Elbuz runs long operations in the background, in parts: bulk text generation, auto-enrichment, AI matching. The “Job queue” shows what's done and what failed.
When you need this:
- “Added to the job queue”: the operation runs in the background; you can close the window.
- A job is complete, but products lack results: the lower grid shows each item's status; retry failures.
- Old jobs piled up: delete one with its row's trash icon; after three days they vanish themselves.
- Enrichment is slow: threads already run in parallel; a second API key speeds it up.
Where to start: open the “Job queue” via the counter in the base catalog's bottom bar.
What the job queue does
- Finishes a long operation in the background — every minute, even when the window you started it from is closed.
- Shows every item of a job — a product, category or row, with its own status and number of attempts.
- Retries an item that ran out of time — until the attempts are used up.
- Clears old jobs on its own — history is kept for three to four days.
- Processes jobs in several threads at once — no separate setting needed.
How to read the “Job queue” window
When an operation is too long for one request, the app creates a job instead of finishing it on the spot and shows the line “The operation is not finished yet and was added to the job queue (runs in the background)” with the link “Open job queue list”. This is not an error — it's normal for bulk work.
The queue itself is worked through by a system schedule every minute. You don't need to create or configure it: it finishes unfinished work even when the process that started it broke off.
1
Status column of the jobs grid2
Number of attempts column3
Job items grid- Status — the job state; the values (new, work, complete) appear here as Latin words.
- Number of attempts — how many times the job was started.
- The lower grid lists the items of the selected job; right now it is empty because the queue has no jobs.
The “Job queue” window has two grids, one above the other. The upper one lists the jobs; its main columns are “Type”, “Subtype”, “Status”, “Date added”, “Start date”, “End date”, “Running time”, “Number of attempts”, “Source ID”, “Note”. The lower one shows the items of the job you selected in the upper grid.
Statuses in both grids are shown exactly as they are stored in the database: new — created and not picked up yet, work — being processed, complete — the app considers the job finished. They are not translated, and this is not a display glitch.
Three things about how the queue behaves explain most questions:
- Running it again replaces the previous job. If you run the same operation for the same source a second time, the previous job disappears together with all its items, and the count starts over.
- History is kept for three to four days. When the next job is created, the app deletes all jobs added before the start of the day three days ago, so a job lives from three to almost four days. Only creating a regular job triggers this cleanup; background AI agent jobs don't, so old jobs stay until new ones appear. You can't find out what happened last week.
- An empty job isn't created. If no product matches the operation's conditions, the job is deleted right after it's created. There's no row in the queue and no log entry — so “nothing happened” sometimes means “nothing was supposed to happen”.
One more situation concerns jobs that use artificial intelligence: bulk text generation, auto-enrichment, AI matching. On a demo account AI works with a daily limit: up to €3 a day for AI, parsing and media generation combined.
Once the limit is used up, and on an account without an key, such jobs are removed entirely, with a line in the log. You'll have to create them again: the next day, on a paid plan, or once the key is entered.
Queue progress is recorded in the “Operation log” under the “Job Queue” type. More details: where to see the result.
The trash icon in the upper toolbar has the tooltip “Delete all”, but what exactly it deletes depends on how the window was opened.
If the window was opened with a filter — from a template card or from the bulk generation window — the button deletes only the jobs of that selection (type and source) together with their items, and the confirmation reads “Delete the jobs of the current selection (the type and source shown in the window)?”.
If the window was opened without a filter, the button wipes the whole account's queue — all jobs and all their items — and the dialog asks “Delete all?”. Neither case can be undone. To remove a single job, use the trash icon in its row.
Why a job is complete but not everything was done
A job is split into items: one product, one category, one row. A job can have thousands of them, and each has its own status.
1
Status2
Number of attempts- Status — the outcome for each item; here all eight items finished as complete, and any that failed would show a different value.
- Number of attempts — how many times the item was processed.
An item that didn't finish in the allotted time or ran out of attempts gets the zombie status. The limits depend on the type of work: a few minutes in some cases, an hour in others; there are from two to five attempts, and the longest branch has no attempt counter at all.
Usually a zombie returns to the queue and is retried until the attempts run out; then it stays zombie for good. The same goes for an item where the AI returns an error every time (for example, a wrong key or no access to the model): after the last attempt it becomes zombie, and the job runs to the end.
- Start
- State
- End
A job is considered finished when all its items have the complete or zombie status. So items that never went through count as processed, the job reaches complete and looks successful in the upper grid. Only the lower grid shows the real picture: filter the items by the zombie status and repeat the operation for exactly those.
What you pay for in AI jobs
The billable unit in such a job is not the job and not the product, but each request to the AI service. A request's price is based on tokens: input (what the app sent to the service) and output (what it replied), each at its own price, with a coefficient that depends on the type of operation.
In bulk text generation, one request writes one field of one product.
- If the service didn't return a token count — for example, the request failed with an error — there's no cost record.
- If an item went for another attempt, that's a new request, and it's billed separately from the first one.
You can see what was spent in the “Operation log”, in the “Cost” column: each request leaves a row there with its own amount, and in “Summary” mode the column shows the total for the whole operation.
Money isn't charged to your balance for each request separately: a background charge adds up all not-yet-charged requests of one type and deducts them in one row, rounding any amount below 0.01 up to 0.01. These rows are on the “Transaction history” tab of the “Balance” window.
How to clear many old jobs of one type
The “Job queue” window has no separate “delete all jobs of this type” button. The job type is shown in the “Type” column, for example “AI product matching” or “AI category matching”. You can clear old AI matching jobs in three ways:
- One at a time. The trash icon in a job's row deletes exactly that row together with its items. The grid lets you select only one row, so you can't delete several jobs at once this way.
- The whole queue at once. The trash icon in the upper toolbar, in a window opened without a selection, wipes all jobs of all types in the account. Do this only when the queue has nothing you need: export, import and text generation will disappear along with matching.
- Wait. Jobs added before the start of the day three days ago are deleted on their own as soon as the next job is created (except background jobs of the AI agent).
A filter in the “Type” column only hides rows from view: the trash icon in the upper toolbar still deletes by the selection the window was opened with, not by the grid filter. Running the same matching again for the same source also removes the previous job — the new one takes its place.
Can the queue be processed in parallel to enrich products faster
There's no button or setting for a second worker, and it wouldn't add speed: parallel processing already works.
- The worker starts on its own — every minute on schedule and right after a job is created. Only the scheduled start sets the “run in progress” mark, so these two starts sometimes already run at the same time.
- The AI product and category matching windows, while open, push processing of their own job type themselves, no more often than once every 8 seconds.
- Within a run, the worker distributes jobs across several simultaneous threads.
- The thread limit is no more than five when AI jobs have used over 100,000 tokens in the last three minutes. Several API keys (OpenAI, Anthropic) raise this limit.
A second API key speeds up enrichment, not a second worker.
Scenarios
Bulk generation finished, but some products have no text
What to do. Open the “Job queue”, select the job in the upper grid, and in the lower grid filter the items by the zombie status. Run the operation again for exactly these products.
What you get. The new job takes only the products that failed the first time. If the cause is the key or access to the model, fix the key first: otherwise the items will become zombie again.
The queue is clogged with old AI matching jobs
What to do. Delete the jobs you want one at a time with the trash icon in their rows, or wait for the automatic cleanup. Click the upper trash icon in a window without a selection only when the queue has nothing you need.
What you get. The jobs disappear together with their items. Jobs older than three days will be cleared by the app itself when the next job is created.
Auto-enrichment of thousands of products is too slow
What to do. Add one more AI service API key to the account.
What you get. The limit of simultaneous threads goes up, and the job is processed faster. There's no need to look for a second queue worker — there isn't one, and you don't need it.
What to do when a job fails
| What you see | Why | What to do |
|---|---|---|
| “The operation is not finished yet and was added to the job queue (runs in the background)” | The operation is too large for one request — the picked it up | Wait; follow the progress via the “Open job queue list” link (about the queue) |
The job has the complete status, but not all the work is done |
Some items became and were counted as processed | Open the lower grid: each is shown there with its status. Filter by zombie and repeat the operation for them (about items) |
| “The daily demo limit is used up: up to €3 a day for AI, parsing and media generation. You can continue tomorrow or on a paid plan.” | The €3 daily limit on a demo account is used up while an AI job was in the queue. Such jobs are removed entirely | Create the job again tomorrow or switch to a paid plan (about the queue) |
| “No AI access: AI service API key is not configured.” | The account has no active AI service key | Enter the key and create the job again (about the queue) |
| “Delete all?” | The confirmation of the trash icon in a “Job queue” window opened without a filter. In such a window the upper trash icon wipes the whole account's queue; in a window with a selection it deletes only that selection's jobs and asks differently | Cancel and delete a single job with the trash icon in its row (about deleting) |
| “Update failed” | An edit made directly in the queue grid wasn't saved | Reload the window with the “Refresh” button and try again |
| Instead of its contents, the “Job queue” shows an access denied message | The user's group has no right to the “Job queue” window | Grant the right in group settings (about access) |
FAQ
I ran an operation a second time — where did the first job go?
Running the same operation again for the same source replaces the previous job together with its items. The count starts over.
Can I see what was in the queue last week?
No. Jobs live from three to almost four days and are deleted when the next one is created. A trace of the operation stays in the “Operation log” under the “Job Queue” type for 14 days.
Can I export the queue to a file?
No, the queue can't be exported to a file. The “Operation log” can be exported as an Excel table with the “Export to file” item (how to search the log).
The new, work, complete statuses are in English — is that a glitch?
No. The statuses are shown as they are stored in the database; they aren't translated. What each one means is explained in the section about the queue window.
I ran an operation, but the queue is empty. Why?
No product matched the operation's conditions, and the empty job was deleted right after it was created. Check the product selection and run it again.
Where to find it and who has access
The job queue has no menu item — it opens from other windows.
- Base catalog
- the counter with a clock icon in the bottom bar
- Other entry points
- the “Open job queue list” link in an operation's message; “Job queue” in the API export template card
- Bulk work windows
- product auto-enrichment and bulk text generation
- Access right
- the “Job queue” window in group permissions
- Prepare
- for AI jobs — a paid plan and an AI service key entered in the account
Related topics
- If you need the overall picture of the schedule and background jobs — automatic scheduled runs.
- If a background operation didn't start on schedule — the schedule didn't run.
- If you need to find who started an operation and when — operation log.
- If you want to know how much was charged for AI jobs — balance transaction history.
