Invoking multiple cache operations in parallel in the same transaction is inherently dangerous, especially when they touch the same keys (e.g. if one of them is a bulk operation).

Invoking multiple cache operations in parallel became much more easier when we added asynchronous operations, and while we "know" it's not ok to access the same transaction in parallel, there is no explicit guard against it.

AbstractCacheTransaction.lookedUpEntries is a HashMap, which will sometimes throw a ConcurrentModificationException, but most of the time the exception is something completely unrelated, making the problem even harder to trace back to its cause (e.g. ISPN-10239).