).not.toContain(`${messageId}.json`)
expect(inboxEntries).not.toContain("processed")
})
test("#given session.error arrives after peer message reached history #when pending live delivery exists #then it does not requeue a duplicate peer message", async () => {
// given
const baseDir = await createTemporaryBaseDir()
const config = createConfig(baseDir)
const teamRunId = randomUUID()
const messageId = randomUUID()
await seedRuntimeState(createRuntimeStateWithPendingMessage(teamRunId, messageId), config)
await seedReservedMessage(teamRunId, config, messageId)
const handler = createTeamMemberErrorHandler(config, {
settleMs: 0,
client: {
session: {
status: async () => ({ data: { "member-session": { type: "idle" } } }),
messages: async () => ({
data: [
{
info: { role: "user" },
parts: [
{
type: "text",
text: `<peer_message from="lead" messageId="${messageId}" kind="message">pending live delivery</peer_message>`,
},
],
},
],
}),
},
},
})
// when
await handler({
event: {
type: "session.error",
properties: { sessionID: "member-session", error: new Error("late session.error after accepted prompt") },
},
})
// then
const runtimeState = await loadRuntimeState(teamRunId, config)
expect(runtimeState.members[0]?.status).toBe("running")
expect(runtimeState.members[0]?.pendingInjectedMessageIds).toEqual([messageId])
const inboxEntries = await readdir(getInboxDir(resolveBaseDir(config), teamRunId, "worker"))
expect(inboxEntries).toContain(`.delivering-${messageId}.json`)
expect(inboxEntries).not.toContain(`${messageId}.json`)
expect(inboxEntries).not.toContain("processed")
})
})
@@ -7,9 +7,23 @@ import {
import { loadRuntimeState, transitionRuntimeState } from "../../features/team-mode/team-state-store/store"
import { resolveSessionEventID } from "../../shared/event-session-id"
import { log } from "../../shared/logger"
import {
DEFAULT_SESSION_IDLE_SETTLE_MS,
isSessionActive,
settleAfterSessionIdle,
} from "../../shared/session-idle-settle"
type HookInput = { event: { type: string; properties?: unknown } }
export type HookImpl = (input: HookInput) => Promise<void>
type TeamMemberErrorHandlerDeps = {
client?: {
session?: {
status?: () => Promise<unknown>
messages?: (input: { path: { id: string } }) => Promise<unknown>
}
}
settleMs?: number
}
function getErroredSessionID(properties: unknown): string | undefined {
return resolveSessionEventID(properties)
@@ -31,7 +45,73 @@ async function requeuePendingLiveDeliveries(
}
}
export function createTeamMemberErrorHandler(config: TeamModeConfig): HookImpl {
async function shouldKeepPendingLiveDeliveries(
deps: TeamMemberErrorHandlerDeps,
sessionID: string,
): Promise<boolean> {
if (typeof deps.client?.session?.status !== "function") {
return false
}
await settleAfterSessionIdle(deps.settleMs ?? DEFAULT_SESSION_IDLE_SETTLE_MS)
return await isSessionActive(deps.client, sessionID)
}
function isRecord(value: unknown): value is Record<string, unknown> {
return typeof value === "object" && value !== null
}
function getMessagesData(response: unknown): unknown[] {
if (isRecord(response) && Array.isArray(response.data)) {
return response.data
}
return Array.isArray(response) ? response : []
}
function valueContainsAnyMessageId(value: unknown, messageIds: ReadonlySet<string>): boolean {
if (typeof value === "string") {
return [...messageIds].some((messageId) => value.includes(messageId))
}
if (Array.isArray(value)) {
return value.some((entry) => valueContainsAnyMessageId(entry, messageIds))
}
if (isRecord(value)) {
return Object.values(value).some((entry) => valueContainsAnyMessageId(entry, messageIds))
}
return false
}
async function sessionHistoryContainsPendingMessage(
deps: TeamMemberErrorHandlerDeps,
sessionID: string,
messageIds: readonly string[],
): Promise<boolean> {
if (messageIds.length === 0 || typeof deps.client?.session?.messages !== "function") {
return false
}
try {
const response = await deps.client.session.messages({ path: { id: sessionID } })
const pendingMessageIds = new Set(messageIds)
return getMessagesData(response).some((message) => valueContainsAnyMessageId(message, pendingMessageIds))
} catch (error) {
log("team member session history check failed", {
event: "team-mode-member-error-history-check-failed",
sessionID,
error: error instanceof Error ? error.message : String(error),
})
return false
}
}
export function createTeamMemberErrorHandler(
config: TeamModeConfig,
deps: TeamMemberErrorHandlerDeps = {},
): HookImpl {
return async ({ event }: HookInput): Promise<void> => {
if (event.type !== "session.error") return
@@ -47,6 +127,29 @@ export function createTeamMemberErrorHandler(config: TeamModeConfig): HookImpl {
const runtimeState = await loadRuntimeState(runtimeMember.teamRunId, config)
const memberEntry = runtimeState.members.find((member) => member.name === runtimeMember.memberName)
const pendingInjectedMessageIds = memberEntry?.pendingInjectedMessageIds ?? []
if (await shouldKeepPendingLiveDeliveries(deps, erroredSessionID)) {
log("team member session error ignored while session remains active", {
event: "team-mode-member-error-active",
teamRunId: runtimeState.teamRunId,
teamName: runtimeState.teamName,
memberName: runtimeMember.memberName,
sessionID: erroredSessionID,
pendingCount: pendingInjectedMessageIds.length,
})
return
}
if (await sessionHistoryContainsPendingMessage(deps, erroredSessionID, pendingInjectedMessageIds)) {
log("team member session error ignored after pending peer message reached history", {
event: "team-mode-member-error-peer-message-accepted",
teamRunId: runtimeState.teamRunId,
teamName: runtimeState.teamName,
memberName: runtimeMember.memberName,
sessionID: erroredSessionID,
pendingCount: pendingInjectedMessageIds.length,
})
return
}
await requeuePendingLiveDeliveries(
runtimeState.teamRunId,
runtimeMember.memberName,