In all three layers. However, what's important is the frequent fact that validation requirements (=the actual data constraints) differ from layer to layer. This because of the differing context and designed system boundaries.
In your example, the validation may be as follows:
- Tier 1: Presentation layer — validate that the name has been entered, i.e. a text box in the user interface is not empty, and has a maximum of 100 characters.
- Tier 2: Business logic layer - validate as above, plus it consists of at least two tokens separated by a white space, plus first and last names are a real first and last names (chek against some name database).
- Tier 3: Data layer - database integrity constraint that the respective field is not empty and has a maximum length of 100 characters.
The result is, that from a database perspective you check for a reasonable amount of constraints to keep the system consistent, but don't assume what's a matter of a higher-order logic. In fact, you don't limit future changes unnecessarily. From the business logic perspective, there's the complete set of constraints enforced. And, finally, from the presentation logic perspective, you don't overvalidate: only the simple validations to reduce unnecessary traffic to the business logic are performed, possibly preventing exceptions from the business logic layer, not duplicating anything more complex.
As a rule of thumb, it's always best practice to provide detailed validations at the facade of the business logic layer. That's the place where the (potentially untrusted) presentation layer and/or 3rd party (which may be just another corporate system) API consumers connect.
Further, some specific comments to the options you outlined in your question:
Option 1 : Unique key constraint in the database
Not only. It would work from the data correctness point of view, but by relying solely on database constraints, you are loosing semantics and would struggle providing a human understandable error message. Further, every bad input would travel down to the database, opening a potential hole for DoS attacks that could harm the whole technology stack.
Option 2 : Checking condition in the Data Service class in the business layer in order to keep the business logic encapsulated in this layer regardless of the data layer being used?
Yes, see above. However, don't compromise security, performance, and user experience by avoiding at least the basic, simple-to-evaluate validations in the presentation layer.
