在SQL数据库中存储非关系整数列表

Question

First, to note, I have read several articles that all say storing a list of values in a single column in SQL is a bad idea and breaks all design protocols. In fact, they all say to redesign the table so that IT IS relational. So I'm not looking for the easy-out solution here, but the correct one.

Here's the problem, I have two variables:

1) the unique userId that relates to other tables and

2) a simple Hashset of integers that relate to items in a JSON file.

Thus, these number do not relate to any other table in my SQL database. The max value in this list will likely not go over 1000, but who knows. Also the list could be out of order or skip multiple values in-between. I will never query the numbers, but I will load them when the user logs in and re-save them when the user logs out.

The options I have read are a comma separated value column, an xml, or a lookup table (which in this case I don't know what I'm looking up to and with a 1000 numbers, there could be 1E-249 permutations).

Therefore, I ask what would be the correct way to save this list of integers.

Answer 1

Using normalization you can create a table relationship that stores nested data as follows:

So following DB normalization, you would use 3 tables;

• User table

  • Id (int)
  • UserId (int)
  • Username (varchar)

• Json File Table

  • Id (int)
  • UserId (varchar)
• Json File Item Table
  • Id (int)
  • JsonFileId (int)
  • Value (int)

Then from this structure, you can store the following:

• User #1
  • Json File #1
    • Json File Value #1
    • Json File Value #2
  • Json File #2
    • Json File Value #3
    • Json File Value #4

This essentially lets you have 'x' users to 'x' json files which each have 'x' values.

In regards to you 'saving and loading' the data, you wouldnt want to load them all and re save them all on login and log out respectivly as this would be a waste of resources. You would track the changes and save the delta between the two. Ie you login, load the data, the user may change value #2 and thus you only update record number 2.

Answer 2

Given that you do not need to preserve order and do not need to preserve duplicates, if you wish to have your database in first normal form, you can easily store this in a two-column table in any database:

create table MyTable (
  UserId int not null,
  Value int not null,
  primary key (UserId, Value) );

If the user with ID 1 holds values [1, 2, 8, 33, 999], and the user with ID 2 holds values [3, 4], you store that as 7 records:

UserId | Value
     1 | 1 
     1 | 2
     1 | 8
     1 | 33
     1 | 999
     2 | 3
     2 | 4

This is similar to your lookup table idea, except you don't have to look up what the values mean, you can store the values directly in that table.

The benefit of this is that there is exactly one canonical representation for any set of numbers, and the database cannot hold anything other than a set of numbers. You do not have to worry in your code about the possibility of some record holding duplicate values that you would have to ignore in your application ( [1, 1] ), two users with identical values being represented differently ( [1, 2] vs. [2, 1] ), or users with completely invalid values ( [1, "abc"] or 1.23 ).

Another benefit is that this can be easily handled with 100% standard SQL. You do not need any non-standard extensions, so you can keep your code portable across DB implementations.

Practical concerns may suggest some other approach. Your suggestions of CSV or XML are valid. Another possibility is JSON. All three have native support in at least one major DB implementation, and although in most cases the table approach should be good enough, depending on the size of your data set and access patterns, it is possible that a denormalised database will allow for better performance. The code to read and update the values may be slightly easier to write as well.

This is a trade off you will need to make yourself. I hope you now have enough information to make it.