test(delegate-task): cover aborted concurrent background launches

This commit is contained in:
YeonGyu-Kim
2026-03-31 15:11:12 -07:00
parent 56cf16c4c5
commit 9a0f2ff9a7
3 changed files with 363 additions and 3 deletions
@@ -7,6 +7,7 @@ const afterEachFn = bunTest.afterEach
const { executeBackgroundTask } = require("./background-task")
const { __setTimingConfig, __resetTimingConfig } = require("./timing")
const { SessionCategoryRegistry } = require("../../shared/session-category-registry")
describeFn("executeBackgroundTask output/session metadata compatibility", () => {
beforeEachFn(() => {
@@ -19,6 +20,7 @@ describeFn("executeBackgroundTask output/session metadata compatibility", () =>
afterEachFn(() => {
__resetTimingConfig()
SessionCategoryRegistry.clear()
})
testFn("does not emit synthetic pending session metadata when session id is unresolved", async () => {
@@ -201,4 +203,226 @@ describeFn("executeBackgroundTask output/session metadata compatibility", () =>
{ permission: "question", action: "deny", pattern: "*" },
])
})
testFn("keeps launched background task alive when parent aborts before session id resolves", async () => {
//#given - parallel tool execution can abort the parent call after launch succeeds
const metadataCalls: any[] = []
const abortController = new AbortController()
const manager = {
launch: async () => ({
id: "bg_abort_after_launch",
sessionID: undefined,
description: "Abort after launch",
agent: "explore",
status: "pending",
}),
getTask: () => {
abortController.abort()
return { sessionID: undefined, status: "pending" }
},
}
//#when
const result = await executeBackgroundTask(
{
description: "Abort after launch",
prompt: "check",
run_in_background: true,
load_skills: [],
},
{
sessionID: "ses_parent",
callID: "call_abort_after_launch",
metadata: async (value: any) => metadataCalls.push(value),
abort: abortController.signal,
},
{ manager },
{ sessionID: "ses_parent", messageID: "msg_abort_after_launch" },
"explore",
undefined,
undefined,
undefined,
)
//#then - background launch should still succeed without fake abort failure
expectFn(result).toContain("Background task launched")
expectFn(result).toContain("Background Task ID: bg_abort_after_launch")
expectFn(result).not.toContain("Task aborted while waiting for session to start")
expectFn(metadataCalls).toHaveLength(1)
expectFn("sessionId" in metadataCalls[0].metadata).toBe(false)
})
testFn("registers late session category even when parent aborts before session id resolves", async () => {
//#given - session wiring should continue after returning early on parent abort
const abortController = new AbortController()
abortController.abort()
let reads = 0
const manager = {
launch: async () => ({
id: "bg_abort_category",
sessionID: undefined,
description: "Abort category",
agent: "explore",
status: "pending",
}),
getTask: () => {
reads += 1
return reads >= 2
? { sessionID: "ses_abort_category", status: "running" }
: { sessionID: undefined, status: "pending" }
},
}
//#when
const result = await executeBackgroundTask(
{
description: "Abort category",
prompt: "check",
run_in_background: true,
load_skills: [],
category: "quick",
},
{
sessionID: "ses_parent",
callID: "call_abort_category",
metadata: async () => {},
abort: abortController.signal,
},
{ manager },
{ sessionID: "ses_parent", messageID: "msg_abort_category" },
"explore",
undefined,
undefined,
[{ providers: ["openai"], model: "gpt-5.4" }],
)
await new Promise(resolve => setTimeout(resolve, 5))
//#then - late session setup should still register category for runtime fallback
expectFn(result).toContain("Background task launched")
expectFn(SessionCategoryRegistry.get("ses_abort_category")).toBe("quick")
})
testFn("prefers child terminal status over parent abort while waiting for session id", async () => {
//#given - failed child launch should not be misreported as a successful background launch
const abortController = new AbortController()
abortController.abort()
const manager = {
launch: async () => ({
id: "bg_abort_terminal",
sessionID: undefined,
description: "Abort terminal",
agent: "explore",
status: "pending",
}),
getTask: () => ({ sessionID: undefined, status: "interrupt" }),
}
//#when
const result = await executeBackgroundTask(
{
description: "Abort terminal",
prompt: "check",
run_in_background: true,
load_skills: [],
},
{
sessionID: "ses_parent",
callID: "call_abort_terminal",
metadata: async () => {},
abort: abortController.signal,
},
{ manager },
{ sessionID: "ses_parent", messageID: "msg_abort_terminal" },
"explore",
undefined,
undefined,
undefined,
)
//#then - terminal child status should win over abort and surface the failure
expectFn(result).toContain("Task failed to start")
expectFn(result).toContain("interrupt")
})
testFn("keeps sibling background launch alive when two tasks start concurrently", async () => {
//#given - one aborted parent call should not interrupt a sibling launch from the same parent session
const firstAbortController = new AbortController()
const secondAbortController = new AbortController()
const states = new Map([
["bg_first", { reads: 0, abortOnFirstRead: true, sessionID: "ses_first" }],
["bg_second", { reads: 0, abortOnFirstRead: false, sessionID: "ses_second" }],
])
let launchCount = 0
const manager = {
launch: async () => {
launchCount += 1
return launchCount === 1
? { id: "bg_first", sessionID: undefined, description: "First", agent: "explore", status: "pending" }
: { id: "bg_second", sessionID: undefined, description: "Second", agent: "explore", status: "pending" }
},
getTask: (taskID: string) => {
const state = states.get(taskID)
if (!state) return undefined
state.reads += 1
if (state.abortOnFirstRead && state.reads === 1) {
firstAbortController.abort()
}
return state.reads >= 2
? { sessionID: state.sessionID, status: "running" }
: { sessionID: undefined, status: "pending" }
},
}
//#when
const [firstResult, secondResult] = await Promise.all([
executeBackgroundTask(
{
description: "First",
prompt: "check",
run_in_background: true,
load_skills: [],
},
{
sessionID: "ses_parent",
callID: "call_first",
metadata: async () => {},
abort: firstAbortController.signal,
},
{ manager },
{ sessionID: "ses_parent", messageID: "msg_first" },
"explore",
undefined,
undefined,
undefined,
),
executeBackgroundTask(
{
description: "Second",
prompt: "check",
run_in_background: true,
load_skills: [],
},
{
sessionID: "ses_parent",
callID: "call_second",
metadata: async () => {},
abort: secondAbortController.signal,
},
{ manager },
{ sessionID: "ses_parent", messageID: "msg_second" },
"explore",
undefined,
undefined,
undefined,
),
])
//#then - both tasks still launch and the sibling is not reported as interrupted
expectFn(firstResult).toContain("Background task launched")
expectFn(firstResult).not.toContain("Task failed to start")
expectFn(secondResult).toContain("Background task launched")
expectFn(secondResult).toContain("session_id: ses_second")
expectFn(secondResult).not.toContain("interrupt")
})
})
+53 -3
View File
@@ -10,6 +10,43 @@ import { SessionCategoryRegistry } from "../../shared/session-category-registry"
import { QUESTION_DENIED_SESSION_PERMISSION } from "../../shared/question-denied-session-permission"
import { setSessionFallbackChain } from "../../hooks/model-fallback/hook"
function continueSessionSetup(args: {
taskID: string
manager: ExecutorContext["manager"]
timing: ReturnType<typeof getTimingConfig>
fallbackChain?: FallbackEntry[]
category?: string
}): void {
if (!args.fallbackChain && !args.category) {
return
}
void (async () => {
const waitStart = Date.now()
while (Date.now() - waitStart < args.timing.WAIT_FOR_SESSION_TIMEOUT_MS) {
await new Promise(resolve => setTimeout(resolve, args.timing.WAIT_FOR_SESSION_INTERVAL_MS))
const updated = args.manager.getTask(args.taskID)
if (!updated) {
return
}
if (updated.status === "error" || updated.status === "cancelled" || updated.status === "interrupt") {
return
}
const sessionId = updated.sessionID
if (!sessionId) {
continue
}
setSessionFallbackChain(sessionId, args.fallbackChain)
if (args.category) {
SessionCategoryRegistry.register(sessionId, args.category)
}
return
}
})()
}
export async function executeBackgroundTask(
args: DelegateTaskArgs,
ctx: ToolContextWithMetadata,
@@ -50,12 +87,25 @@ export async function executeBackgroundTask(
const waitStart = Date.now()
let sessionId = task.sessionID
while (!sessionId && Date.now() - waitStart < timing.WAIT_FOR_SESSION_TIMEOUT_MS) {
const updated = manager.getTask(task.id)
if (updated?.status === "error" || updated?.status === "cancelled" || updated?.status === "interrupt") {
return `Task failed to start (status: ${updated.status}).\n\nTask ID: ${task.id}`
}
sessionId = updated?.sessionID
if (sessionId) {
break
}
if (ctx.abort?.aborted) {
return `Task aborted while waiting for session to start.\n\nTask ID: ${task.id}`
continueSessionSetup({
taskID: task.id,
manager,
timing,
fallbackChain,
category: args.category,
})
break
}
await new Promise(resolve => setTimeout(resolve, timing.WAIT_FOR_SESSION_INTERVAL_MS))
const updated = manager.getTask(task.id)
sessionId = updated?.sessionID
}
if (sessionId) {
+86
View File
@@ -1453,6 +1453,92 @@ describe("sisyphus-task", () => {
expect(launchCalled).toBe(true)
expect(result).toContain("Background task launched")
}, { timeout: 10000 })
test("#given concurrent background launches from the same parent #when one parent call aborts during session wait #then sibling launch is not interrupted", async () => {
// given
const { createDelegateTask } = require("./tools")
const firstAbortController = new AbortController()
const secondAbortController = new AbortController()
const taskStates = new Map([
["bg_tool_first", { reads: 0, abortOnFirstRead: true, sessionID: "ses_tool_first" }],
["bg_tool_second", { reads: 0, abortOnFirstRead: false, sessionID: "ses_tool_second" }],
])
let launchCount = 0
const mockManager = {
launch: async () => {
launchCount += 1
return launchCount === 1
? {
id: "bg_tool_first",
sessionID: undefined,
description: "Tool first",
agent: "Sisyphus-Junior",
status: "running",
}
: {
id: "bg_tool_second",
sessionID: undefined,
description: "Tool second",
agent: "Sisyphus-Junior",
status: "running",
}
},
getTask: (taskID: string) => {
const state = taskStates.get(taskID)
if (!state) return undefined
state.reads += 1
if (state.abortOnFirstRead && state.reads === 1) {
firstAbortController.abort()
}
return state.reads >= 2
? { sessionID: state.sessionID, status: "running" }
: { sessionID: undefined, status: "pending" }
},
}
const mockClient = {
app: { agents: async () => ({ data: [] }) },
config: { get: async () => ({ data: { model: SYSTEM_DEFAULT_MODEL } }) },
model: { list: async () => [] },
session: {
create: async () => ({ data: { id: "ses_bg_explicit_true" } }),
prompt: async () => ({ data: {} }),
promptAsync: async () => ({ data: {} }),
messages: async () => ({ data: [] }),
},
}
const tool = createDelegateTask({ manager: mockManager, client: mockClient })
// when
const [firstResult, secondResult] = await Promise.all([
tool.execute(
{
description: "Tool first",
prompt: "Run background",
category: "quick",
run_in_background: true,
load_skills: [],
},
{ sessionID: "parent-session", messageID: "parent-message-1", agent: "sisyphus", abort: firstAbortController.signal }
),
tool.execute(
{
description: "Tool second",
prompt: "Run background",
category: "quick",
run_in_background: true,
load_skills: [],
},
{ sessionID: "parent-session", messageID: "parent-message-2", agent: "sisyphus", abort: secondAbortController.signal }
),
])
// then
expect(firstResult).toContain("Background task launched")
expect(firstResult).not.toContain("Task failed to start")
expect(secondResult).toContain("Background task launched")
expect(secondResult).toContain("session_id: ses_tool_second")
expect(secondResult).not.toContain("interrupt")
}, { timeout: 10000 })
})
describe("session_id with background parameter", () => {