fix(task): wait up to 72 hours before timing out tasks

Some vendors keep a task running for more than a day: Volcengine Ark waits
48 hours by default and up to 72 when configured, and Vidu off-peak jobs
can take 48 hours. With the 24-hour default the gateway failed and refunded
those tasks while the vendor was still producing, and charging for, the
result. TASK_TIMEOUT_MINUTES now defaults to 4320. A task that is truly
stuck is refunded after 72 hours instead of 24.

.env.example

@@ -59,7 +59,7 @@
 # 更新任务启用
 # UPDATE_TASK=true
 # 异步任务硬超时（分钟），按提交时间计算，超时未完成的任务标记失败并退款；0 表示禁用
-# TASK_TIMEOUT_MINUTES=1440
+# TASK_TIMEOUT_MINUTES=4320
 # 异步任务连续轮询失败阈值（上游 429/5xx/401/403、网络错误、无法识别的响应），达到后任务标记失败并退款；正常轮询成功一次即归零
 # TASK_POLL_MAX_FAILURES=20
 

common/init.go

@@ -206,7 +206,7 @@ func initConstantEnv() {
 	// 任务轮询时查询的最大数量
 	constant.TaskQueryLimit = GetEnvOrDefault("TASK_QUERY_LIMIT", 1000)
 	// 异步任务超时时间（分钟），超过此时间未完成的任务将被标记为失败并退款。0 表示禁用。
-	constant.TaskTimeoutMinutes = GetEnvOrDefault("TASK_TIMEOUT_MINUTES", 1440)
+	constant.TaskTimeoutMinutes = GetEnvOrDefault("TASK_TIMEOUT_MINUTES", 4320)
 	// Consecutive unrecognized/transient poll failures before the task is failed and refunded.
 	constant.TaskPollMaxFailures = GetEnvOrDefault("TASK_POLL_MAX_FAILURES", 20)
 	// 声明式任务协议桥只观察数据库；这些值控制一次客户端观察连接，

service/task_polling.go

@@ -776,7 +776,8 @@ func unrecognizedPollDetail(reason string, body []byte) string {
 func recordPollFailure(ctx context.Context, adaptor TaskPollingAdaptor, task *model.Task, fromStatus model.TaskStatus, class string, statusCode int, detail string) error {
 	task.PrivateData.PollFailures++
 	// TASK_POLL_MAX_FAILURES <= 0 disables the consecutive-failure cutoff, matching
-	// TASK_TIMEOUT_MINUTES semantics; the 24h sweep remains the only backstop.
+	// TASK_TIMEOUT_MINUTES semantics; the 72h sweep (TASK_TIMEOUT_MINUTES default
+	// 4320) remains the only backstop.
 	giveUp := constant.TaskPollMaxFailures > 0 && task.PrivateData.PollFailures >= constant.TaskPollMaxFailures
 	if giveUp || class == pollClassUnrecognized || class == pollClassHookError {
 		// The redacted body is intentionally not persisted to Task.Data on these

service/task_polling_test.go

@@ -775,6 +775,40 @@ func TestSweepTimedOutTasksHonorsRefundRolloutBoundary(t *testing.T) {
 	assert.Equal(t, int64(1), countLogs(t))
 }
 
+// TASK_TIMEOUT_MINUTES defaults to 72 hours, the longest a vendor keeps a
+// task running (Ark can be configured to 72h, Vidu off-peak runs 48h).
+func TestSweepTimedOutTasksAtTheDefaultDeadline(t *testing.T) {
+	truncate(t)
+
+	const userID, initialQuota, taskQuota = 404, 10_000, 1_500
+	seedUser(t, userID, initialQuota)
+	running := makeTask(userID, 0, taskQuota, 0, BillingSourceWallet, 0)
+	running.TaskID = "running_for_60_hours"
+	running.Progress = "50%"
+	running.SubmitTime = time.Now().Add(-60 * time.Hour).Unix()
+	require.NoError(t, model.DB.Create(running).Error)
+	stuck := makeTask(userID, 0, taskQuota, 0, BillingSourceWallet, 0)
+	stuck.TaskID = "running_for_73_hours"
+	stuck.Progress = "50%"
+	stuck.SubmitTime = time.Now().Add(-73 * time.Hour).Unix()
+	require.NoError(t, model.DB.Create(stuck).Error)
+
+	previousTimeout := constant.TaskTimeoutMinutes
+	constant.TaskTimeoutMinutes = 4320
+	t.Cleanup(func() { constant.TaskTimeoutMinutes = previousTimeout })
+
+	sweepTimedOutTasks(context.Background())
+
+	var reloadedRunning, reloadedStuck model.Task
+	require.NoError(t, model.DB.First(&reloadedRunning, running.ID).Error)
+	require.NoError(t, model.DB.First(&reloadedStuck, stuck.ID).Error)
+	assert.NotEqualValues(t, model.TaskStatusFailure, reloadedRunning.Status)
+	assert.Equal(t, taskQuota, reloadedRunning.Quota)
+	assert.EqualValues(t, model.TaskStatusFailure, reloadedStuck.Status)
+	assert.Equal(t, "Task timed out", reloadedStuck.FailReason)
+	assert.Equal(t, initialQuota+taskQuota, getUserQuota(t, userID))
+}
+
 type scriptedPollingAdaptor struct {
 	statusCode int
 	body       []byte