marcel/familienarchiv
1
0
Fork 0
Code Issues 114 Pull Requests 3 Actions Packages Projects Releases Wiki Activity

fix(ocr): eliminate race window in runOrQueueSenderTraining by creating RUNNING row atomically

Browse Source 
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Marcel
2026-04-17 20:11:56 +02:00
parent 8d041a377d
commit 1b34a36a77
3 changed files with 49 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
backend/src/main/java/org/raddatz/familienarchiv/repository/OcrTrainingRunRepository.java
View File

@@ -14,6 +14,8 @@ public interface OcrTrainingRunRepository extends JpaRepository<OcrTrainingRun,
Optional<OcrTrainingRun> findFirstByStatusOrderByCreatedAtAsc(TrainingStatus status);
Optional<OcrTrainingRun> findFirstByPersonIdAndStatus(UUID personId, TrainingStatus status);
boolean existsByPersonIdAndStatus(UUID personId, TrainingStatus status);
List<OcrTrainingRun> findTop20ByOrderByCreatedAtDesc();

35
backend/src/main/java/org/raddatz/familienarchiv/service/SenderModelService.java
View File

@@ -72,9 +72,10 @@ public class SenderModelService {
}
/**
* Inserts a QUEUED row when training is already running (coalescing: at most one QUEUED per
* person_id). Runs immediately if idle. Returns true if training should start now, false if
* queued or skipped.
* Atomically checks the queue state and either creates a RUNNING row (returns true) or a
* QUEUED row (returns false). All three operations — idle check, duplicate-queue guard, and
* RUNNING row creation — happen in one transaction, eliminating the race window that would
* otherwise exist between the check and a separate RUNNING row creation.
*/
@Transactional
public boolean runOrQueueSenderTraining(UUID personId, int correctedLines) {
@@ -96,23 +97,29 @@ public class SenderModelService {
return false;
}
long blockCount = blockRepository.countManualKurrentBlocksByPerson(personId);
trainingRunRepository.save(OcrTrainingRun.builder()
.status(TrainingStatus.RUNNING)
.personId(personId)
.blockCount((int) blockCount)
.documentCount(0)
.modelName("sender_" + personId)
.build());
return true;
}
/** Executes sender training synchronously. Caller must run this on a background thread. */
/**
* Executes sender training synchronously. Caller must run this on a background thread.
* The RUNNING row is expected to already exist — created atomically by
* runOrQueueSenderTraining (for new runs) or by promoteNextQueuedRun (for promoted runs).
*/
public void triggerSenderTraining(UUID personId, int correctedLines) {
String outputModelPath = "/app/models/sender_" + personId + ".mlmodel";
OcrTrainingRun run = Objects.requireNonNull(txTemplate.execute(status -> {
long blockCount = blockRepository.countManualKurrentBlocksByPerson(personId);
return trainingRunRepository.save(OcrTrainingRun.builder()
.status(TrainingStatus.RUNNING)
.personId(personId)
.blockCount((int) blockCount)
.documentCount(0)
.modelName("sender_" + personId)
.build());
}));
OcrTrainingRun run = Objects.requireNonNull(txTemplate.execute(status ->
trainingRunRepository.findFirstByPersonIdAndStatus(personId, TrainingStatus.RUNNING)
.orElseThrow(() -> new IllegalStateException(
"Expected RUNNING row for person " + personId + " but none found"))));
String runId = run.getId().toString();
MDC.put("trainingRunId", runId);