Merge pull request #4228 from code-yeongyu/fix/delegate-stale-activity

Fix delegate stale timeout activity checks
This commit is contained in:
YeonGyu-Kim
2026-05-21 12:59:11 +09:00
committed by GitHub
9 changed files with 491 additions and 16 deletions
@@ -0,0 +1,154 @@
import { afterEach, describe, expect, mock, spyOn, test } from "bun:test"
import { tmpdir } from "node:os"
import type { PluginInput } from "@opencode-ai/plugin"
import { unsafeTestValue } from "../../../test-support/unsafe-test-value"
import { BackgroundManager } from "./manager"
import type { BackgroundTask } from "./types"
type PollingManager = {
readonly pollRunningTasks: () => Promise<void>
readonly tasks: Map<string, BackgroundTask>
}
function createPluginContext(client: unknown): PluginInput {
const directory = tmpdir()
return unsafeTestValue<PluginInput>({
project: {
id: "test-project",
worktree: directory,
time: { created: Date.now() },
},
directory,
worktree: directory,
serverUrl: new URL("http://localhost:4096"),
$: {},
client,
})
}
function createRunningTask(overrides: Partial<BackgroundTask> = {}): BackgroundTask {
return {
id: "bg_test_session_activity",
sessionId: "ses-active",
parentSessionId: "parent-session",
parentMessageId: "parent-message",
description: "test task",
prompt: "test prompt",
agent: "explore",
status: "running",
startedAt: new Date(Date.now() - 120_000),
progress: { toolCalls: 0, lastUpdate: new Date() },
...overrides,
}
}
describe("BackgroundManager persisted session activity stale checks", () => {
const originalDateNow = Date.now
const fixedTime = new Date("2026-05-21T03:00:00.000Z").getTime()
afterEach(() => {
Date.now = originalDateNow
mock.restore()
})
test("keeps a busy task running when session.get reports fresh activity", async () => {
//#given - live event progress is stale, but OpenCode session metadata was updated recently
spyOn(globalThis.Date, "now").mockReturnValue(fixedTime)
let abortCallCount = 0
const sessionGet = mock(async () => ({
data: {
id: "ses-active",
time: { updated: fixedTime - 10_000 },
},
}))
const client = {
session: {
status: async () => ({ data: { "ses-active": { type: "busy" } } }),
get: sessionGet,
prompt: async () => ({}),
promptAsync: async () => ({}),
abort: async () => {
abortCallCount += 1
return {}
},
todo: async () => ({ data: [] }),
messages: async () => ({ data: [] }),
},
}
const manager = new BackgroundManager({
pluginContext: createPluginContext(client),
config: { staleTimeoutMs: 180_000 },
enableParentSessionNotifications: false,
})
const task = createRunningTask({
startedAt: new Date(Date.now() - 45 * 60 * 1000),
progress: {
toolCalls: 3,
lastUpdate: new Date(Date.now() - 45 * 60 * 1000),
},
})
const pollingManager = unsafeTestValue<PollingManager>(manager)
pollingManager.tasks.set(task.id, task)
//#when - polling reaches stale confirmation for the active child session
await pollingManager.pollRunningTasks()
//#then - persisted activity refreshes the task instead of aborting it
expect(task.status).toBe("running")
expect(task.error).toBeUndefined()
expect(task.progress?.lastUpdate).toEqual(new Date(fixedTime - 10_000))
expect(abortCallCount).toBe(0)
expect(sessionGet).toHaveBeenCalledTimes(1)
await manager.shutdown()
})
test("keeps a busy task running when session.get returns an error response", async () => {
//#given - live event progress is stale and OpenCode session lookup fails without throwing
spyOn(globalThis.Date, "now").mockReturnValue(fixedTime)
let abortCallCount = 0
const sessionGet = mock(async () => ({
error: "lookup failed",
data: undefined,
}))
const client = {
session: {
status: async () => ({ data: { "ses-active": { type: "busy" } } }),
get: sessionGet,
prompt: async () => ({}),
promptAsync: async () => ({}),
abort: async () => {
abortCallCount += 1
return {}
},
todo: async () => ({ data: [] }),
messages: async () => ({ data: [] }),
},
}
const manager = new BackgroundManager({
pluginContext: createPluginContext(client),
config: { staleTimeoutMs: 180_000 },
enableParentSessionNotifications: false,
})
const task = createRunningTask({
startedAt: new Date(Date.now() - 45 * 60 * 1000),
progress: {
toolCalls: 3,
lastUpdate: new Date(Date.now() - 45 * 60 * 1000),
},
})
const pollingManager = unsafeTestValue<PollingManager>(manager)
pollingManager.tasks.set(task.id, task)
//#when - polling tries to confirm stale activity through the SDK response
await pollingManager.pollRunningTasks()
//#then - the lookup failure defers cancellation for the active child session
expect(task.status).toBe("running")
expect(task.error).toBeUndefined()
expect(abortCallCount).toBe(0)
expect(sessionGet).toHaveBeenCalledTimes(1)
await manager.shutdown()
})
})
@@ -0,0 +1,56 @@
import { isRecord, log } from "../../shared"
import type { OpencodeClient } from "./opencode-client"
export type SessionActivityLookup =
| { readonly type: "activity"; readonly activity: Date }
| { readonly type: "missing" }
| { readonly type: "unavailable" }
export type SessionActivityResolver = (sessionID: string) => Promise<SessionActivityLookup>
function dateFromMillis(value: unknown): Date | undefined {
if (typeof value !== "number") return undefined
if (!Number.isFinite(value) || value < 0) return undefined
return new Date(value)
}
export function extractSessionActivityDate(sessionInfo: unknown): Date | undefined {
if (!isRecord(sessionInfo)) return undefined
const time = isRecord(sessionInfo.time) ? sessionInfo.time : undefined
return dateFromMillis(time?.updated) ?? dateFromMillis(sessionInfo.time_updated)
}
function sessionActivityLookupFromInfo(sessionInfo: unknown): SessionActivityLookup {
const activity = extractSessionActivityDate(sessionInfo)
return activity ? { type: "activity", activity } : { type: "missing" }
}
export async function getSessionActivityFromClient(
client: OpencodeClient,
sessionID: string,
directory?: string,
): Promise<SessionActivityLookup> {
const sessionGet = client.session.get
if (typeof sessionGet !== "function") return { type: "missing" }
try {
const response = await sessionGet({
path: { id: sessionID },
...(directory ? { query: { directory } } : {}),
})
if (isRecord(response) && response.error !== undefined && response.error !== null) {
log("[background-agent] Failed to read session activity:", { sessionID, error: response.error })
return { type: "unavailable" }
}
const sessionInfo = isRecord(response) && "data" in response ? response.data : response
return sessionActivityLookupFromInfo(sessionInfo)
} catch (error) {
if (error instanceof Error) {
log("[background-agent] Failed to read session activity:", { sessionID, error: error.message })
return { type: "unavailable" }
}
log("[background-agent] Failed to read session activity:", { sessionID, error })
return { type: "unavailable" }
}
}
@@ -0,0 +1,49 @@
import { log } from "../../shared"
import type { SessionActivityLookup, SessionActivityResolver } from "./session-activity"
import type { BackgroundTask } from "./types"
export type TaskActivityRefreshResult =
| { readonly type: "activity"; readonly activityTime: number }
| { readonly type: "missing" }
| { readonly type: "unavailable" }
function updateTaskActivityFromLookup(
task: BackgroundTask,
lookup: SessionActivityLookup,
): TaskActivityRefreshResult {
if (lookup.type !== "activity") return lookup
const activityTime = lookup.activity.getTime()
if (!Number.isFinite(activityTime)) return { type: "missing" }
const baseline = task.progress?.lastUpdate.getTime() ?? task.startedAt?.getTime()
if (baseline !== undefined && activityTime <= baseline) return { type: "activity", activityTime }
if (!task.progress) {
task.progress = { toolCalls: 0, lastUpdate: new Date(activityTime) }
} else {
task.progress.lastUpdate = new Date(activityTime)
}
return { type: "activity", activityTime }
}
export async function refreshTaskActivityFromSession(
task: BackgroundTask,
getSessionActivity: SessionActivityResolver,
): Promise<TaskActivityRefreshResult> {
if (!task.sessionId) return { type: "missing" }
let lookup: SessionActivityLookup
try {
lookup = await getSessionActivity(task.sessionId)
} catch (error) {
if (error instanceof Error) {
log("[background-agent] Error refreshing task session activity:", { taskId: task.id, error: error.message })
return { type: "unavailable" }
}
log("[background-agent] Error refreshing task session activity:", { taskId: task.id, error })
return { type: "unavailable" }
}
return updateTaskActivityFromLookup(task, lookup)
}
@@ -0,0 +1,160 @@
import { afterEach, describe, expect, mock, spyOn, test } from "bun:test"
import { unsafeTestValue } from "../../../test-support/unsafe-test-value"
import { checkAndInterruptStaleTasks } from "./task-poller"
import type { BackgroundTask } from "./types"
function createRunningTask(overrides: Partial<BackgroundTask> = {}): BackgroundTask {
return {
id: "task-1",
sessionId: "ses-1",
parentSessionId: "parent-ses-1",
parentMessageId: "msg-1",
description: "test",
prompt: "test",
agent: "explore",
status: "running",
startedAt: new Date(Date.now() - 120_000),
...overrides,
}
}
describe("checkAndInterruptStaleTasks persisted session activity", () => {
const mockClient = unsafeTestValue<Parameters<typeof checkAndInterruptStaleTasks>[0]["client"]>({
session: {
abort: mock(() => Promise.resolve()),
get: mock(() => Promise.resolve({ data: { id: "ses-1" } })),
},
})
const mockConcurrencyManager = unsafeTestValue<Parameters<typeof checkAndInterruptStaleTasks>[0]["concurrencyManager"]>({
release: mock(() => {}),
})
const mockNotify = mock(() => Promise.resolve())
const originalDateNow = Date.now
const fixedTime = new Date("2026-05-21T03:00:00.000Z").getTime()
afterEach(() => {
Date.now = originalDateNow
mockClient.session.abort.mockClear()
mockConcurrencyManager.release.mockClear()
mockNotify.mockClear()
})
test("keeps a busy task running when persisted session activity is fresh", async () => {
//#given - in-memory progress is stale, but OpenCode storage shows recent child-session activity
spyOn(globalThis.Date, "now").mockReturnValue(fixedTime)
const staleActivity = new Date(Date.now() - 45 * 60 * 1000)
const freshActivity = new Date(Date.now() - 10_000)
const task = createRunningTask({
startedAt: staleActivity,
progress: {
toolCalls: 2,
lastUpdate: staleActivity,
},
})
//#when - stale checking can refresh activity from the persisted session timestamp
await checkAndInterruptStaleTasks({
tasks: [task],
client: mockClient,
config: { staleTimeoutMs: 180_000 },
concurrencyManager: mockConcurrencyManager,
notifyParentSession: mockNotify,
sessionStatuses: { "ses-1": { type: "busy" } },
getSessionActivity: async () => ({ type: "activity", activity: freshActivity }),
})
//#then - task stays running and future stale checks use the persisted activity timestamp
expect(task.status).toBe("running")
expect(task.progress?.lastUpdate).toEqual(freshActivity)
expect(mockNotify).not.toHaveBeenCalled()
})
test("keeps a busy task with no local progress running when persisted session activity is fresh", async () => {
//#given - no progress event reached the manager, but OpenCode storage shows recent activity
spyOn(globalThis.Date, "now").mockReturnValue(fixedTime)
const staleStart = new Date(Date.now() - 15 * 60 * 1000)
const freshActivity = new Date(Date.now() - 10_000)
const task = createRunningTask({
startedAt: staleStart,
progress: undefined,
})
//#when - message staleness confirmation refreshes from persisted session metadata
await checkAndInterruptStaleTasks({
tasks: [task],
client: mockClient,
config: { messageStalenessTimeoutMs: 180_000 },
concurrencyManager: mockConcurrencyManager,
notifyParentSession: mockNotify,
sessionStatuses: { "ses-1": { type: "busy" } },
getSessionActivity: async () => ({ type: "activity", activity: freshActivity }),
})
//#then - the task stays running and receives a progress timestamp for future stale checks
expect(task.status).toBe("running")
expect(task.progress?.lastUpdate).toEqual(freshActivity)
expect(mockNotify).not.toHaveBeenCalled()
})
test("cancels a busy task when persisted session activity is also stale", async () => {
//#given - local progress is older than persisted activity, but both are outside the stale window
spyOn(globalThis.Date, "now").mockReturnValue(fixedTime)
const localActivity = new Date(Date.now() - 45 * 60 * 1000)
const stalePersistedActivity = new Date(Date.now() - 10 * 60 * 1000)
const task = createRunningTask({
startedAt: localActivity,
progress: {
toolCalls: 2,
lastUpdate: localActivity,
},
})
//#when - the persisted timestamp confirms the child session is truly stale
await checkAndInterruptStaleTasks({
tasks: [task],
client: mockClient,
config: { staleTimeoutMs: 180_000 },
concurrencyManager: mockConcurrencyManager,
notifyParentSession: mockNotify,
sessionStatuses: { "ses-1": { type: "busy" } },
getSessionActivity: async () => ({ type: "activity", activity: stalePersistedActivity }),
})
//#then - cancellation still happens, but the stale age reflects persisted activity
expect(task.status).toBe("cancelled")
expect(task.error).toContain("Stale timeout")
expect(task.error).toContain("10min")
expect(task.progress?.lastUpdate).toEqual(stalePersistedActivity)
expect(mockNotify).toHaveBeenCalledWith(task)
})
test("keeps a busy task running when persisted session activity lookup is unavailable", async () => {
//#given - the child session is active, but session.get returned an error response during confirmation
spyOn(globalThis.Date, "now").mockReturnValue(fixedTime)
const staleActivity = new Date(Date.now() - 45 * 60 * 1000)
const task = createRunningTask({
startedAt: staleActivity,
progress: {
toolCalls: 2,
lastUpdate: staleActivity,
},
})
//#when - stale checking cannot verify persisted activity for this poll
await checkAndInterruptStaleTasks({
tasks: [task],
client: mockClient,
config: { staleTimeoutMs: 180_000 },
concurrencyManager: mockConcurrencyManager,
notifyParentSession: mockNotify,
sessionStatuses: { "ses-1": { type: "busy" } },
getSessionActivity: async () => ({ type: "unavailable" }),
})
//#then - active-session cancellation is deferred instead of treating lookup failure as inactivity
expect(task.status).toBe("running")
expect(task.progress?.lastUpdate).toEqual(staleActivity)
expect(mockNotify).not.toHaveBeenCalled()
})
})
+27 -1
View File
@@ -19,6 +19,8 @@ import { removeTaskToastTracking } from "./remove-task-toast-tracking"
import { MIN_SESSION_GONE_POLLS, verifySessionExists } from "./session-existence"
import { isActiveSessionStatus } from "./session-status-classifier"
import { getSessionActivityFromClient, type SessionActivityResolver } from "./session-activity"
import { refreshTaskActivityFromSession } from "./task-activity-refresh"
const TERMINAL_TASK_STATUSES = new Set<BackgroundTask["status"]>([
"completed",
@@ -120,6 +122,7 @@ export async function checkAndInterruptStaleTasks(args: {
notifyParentSession: (task: BackgroundTask) => Promise<void>
sessionStatuses?: SessionStatusMap
onTaskInterrupted?: (task: BackgroundTask) => void
getSessionActivity?: SessionActivityResolver
}): Promise<void> {
const {
tasks,
@@ -137,6 +140,8 @@ export async function checkAndInterruptStaleTasks(args: {
const abortPromises: Array<Promise<unknown>> = []
const messageStalenessMs = config?.messageStalenessTimeoutMs ?? DEFAULT_MESSAGE_STALENESS_TIMEOUT_MS
const getSessionActivity = args.getSessionActivity
?? ((id: string) => getSessionActivityFromClient(client, id, directory))
for (const task of tasks) {
if (task.status !== "running") continue
@@ -157,6 +162,9 @@ export async function checkAndInterruptStaleTasks(args: {
const sessionGone = sessionMissing && (task.consecutiveMissedPolls ?? 0) >= MIN_SESSION_GONE_POLLS
const shouldSkipInactivityTimeout = task.teamRunId !== undefined && !sessionGone
const shouldRefreshFromSessionActivity = !sessionGone
&& sessionStatus !== undefined
&& isActiveSessionStatus(sessionStatus)
if (!task.progress?.lastUpdate) {
if (shouldSkipInactivityTimeout) continue
@@ -164,6 +172,12 @@ export async function checkAndInterruptStaleTasks(args: {
const effectiveTimeout = sessionGone ? sessionGoneTimeoutMs : messageStalenessMs
if (runtime <= effectiveTimeout) continue
if (shouldRefreshFromSessionActivity) {
const activityRefresh = await refreshTaskActivityFromSession(task, getSessionActivity)
if (activityRefresh.type === "unavailable") continue
if (activityRefresh.type === "activity" && now - activityRefresh.activityTime <= effectiveTimeout) continue
}
if (sessionGone && await verifySessionExists(client, sessionID, directory)) {
task.consecutiveMissedPolls = 0
continue
@@ -197,9 +211,21 @@ export async function checkAndInterruptStaleTasks(args: {
if (runtime < MIN_RUNTIME_BEFORE_STALE_MS) continue
const timeSinceLastUpdate = now - task.progress.lastUpdate.getTime()
let timeSinceLastUpdate = now - task.progress.lastUpdate.getTime()
const effectiveStaleTimeout = sessionGone ? sessionGoneTimeoutMs : staleTimeoutMs
if (timeSinceLastUpdate <= effectiveStaleTimeout) continue
if (shouldRefreshFromSessionActivity) {
const activityRefresh = await refreshTaskActivityFromSession(task, getSessionActivity)
if (activityRefresh.type === "unavailable") continue
const refreshedLastUpdate = task.progress?.lastUpdate.getTime()
?? (activityRefresh.type === "activity" ? activityRefresh.activityTime : undefined)
if (refreshedLastUpdate !== undefined && now - refreshedLastUpdate <= effectiveStaleTimeout) continue
if (refreshedLastUpdate !== undefined) {
timeSinceLastUpdate = now - refreshedLastUpdate
}
}
if (task.status !== "running") continue
if (sessionGone && await verifySessionExists(client, sessionID, directory)) {
@@ -191,9 +191,10 @@ describe("remapAgentKeysToDisplayNames", () => {
const result = remapAgentKeysToDisplayNames(agents)
// then exactly one row is emitted under the clean literal display name
expect(Object.keys(result)).toEqual(["Sisyphus - Ultraworker"])
expect(result["Sisyphus - Ultraworker"]).toEqual({
name: "Sisyphus - Ultraworker",
const displayName = getAgentListDisplayName("sisyphus")
expect(Object.keys(result)).toEqual([displayName])
expect(result[displayName]).toEqual({
name: displayName,
foo: "bar",
})
})
+24 -11
View File
@@ -61,6 +61,7 @@ Original error: ${createResult.error}`
log(`[look_at] Created session: ${sessionID}`)
log(`[look_at] Sending prompt with ${isBase64Input ? "base64 image" : "file"} to session ${sessionID}`)
let promptFailed = false
try {
await promptSyncWithModelSuggestionRetry(ctx.client, {
path: { id: sessionID },
@@ -83,27 +84,39 @@ Original error: ${createResult.error}`
queueBehavior: "defer",
})
} catch (promptError) {
promptFailed = true
log("[look_at] Prompt error (ignored, will still fetch messages):", promptError)
}
let observedMessages: unknown[] | undefined
let observedText: string | undefined
if (typeof ctx.client.session.status === "function") {
await waitForLookAtSessionResult(ctx.client, sessionID)
const waitResult = await waitForLookAtSessionResult(ctx.client, sessionID, {
allowStableIdleWithoutActivity: true,
allowEmptyStableIdleWithoutActivity: promptFailed,
})
observedText = waitResult.outcome.text ?? undefined
if (observedText) {
observedMessages = waitResult.messages
}
}
log(`[look_at] Fetching messages from session ${sessionID}...`)
const messagesResult = await ctx.client.session.messages({
path: { id: sessionID },
})
let messages = observedMessages
if (!messages) {
log(`[look_at] Fetching messages from session ${sessionID}...`)
const messagesResult = await ctx.client.session.messages({
path: { id: sessionID },
})
if (messagesResult.error) {
log("[look_at] Messages error:", messagesResult.error)
return `Error: Failed to get messages: ${messagesResult.error}`
if (messagesResult.error) {
log("[look_at] Messages error:", messagesResult.error)
return `Error: Failed to get messages: ${messagesResult.error}`
}
messages = messagesResult.data
}
const messages = messagesResult.data
log(`[look_at] Got ${messages.length} messages`)
const responseText = extractLatestAssistantText(messages)
const responseText = observedText ?? extractLatestAssistantText(messages)
if (!responseText) {
log("[look_at] No assistant message found")
return "Error: No response from multimodal-looker agent"
+12
View File
@@ -78,6 +78,18 @@ describe("waitForLookAtSessionResult", () => {
expect(result.outcome.text).toBe("done")
})
test("#given session is absent and has no assistant output #when stable idle is allowed #then keeps polling", async () => {
const client = createMockClient([{ data: {} }], [])
await expect(
waitForLookAtSessionResult(unsafeTestValue(client), "ses_test", {
pollIntervalMs: 10,
timeoutMs: 50,
allowStableIdleWithoutActivity: true,
}),
).rejects.toThrow("timed out")
})
test("#given session never becomes idle #when polling exceeds timeout #then rejects", async () => {
const client = createMockClient(
[{ data: { ses_test: { type: "busy" } } }],
+5 -1
View File
@@ -9,6 +9,7 @@ export interface PollOptions {
timeoutMs?: number
abortSignal?: AbortSignal
allowStableIdleWithoutActivity?: boolean
allowEmptyStableIdleWithoutActivity?: boolean
}
const DEFAULT_POLL_INTERVAL_MS = 1000
@@ -136,7 +137,10 @@ export async function waitForLookAtSessionResult(
const canConcludeIdle =
sawNonIdleStatus ||
!status.supported ||
Boolean(options?.allowStableIdleWithoutActivity)
(
Boolean(options?.allowStableIdleWithoutActivity)
&& (outcome.hasAssistant || Boolean(options?.allowEmptyStableIdleWithoutActivity))
)
if (canConcludeIdle && stableIdlePolls >= IDLE_STABILITY_POLLS_REQUIRED) {
return { messages, outcome, statusType }