---
title: "Watch and retry the background job queue"
description: "Read the Background Jobs screen to see what is waiting and what has broken, open the error on a single job, and then retry it or cancel it."
updated: "2026-08-28"
audience: [administrator]
source: "https://helpdesk-docs.svr1.feedapp.click/en/quan-tri/cong-viec-nen"
---

Email, notifications, webhooks and AI indexing do not run during the click that
asks for them — they queue. This page reads that queue and deals with a job that
has broken.

## Goal [#goal]

You can answer "has that email gone out yet" with evidence on screen, and you
can put a broken job back on the queue.

## Before you begin [#before-you-begin]

* This screen is reserved for the **administrator** role and cannot be granted through a permission group.
* Know which queue your job belongs to; the filter splits the list six ways.
* Redis has to be running, because the queue lives there. Without it nothing leaves the pending state.
* If you are chasing an email that never arrived, read [Configure notifications for the workspace](/en/quan-tri/thong-bao) first.

## Steps [#steps]

1. Open **Administration** → **Background Jobs**. Four counters sit under the
   heading, with the job table below them.

   ![The Background Jobs screen with Pending, Active, Completed and Failed counters, two filter selects, and a job table listing queue, job name, status and duration](https://helpdesk-docs.svr1.feedapp.click/images/docs/quan-tri/cong-viec-nen/01-hang-doi-tac-vu.en.webp)

   *Counters on top, two filters in the middle, the job table underneath.*
2. Read the four counters: `Pending`, `Active`, `Completed` and `Failed`. Those
   labels are not translated and read in English in either console language.
3. Narrow the list with the left-hand select. The six queues are `AI Source
   Processing`, `AI Auto Reply`, `Notifications`, `Email Notifications` and
   `Outbound Webhooks`.
4. Narrow further with the status select on the right. Pick `Failed` to leave only
   the jobs that broke.
5. Click a row. The **Job Detail** panel slides in from the right edge with
   **Attempts**, **Error** and **Payload** on it.
6. Read **Error** first. It is the message the worker actually received, and it
   usually names the cause — a wrong key, an expiry, a refusal from the far end.
7. Press **Retry** to put the job back on its queue. The button appears on failed
   jobs only.
8. Press **Cancel** to stop a job that is waiting or running. That button appears
   on those two states only.
9. Clear the table with **Purge completed** at the top right. The dialog states the
   scope: every completed, failed and cancelled job older than 30 days.

## Verify [#verify]

After **Retry**, the row leaves the failed state and the `Failed` counter drops
by one. The page refreshes itself every eight seconds — the line **Auto-refreshing
every 8s** at the right of the filter row says so — so there is no need to press
**Refresh**. If the job finished, reopen the detail panel: **Attempts** has gone
up and the **Error** block is gone.

> **The detail panel can hold customer data:** The **Payload** block prints the data the job was handed verbatim, and for a
> notification job that includes a conversation id and contact details. Think
> before you screenshot or share your screen on that block.

> **Administrators only, by design:** There is no way to grant access to this screen from the **Access Rights**
> screen. The console checks the workspace role on the spot and answers **Access
> Denied** to an agent, whatever their groups hold. The detail is in
> [Grant per-feature permissions to a group](/en/quan-tri/phan-quyen).

## Troubleshooting [#troubleshooting]

| Symptom                                   | Cause                                            | What to do                                                                   |
| ----------------------------------------- | ------------------------------------------------ | ---------------------------------------------------------------------------- |
| The table is completely empty             | No job exists yet, or the filters are too narrow | Return both selects to their first option                                    |
| Every job sits at `Pending`               | Nothing is consuming the queue                   | Ask whoever runs the deployment to check the worker process and Redis        |
| **Retry** runs and the job breaks again   | The underlying cause is still there              | Read **Error** in the detail panel and fix the configuration before retrying |
| A job runs out of **Attempts** unfinished | It has used its maximum attempt count            | Fix the cause, then press **Retry** once more                                |
| The table is full of old rows             | Nobody has ever cleared the queue                | Press **Purge completed**; the action cannot be undone                       |

## Next steps [#next-steps]

* [Configure notifications for the workspace](/en/quan-tri/thong-bao)
* [Create an AI agent and feed it sources](/en/quan-tri/ai-agent)
* [Grant per-feature permissions to a group](/en/quan-tri/phan-quyen)