Step executions not saved properly in production job with Spring Batch 5.2.x (works in 5.1.3)

[zenith7781]

Bug description

This issue is the same as discussion #4683, but with additional details and version isolation. Currently I am on spring boot 3.5.6 and the issue occurs only with Spring Batch 5.2.x (e.g., 5.2.3). When using Spring Batch 5.1.3 (explicitly set via <spring-batch.version> with Spring Boot 3.5.6), the problem does not occur. By default, Spring Boot 3.5.6 pulled Spring Batch 5.2.3, which triggers the failure. This confirms the issue is related to changes in Spring Batch after 5.1.3.

Additional debugging details

Through debugging, I found that the class org.springframework.batch.core.repository.dao.JdbcStepExecutionDao fails to properly save a particular step to the database which was being routed through the method public void saveStepExecutions(final Collection<StepExecution> stepExecutions).
The count variable on line 284 of JdbcStepExecutionDao is set to 0 for the faulty step, and after several more method calls, the following error occurs:

org.springframework.dao.EmptyResultDataAccessException: Incorrect result size: expected 1, actual 0

The faulty step eventually gets saved with version 0, but it is expected to be 3 (as seen in a recreated scenario for Spring Batch 5.1.3).
Other steps that were saved via public void saveStepExecution(StepExecution stepExecution) worked fine.

Database and environment

• Database: PostgreSQL 42.7.7
• Backend: Spring Boot 3.5.6

Environment

• Spring Batch: 5.2.x (confirmed failing)
• Spring Batch: 5.1.3 (confirmed working)
• Spring Boot: 3.5.6

Expected behavior
Job should not fail with EmptyResultDataAccessException when updating step status; should behave as in 5.1.3.

Please let me know if any further specific information is required.

[zenith7781]

Hi @fmbenhassine , this issue is blocking our planned release. Any help will be gladly appreciated, thanks

[fmbenhassine]

Thank you for reporting this. The discussion you linked to seems to use DoInTransaction basically runs the code in a @Transactional method., which could be the issue. Are you in the same case? Without an example, it's really hard to conclude if it's a bug in Spring Batch or a misuse (it is known that using @Transactional methods with Spring Batch can cause issues).

Please let me know if any further specific information is required.

Yes please. Since you use PostgreSQL, can you please provide a test with a Docker-based setup that reproduces the issue? We have an integration test here that you can start with. If you share your use case, I will dig in further. Thank you.