I am working on project, where a smaller datamart (maybe 30 tables) was implemented from the scratch. Now, the colleague with deep knowledge of this mart is going to do another project, left me alone in this project (with some support from him).
I just thought that I generate database diagram and thus I will not need any extra help when I would modify an ETL and do some joins etc. But, surprisingly for me, newbie in this profession…
I generated diagram, and no star or snowflake schema was there, just plain tables without primary and foreign keys. So My work was to try to imagine how the tables are related, then consult it if it is true, then remake my script etc. ANNOYING.
When I ask why it is like this (without relations between tables), I got this answer: “It is because of performance.”
Is it usual to solve it like this? If no, how to solve it with relations and still good performance?
Foreign keys are constraints used to ensure consistency of data in a database – their purpose is not to document the structure of your database, rather it is to enforce data consistency rules by controlling what changes are allowed to the database.
This is all good in a live database where data integrity is key, but in a datamart there is no need to enforce these rules – we know the data is consistent because it’s a copy / extract of the live database where these rules are enforced.
Foreign keys also come with some disadvantages:
In short, they reduce performance and provide no real benefit, so why bother? Just make sure that your datamart is documented properly elsewhere.
You might be interested in these questions: