As I understand postgres does not reserve or use space for columns having no content. If that's so then the bloat will be insignificant . But off course the pros and cons of embedding the columns inline have to be weighed
Wouldn’t page size performance be decreased with a very wide table? Unless the columns are tight. If you are on a VM subsystem, that page bloat will be over the network from the memory controller.
maybe unorthodox but if its certain to be 1:1 then why not flatten it to a single table with all the possible columns (tblMain+tblOne+tblTwo+tblThree). You can keep a flag to indicate the type. there may be a gain in simplicity without losing anything.
I have one table that can have relations to only 1 of 3 possible tables. For example: tblMain, tblOne, tblTwo and tblThree.
I will always have 1 record in tblMain but each record in this table will be related to one record in tblOne OR one record in tblTwo OR one record in tblThree.
The relation between tblMain and tblOne is 1:1.
The relation between tblMain and tblTwo is 1:1.
The relation between tblMain and tblThree is 1:1.
Is it better to set tblMain as parent or child?
This can be seen as if tblOne, tblTwo and tblThree extend tblMain depending on a specific criteria.
Let's say that tableMain has a string field called "type" with the following possible values: "residential", "industrial" and "energy".
I will always insert a record in tblMain but:
* If type is "residential" then a record in tblOne is created and it is associated to tblMain,
* If type is "industrial" a record in tblTwo is created and it is associated to tblMain
* If type is "energy" a record in tblThree is created and it is associated to tblMain.
I am not sure how to design a case like this. I will very much appreciate your feedback.
