问题
I am developing a stock control system for a construction project. The storeman is responsible for adding new stock and distributing/returning it to/from employees. The items (and hence their attributes) will be very varied; e.g. steelwork, clothing, plant/machinery, tools, etc.
My question is whether to go for a Class/Concrete Table Inheritance or EAV based schema. I don't know what attributes items will have, so presumably a Class/Concrete Table Inheritance approach would require the end user to add new tables to the database (is this possible?).
Please see my proposed EAV schema. Would this be a reasonable solution? I think I'm right in saying that to identify an item of stock, it would be necessary to perform multiple self joins in a query on the 'EV' table?
N.B. Developing using PHP, MySQL and Zend Framework.
回答1:
If attribute changes are few and far between, go for Table Inheritance, but the changes to the schema should be done by yourself or a DBA. Programmatically modifying your schema based on user input seems like a bad idea.
If attribute changes are fairly common, consider using a document-oriented database like MongoDB or CouchDB.
If attribute changes are common and you are restricted to relational databases, go with EAV.
回答2:
I'd avoid EAV unless I'm building a medical repository with thousands of hundreds of products. Class Table Inheritance is a proper solution and altering schema at run-time isn't that bad idea.
Assume you're building an online shop were you have different products with different attributes. Usually a product is using a set of attributes. By using Class Table Inheritance pattern you can have one table per product set inheriting a common table with common product attributes. When a new attribute is needed you have two options:
- alter the set table and add a new column with the attribute name
- create a new set (i.e new table) with the new column and set the current product set to inherit the new one, thus adding the new attribute.
Altering table at run-time may indeed cause problems due to locking occurring during alter. However as of MySQL 5.6 a programmer can specify a LOCK type i.e LOCK=NONE and also ALTER ONLINE|OFFLINE. Obviously DB vendors are working in this direction to allow for database alternations during run-time.
Table Locking issues can also be avoided/minimized. Lock issues usually occur when altering huge tables i.e 100k+ records. However inheritance allows for distributing products in different sets and thus having a relatively small amount of records per table.
For instance if we have 10 different types of products, we have 10 different tables. Assume we have a 10 000 products per type, this means that we have over 100 000 products, however adding a new attribute in a set will be in effect adding a new column in a table with 10 000 rows only. Adding a new column in a table with 10k rows will take less then a second, so is altering database schema at run-time when dealing with Class Table Inheritance really bad idea?
I'd be happy to hear more opinions here on this one. Thanks guys.
来源：https://stackoverflow.com/questions/3410362/design-of-an-eav-or-class-concrete-table-inheritance-database-for-stock-cont