That Day, Why Did Our Batch Job Stop?

A specific Batch job at Oliveyoung failed to terminate and hung. Initially, a DB Deadlock was suspected, but log review confirmed that the process had stopped after a specific data update.

Root cause analysis revealed that the @Transactional setting was missing on a service method called within the Spring Batch Tasklet. As a result, Spring's TransactionSynchronizationManager created an independent transaction for each thread via ThreadLocal, ultimately producing an unintended effect similar to Propagation.REQUIRES_NEW.

In addition, the RejectedExecutionHandler configuration and the structure of the work queue used when processing tasks in parallel with a multithreaded ThreadPoolExecutor turned out to be one of the core causes of the failure.