Question

Here is the query:

SELECT name, SUM(  `count` ) AS Total
FROM  `identdb` 
WHERE MBRCONTAINS( GEOMFROMTEXT(  'LineString(34.4 -119.9, 34.5 -119.8)' ) , latlng ) 
AND MOD( DAYOFYEAR( CURDATE( ) ) - DAYOFYEAR(  `date` ) +365, 365 ) <=14
OR MOD( DAYOFYEAR(  `date` ) - DAYOFYEAR( CURDATE( ) ) +365, 365 ) <=14
AND MBRCONTAINS( GEOMFROMTEXT(  'LineString(34.4 -119.9, 34.5 -119.8)' ) , latlng ) 
GROUP BY  `name`

It essentially finds any rows where the day of year is plus or minus 14 of today’s day, and rows that the latlng spatial column is in the rectangle.

Here is what my database looks like:

#   Column  Type        Collation   
1   name    varchar(66) utf8_general_ci 
2   count   tinyint(3)
3   date    date    
4   latlng  geometry
5   lat1    varchar(15) latin1_swedish_ci
6   long1   varchar(15) latin1_swedish_ci

Keyname Type    Unique  Packed  Column  Cardinality Collation   Null    Comment
PRIMARY BTREE   Yes No  name    0   A       
                        count   0   A   
                        date    0   A   
                        lat1    0   A   
                        long1   6976936 A   
sp_index SPATIAL    No  No  latlng (32) 0   A

There are 7 million records and the query is taking about 7 seconds. I have no clue how to speed this up, thanks in advance!

EXPLAIN:

id  select_type table   type    possible_keys   key     key_len ref  rows       Extra
1   SIMPLE      identdb ALL     sp_index        NULL    NULL    NULL 6976936    Using where; Using temporary; Using filesort

UPDATED explanation of query:
I believe MBRCONTAINS creates a rectangle where I can compare whether the latlng spatial point is inside or not. The date part is finding dayofyear + or – 14 days. It is using modular arithmetic so that it won’t mess up around the new years. I had to put the MBRCONTAINS part in twice because of the use of OR.

My needs of the query are to find find all names that have a day of the year + or – 14 days, and are within the given lat/long pairs, and then total the counts for each.

I’m dumb at this stuff so please correct me if I’m doing something dumb. Thanks guys!

Answer 1

Rewrite it so that your calculations happen once per query, rather than once per row by expressing your predicates such that the column is not part of the calculation.

For example, this expression:

MOD( DAYOFYEAR( CURDATE( ) ) - DAYOFYEAR(  `date` ) +365, 365 ) <= 14

which requires 7 millions calculations on date, can be expressed as

`date` between SUBDATE( CURDATE( ), 14) and ADDDATE( CURDATE( ), 14)

which requires only 1 calculation and further would allow an index on the date column to be used.
That change alone will speed up your query.

if you don’t have an index on date, put one and your query will fly:

create index mytable_date on mytable(`date`);

I don’t know what MBRCONTAINS does, but try to refactor it too so that the column value is not part of the calculation.