Question

I have a table structure like this (vertical design). I can have unlimited number of attributes (eg: city, phone etc.) for each user.

Table: tbl_UserAttributes

┌────────┬───────────┬────────────┐
| UserID │ FieldName │ Value      |
├────────┼───────────┼────────────┤
│ 341    │ city      │ MyCity1    │
│ 772    │ phone     │ 1234567890 │
│ 033    │ city      │ MyCity2    │
│ 044    │ sex       │ M          │
│ 772    │ firstname │ MyName     │
│ ---    │ ---       │ ---        │
└────────┴───────────┴────────────┘

I have to implement a search feature which should output rows which we apply query like this for a horizontally designed table:

SELECT 
    FieldName 
FROM 
    tbl_UserAttributes 
WHERE 
    city='%Mumbai%' AND 
    sex='M' AND ...

Please dont ask me to change the database design.

UPDATE: At present, I have a JOIN solution in place which is very slow and it hangs the server some times. Any alternate methods?

Answer 1

EAV table is a good thing as long as you don’t need to search for multiple values at once in which case it becomes a bad thing.

You cannot index several values at once because they are located in different records.

In an SQL Server table you could create an indexed view over multiple values and use it for the searches.

In Oracle, you could cluster the table by UserID which would keep all records with the same UserID within one data page which would use an index on the most selective value and quickly scan for the other values.

In PostgreSQL, you could store all value in a single array and index it with a GIN index.

In MySQL, you can do neither of this.

Here’s a query which will return the values:

SELECT  *
FROM    tbl_UserAttributes tcity
JOIN    tbl_UserAttributes tsex
ON      tsex.userid = tcity.userid
WHERE   tcity.fieldname = 'city'
        AND tcity.value LIKE '%Mumbai%'
        AND tsex.fieldname = 'sex'
        AND tsex.value = 'M'

but don’t expect it to be very fast.

Update:

If you needed to have the exact matching, you could create a composite index on (fieldname, value, userid), put the most selective fieldname into the first table and use STRAIGHT_JOIN to force the order:

SELECT  *
FROM    tbl_UserAttributes tcity
STRAIGHT_JOIN
        tbl_UserAttributes tsex
ON      tsex.userid = tcity.userid
WHERE   tcity.fieldname = 'city'
        AND tcity.value = 'Mumbai'
        AND tsex.fieldname = 'sex'
        AND tsex.value = 'M'

However, this won’t help with your current query, since you are looking for a wildcard match in which case the indexes are not very helpful. And your second table won’t benefit much from the index unless you are querying a maternity hospital database.

Still it will save you some time since the index scan can be used instead of a table scan.