Distributed sessions: Keep track of modified attributes using a hashcode

[cassand]

Current Behavior:

Each session keep track of the accessed attributes and only saves to the Datastore those attributes.

New behavior:

The session would keep track of the modified attributes using a hashcode.

• The hashcode could be stored in the same Datastore entity as the attribute.
• The hashcode would be computed during the serialization of the attributes.
• Only the attributes with a hashcode which differs from the loaded hashcode would be sent to the Datastore.

[meltsufin]

There is a potential anomaly that can happen with the new behavior.

Lets say you have two bank accounts A and B, each of which has $100, and the bank requires the combined balance A+B to be greater than zero. Then you initiate two concurrent requests to withdraw $100 from A & B. Each request starts by reading (A = 100, B = 100), and after subtracting 100 from A or B respectively, verifies that the combined balance is greater than 0 before writing back to the database. However, after both transactions write the result back (A = 0, [B=100]) and ([A=100], B=0), the state in the database is (A = 0, B = 0), with the combined balance of $0 in violation of the requirement.

With the current behavior, you would end up with either (A = 100, B = 0) or (A = 0, B = 100), which would not violate the constraint, but would lose one of the withdrawals.

That being said, I'm not sure which is better.