Skip to content

Recovery After Interrupted Processing

sajidbuet edited this page Sep 24, 2026 · 1 revision

Recovery After Interrupted Processing

Batch processing is designed to be interrupted. Nothing already read is lost, whether the run was cancelled, the window closed, the application crashed or the machine lost power.

Why it works

Each sheet's result is committed to the project database as that sheet finishes, not accumulated in memory and written at the end. So at any instant the database holds every sheet completed so far, and the rest are still marked as outstanding.

What to do

Reopen the project and go to the Scan stage.

Button Use it when
Resume Batch Continue the run. Sheets already completed are not reprocessed
Retry Failed Re-attempt only the sheets that failed

Both act on the stored batch, so they work after a restart and are enabled only when there is genuinely something left to do.

When a project with an interrupted batch is opened, OMRFlow recovers the outstanding work and reports how many sheets it picked up.

Closing while a batch runs

You are asked to confirm, and told that sheets already read are saved and the batch can be resumed. On confirmation the batch is stopped and waited for before the project is released — the worker processes own the database writes, and closing underneath them would abort the very writes that make the run resumable.

Checking afterwards

Application menu → Tools → Project Health / Recovery… reports structural integrity, foreign keys, stale jobs left in a running state, and missing or changed source scans. It reports and never silently repairs.

Worker processes

Recognition runs in a pool of worker processes. If the coordinator dies, the workers die with it — they are bound to its lifetime, so a crashed run cannot leave processes consuming CPU in the background. Workers never have database write access, so an abrupt end cannot corrupt the project.

What is not covered

Recovery is designed against process interruption — cancellation, crashes, a lost coordinator. It is not a defence against storage failure or filesystem corruption; nothing in OMRFlow is. That is what backups are for.

Related

Clone this wiki locally