Unanswered: Transactions and the "lost update" problem
I am reading several books about the theory of databases and SQL Server. In the issues of concurrency problems they provide the transaction solution and the isolation levels. One problem is the "lost updates". There is a big mistake with this problem. They leave to be meant that if we have full serializable transactions then we haven't "lost updates". IT IS WRONG. This specific problem occurs even if we have full isolation and locking. For example supposed we have two transactions A and B. A is executed first (if we have full serialization) and make some updates. B is executed after the execution of A and makes some updates on the same data. The result is that the modifications madden by A disappeared without A knows this. This problem is very obvious when we have a web base application when the user A and B have the same record on its own screens. If A changes the telephone field and B changes the fax field then A will see that his change didn't happen and it was overwritten by B.