I have a 3-tier architecture that looks roughly like this:
Client -> Business -> Data
Where should transactions ideally start?
One school of thought says that transactions should only start at the top of the Data layer. The Business layer only manipulates business objects with business logic, and never knows about transactions. The business does all of its work to manipulate objects, and then hands them to the Data layer to be persisted. It's a somewhat RESTful philosophy applied to lower layers.
Another school of thought says that transactions should start at the top of the Business layer. The Business layer defines logical units of work, not the data layer, because a logical unit of work sometimes contains business logic, not just data logic.
I do like the idea of pushing transaction concerns as low as possible. But I also find it can require extra effort and design challenges to try and keep business logic out of the data layer, unless it's just CRUD operations. If you apply RESTful design patterns with a sledgehammer, you can make it so that your applications have very few non-CRUD operations.
There is even a 3rd school of thought that says that the Client could start transactions so that it can combine multiple business operations when it needs to. But now the Client is defining the unit-of-work? Isn't that a business concern?
A 4th school of thought says that my Clients can be just SOA components that could participate in an XA transaction started even outside the client!!
Our developers would like some standards more concrete than just "Start transactions wherever you feel like"
Does anyone have any opinions or suggestions on this subject?
Thanks!
In the
Client -> Business -> Data
architecture, it is always better to define the transaction on the business layer. I would suggest that the transaction be so defined that the business service either starts a new transaction or participates in the existing transaction if one is already started. This takes care of cases where a business service is invoked by another business service.
Having the transaction boundary at the data layer fails if the business layer make multiple data layer calls as part of the same request, as
client1-> business1 => data1 , business1 => data2
Transaction is a business concept and it should be coordinated from within the Business Tier.
Manipulating objects in isolation usually makes little benefit and spanning the manipulation between multiple types of objects already is a transaction. So first school of thought is dealing with really basic cases.
When your Business Tier is handling transactions it doesn't really matter who starts the transaction: client or other service. Also long running (distributed) transactions can only be supported when Business Tier is aware of them.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With