Does it make sense to use an index that will have a low cardinality?

I'm mainly an Actionscript developer and by no means an expert in SQL, but from time to time I have to develop simple server side stuff. So, I thought I'd ask more experienced people about the question in the title.

My understanding is that you don't gain much by setting an index in a column that will hold few distinct values. I have a column that holds a boolean value (actually it's a small int, but I'm using it as a flag), and this column is used in the WHERE clauses of most of the queries I have. In a theoretical "average" case, half of the records' values will be 1 and the other half, 0. So, in this scenario, the database engine could avoid a full table scan, but will have to read a lot of rows anyway (total rows/2).

So, should I make this column an index?

For the record, I'm using Mysql 5, but I'm more interested in a general rationale on why it does / does not make sense indexing a column that I know that will have a low cardinality.

Thanks in advance.

Solution 1:

An index can help even on low cardinality fields if:

1. When one of possible values is very infrequent compared to the other values and you search for it.

   For instance, there are very few color blind women, so this query:

   SELECT  *
   FROM    color_blind_people
   WHERE   gender = 'F'
   

   would most probably benefit from an index on gender.

2. When the values tend to be grouped in the table order:

   SELECT  *
   FROM    records_from_2008
   WHERE   year = 2010
   LIMIT 1
   

   Though there are only 3 distinct years here, records with earlier years are most probably added first so very many records would have to be scanned prior to returning the first 2010 record if not for the index.

3. When you need ORDER BY / LIMIT:

   SELECT  *
   FROM    people
   ORDER BY
           gender, id
   LIMIT 1
   

   Without the index, a filesort would be required. Though it's somewhat optimized do to the LIMIT, it would still need a full table scan.

4. When the index covers all fields used in the query:

   CREATE INDEX (low_cardinality_record, value)
   
   SELECT  SUM(value)
   FROM    mytable
   WHERE   low_cardinality_record = 3
   

5. When you need DISTINCT:

   SELECT  DISTINCT color
   FROM    tshirts
   

   MySQL will use INDEX FOR GROUP-BY, and if you have few colors, this query will be instant even with millions of records.

   This is an example of a scenario when the index on a low cardinality field is more efficient than that on a high cardinality field.

Note that if DML performance is not much on an issue, then it's safe to create the index.

If optimizer thinks that the index is inefficient, the index just will not be used.

Solution 2:

It might be worth including the boolean field in a composite index. For example if you have a large table of messages which typically need to be ordered by Date but you also have a boolean Deleted field, so you often query it like this:

SELECT ... FROM Messages WHERE Deleted = 0 AND Date BETWEEN @start AND @end

You will definitely benefit from having a composite index on the Deleted and Date fields.

Solution 3:

I usually do a simple "have index" vs "don't have" index test. In my experience you get most of the performance on queries that use ORDER BY the indexed column. In case you have any sorting on that column, indexing will most likely help.

Solution 4:

When half of the records' values will be 1 and the other half 0, no point of putting an index on that column. The query optimizer is likely not to make use of it.

Typically, however, you have a small set of "active" records and an increasingly larger set of "inactive". For example in a bug tracking system, you care about active bugs and hardly every look at the completed and archived ones. For such a case, the trick is to use "dateInactivated" column that stores the timestamp of when the record is inactivated/deleted. As the name implies, the value is NULL while the record is active, but once inactivated, write in the system datetime. Thus, an index on that column ends up having high selectivity as the number of "deleted" records grows since each record will have a unique (not strictly speaking) value. The query would have

"... AND dateInactivated is NULL ..." 

as part of the predicate and the index will pull in just the right set of rows that you care about.