Jobs and Notifications

What runs as a job

Long work does not block the interface. These run as jobs: dataset imports, archive extraction, format conversion, profiling, transforms, exports, storage clean-up and sync, pipeline steps and Research figure generation.

Jobs are durable: they are saved, so a reload does not forget a queue. They still only run while the Studio is open in a tab. Closing the browser pauses everything; nothing runs on a server.

Reading the Job Centre

The Job Centre has a summary rail and the queue as a ledger. A job is in one of five states: running, queued, failed, cancelled or done. Running jobs are listed first, then queued, failed, cancelled and done. The rail shows counts, how many workers are busy, bytes processed and the current throughput. Pipeline steps are grouped under the run they belong to, so a ten-step run is one row.

Controlling the queue

• Cancel stops a running or queued job. Cancelling a pipeline row cancels the whole run.
• Retry runs a failed job again. The attempt counter advances, and the failure's diagnostic stays visible so you can see what changed.
• Pause queue stops new jobs from starting; Resume lets them continue. Use it before a heavy training run or while you reorganise a dataset.
• Queued imports can be prioritised or removed without touching anything that is already running.

Notifications

While the Studio is on screen, results appear as in-app toasts. Because jobs only run while a tab is open, the useful extra signal is an alert while that tab is in the background. Under Settings → Notifications you choose:

• Desktop alerts, with an option to send them only when the Studio is in the background.
• Which outcomes to announce: finished jobs and pipeline runs, and failed ones (a run held at a quality gate, or finished with a warning, counts with the failures).
• A webhook, which your browser calls itself when a job finishes or fails. Nothing goes through the DLWAY server.
• A minimum duration: jobs and runs shorter than this are skipped, so quick work does not interrupt you.

Housekeeping jobs (clean-up and sync) never notify, and a pipeline raises one alert when the whole run ends instead of one per step. After a reload, jobs that were already finished are not announced again.

When a job fails

Open the failed row for its diagnostic. A damaged archive, an unreachable URL and an unsupported format all say so in plain words. Fix the cause and press Retry. If the failure is a storage problem, see Storage in Profiles, versions and lineage.