Sql query how to make it faster? - mysql

Question

I have an SQL query. Is it possible to change somehow this query, so it has better performance, but with the same result? This query is working, but it is very slow, and I don't have an idea on improving its performance.
SELECT keyword, query
FROM url_alias ua
JOIN product p
on (p.manufacturer_id =
CONVERT(SUBSTRING_INDEX(ua.query,'=',-1),UNSIGNED INTEGER))
JOIN oc_product_to_storebk ps
on (p.product_id = ps.product_id)
AND ua.query LIKE 'manufacturer_id=%'
AND ps.store_id= '9'
GROUP BY ua.keyword
Table structure:
URL_ALIAS
+-----------------------------------------------+
| url_alias_id | query | keyword |
+--------------+---------------------+----------+
| 1 | manufacturer_id=100 | test |
+--------------+---------------------+----------+
PRODUCT
+-----------------+------------+
| manufacturer_id | product_id |
+-----------------+------------+
| 100 | 1000 |
+-----------------+------------+
OC_PRODUCT_TO_STOREBK
+------------+----------+
| product_id | store_id |
+------------+----------+
| 1000 | 9 |
+------------+----------+
I want all the keywords from the url_alias keyword column, when the following condition is met: LIKE 'manufacturer_id=%' AND ps.store_id='9'

Answer 1

You should avoid the convert function as it will be expensive and provides no way you could profit from indexes on the url_alias table.
Extend your url_alias table so it has additional fields for the parts of the query. You will probably hesitate to go this way, but you will not regret it once you have done it. So your url_alias table should look like this:
create table url_alias (
url_alias_id int,
query varchar(200),
keyword varchar(100),
query_key varchar(200),
query_value_str varchar(200),
query_value_int int
);
If you don't want to recreate it, then add the fields as follows:
alter table url_alias add (
query_key varchar(200),
query_value_str varchar(200),
query_value_int int
);
Update these new columns for the existing records with this statement (only to execute once):
update url_alias
set query_key = substring_index(query, '=', 1),
query_value_str = substring_index(query, '=', -1),
query_value_int = nullif(
convert(substring_index(query,'=',-1),unsigned integer), 0);
Then create a trigger so that these 3 extra fields are updated automatically when you insert a new record:
create trigger ins_sum before insert on url_alias
for each row
set new.query_key = substring_index(new.query, '=', 1),
new.query_value_str = substring_index(new.query, '=', -1),
new.query_value_int = nullif(
convert(substring_index(new.query,'=',-1),unsigned integer), 0);
Note the additional nullif() which will make sure the last field is null when the value after the equal sign is not numerical.
If ever you also update such records, then also create a similar update trigger.
With this set-up, you can still insert records like before:
insert into url_alias (url_alias_id, query, keyword)
values (1, 'manufacturer_id=100', 'test');
When you then select this record, you will see this:
+--------------+---------------------+---------+-----------------+-----------------+-----------------+
| url_alias_id | query | keyword | query_key | query_value_str | query_value_int |
+--------------+---------------------+---------+-----------------+-----------------+-----------------+
| 1 | manufacturer_id=100 | test | manufacturer_id | 100 | 100 |
+--------------+---------------------+---------+-----------------+-----------------+-----------------+
Now the work of extraction and conversion has been done once, and does not have to be repeated any more when you select records. You can rewrite your original query like this:
select ua.keyword, ua.query
from url_alias ua
join product p
on p.manufacturer_id = ua.query_value_int
join oc_product_to_storebk ps
on p.product_id = ps.product_id
and ua.query_key = 'manufacturer_id'
and ps.store_id = 9
group by ua.keyword, ua.query
And now you can improve the performance by creating indexes on both elements of the query:
create index query_key on url_alias(query_key, query_value_int, keyword);
You might need to experiment a bit to get the order of fields right in the index before it gets used by the SQL plan.
See this SQL fiddle.

Answer 2

I asume that you use indexes on the store_id, product_id and keyword columns?
Focus on changing your datamodel to avoid the CONVERT and the LIKE operators. Both of them will cause that the query will not utilize indexes on relevant columns
Also, take a good look on the data that is stored in the ua.query colomn. Possibly you need to distribute data in that column to multiple columns so you can use indexes