The recently added Exchange#lock method has a flaw that was exploited by Akiban Server Data Loading Test. It is possible for transactions inserting locks with sequentially varying keys to accumulate in the lock volume in such a way that pruning can't keep up. The result is that the lock volume, which is intended to be a very small in-memory structure, instead exceeds 2GB in size. To fix this will require some additional logic in the pruning code.
The recently added Exchange#lock method has a flaw that was exploited by Akiban Server Data Loading Test. It is possible for transactions inserting locks with sequentially varying keys to accumulate in the lock volume in such a way that pruning can't keep up. The result is that the lock volume, which is intended to be a very small in-memory structure, instead exceeds 2GB in size. To fix this will require some additional logic in the pruning code.