I'm trying to found out how hibernate deals with this situation.

2 + tomcat servers exporting the same interface via SOAP behind a load balanced switch.

First SOAP request comes in the row is pulled from the DB and a single column is changed, say column A1. The second SOAP request hits a different machine but pulls the same row and changes a different column, B1. They both commit but one the first update is lost.

Original row data is
A1 = 0
B1 = 1

First request gets the row, makes an object, sets the property, and commits it back
A1 = 1
B1 = 1

Second request is waiting on the lock, gets it and does an update with with original A1 val and its new B1 val.

A1 =0
B1 = 2

What we should have is A1=1 and B1= 2. I could write JDBC queries for this and just update the specific column. I can't really do too much locking because there are also thousands of simultaneous SELECT requests happening per second on the same data.

-Matt

You need to look at optimistic locking as this will provide the performance for reads (not locking the tables) and it will detect (and report) an update that would be an issue. See version tag in hibernate docs.