During validation, the processedPathsPerBean always consumes more memory, regardless of whether there are actually valid constraints. In this case, hibernate occupies a large amount of memory. In the concurrent scenario, when the object size is large, OOM occurs. I think this is a bug, and some strategy should be taken to constrain the size of processedPathsPerBean, such as only validating if there is an actual constraint.

According to the analysis, traversableResolver and failingConstraintViolations occupy a large amount of memory during the verification process, but they can be used.
addProperty("hibernate.validator.enable_traversable_resolver_result_cache", "false") and addProperty("hibernate.validator.fail_fast", "true") to prevent them from consuming memory. But there is no way to reduce the memory footprint of processedPathsPerBean.

This is the result of one of our hprof memory analyses.

The demo code that is causing the problem is this, and run with -Xmx5M