feat(ocr): per-sender specialized Kurrent models with automatic active-learning retraining #263

Merged
marcel merged 32 commits from feat/issue-253-sender-models into main 2026-04-18 12:30:57 +02:00
Showing only changes of commit e16dcdb7dc - Show all commits

View File

@@ -173,6 +173,11 @@ public class SenderModelService {
* Promotes the oldest QUEUED sender run to RUNNING and triggers its training.
* Called in the finally block of triggerSenderTraining, creating a sequential chain:
* each run promotes the next only after it fully completes (success or failure).
*
* This is intentionally tail-recursive via the @Async thread: the same thread holds the
* full queue drain, serialising all sender training runs naturally without an external
* scheduler. With N queued runs the thread stays occupied for N sequential trainings —
* acceptable because the @Async executor is dedicated to long-running background work.
*/
private void promoteNextQueuedRun() {
Optional<OcrTrainingRun> queuedOpt = txTemplate.execute(status ->