How to optimize slow query with many joins [closed]

Answer 1

To clarify this as an answer: if you do not already have these indexes, you should consider adding them

do you also have indexes on these:

vehicles.status
vehicles.date_from
vehicles.date_to
vehicles.type_id
vehicles.price

Answer 2

Improving the WHERE clause

Your EXPLAIN shows that MySQL is only utilizing one index (type_id) for selecting the rows that match the WHERE clause, even though you have multiple criteria in the clause.

To be able to utilize an index for all of the criteria in the WHERE clause, and to reduce the size of the result set as quickly as possible, add a multi-column index on the following columns on the vehicles table:

(status, date_from, date_to, type_id, price)

The columns should be in order of highest cardinality to least.

For example, vehicles.date_from is likely to have more distinct values than status, so put the date_from column before status, like this:

(date_from, date_to, price, type_id, status)

This should reduce the rows returned in the first part of the query execution, and should be demonstrated with a lower row count on the first line of the EXPLAIN result.

You will also notice that MySQL will use the multi-column index for the WHERE in the EXPLAIN result. If, by chance, it doesn't, you should hint or force the multi-column index.

Removing the unnecessary JOINs

It doesn't appear that you are using any fields in any of the joined tables, so remove the joins. This will remove all of the additional work of the query, and get you down to one, simple execution plan (one line in the EXPLAIN result).

Each JOINed table causes an additional lookup per row of the result set. So, if the WHERE clause selects 5,000 rows from vehicles, since you have 8 joins to vehicles, you will have 5,000 * 8 = 40,000 lookups. That's a lot to ask from your database server.

Answer 3

Instead of expensive calculation of precise distance for all of the rows use a bounding box and calculate the exact distance only for rows inside the box.

The simplest possible example is to calculate min/max longitude and latitude that interests you and add it to WHERE clause. This way the distance will be calculated only for a subset of rows.

WHERE
    vehicles.gps_lat > min_lat ANDd vehicles.gps_lat < max_lat AND
    vehicles.gps_lon > min_lon AND vehicles.gps_lon < max_lon

For more complex solutions see:

• MySQL spatial extensions
• How to use MySQL spatial extensions
• https://stackoverflow.com/a/5237509/342473

Answer 4

Is you SQL faster without this?

Round(3959 * Acos(Cos(Radians(51.465436)) *
  Cos(Radians(vehicles.gps_lat)) *
  Cos(Radians(vehicles.gps_lon) - 
  Radians(-0.296482)) + 
  Sin(Radians(51.465436)) * 
  Sin(Radians(vehicles.gps_lat)))) AS distance

performing math equation is very expensive

Maybe you should consider a materialized view that pre-calculate you distance, and you can select from that view. Depending on how dynamic you data is, you may not have to refresh you data too often.

Answer 5

To be a little more specific than @Randy of indexes, I believe his intention was to have a COMPOUND index to take advantage of your querying critieria... One index that is built on a MINIMUM of ...

( status, type_id, date_from )

but could be extended to include the date_to and price too, but don't know how much the index at that granular level might actually help

( status, type_id, date_from, date_to, price )

EDIT per Comments

You shouldn't need all those individual indexes... Yes, the Primary Key by itself. However, for the others, you should have compound indexes based on what your common querying criteria might be and remove the others... the engine might get confused on which might be best suited for the query. If you know you are always looking for a certain status, type and date (assuming vehicle searches), make that as one index. If the query is looking for such information, but also prices within that criteria it will already be very close on the few indexed records that qualify and fly through the price as just an extra criteria.

If you offer querying like Only Automatic vs Manual transmission regardless of year/make, then yes, that could be an index of its own. However, if you would TYPICALLY have some other "common" criteria, tack that on as a secondary that MAY be utilized in the query. Ex: if you look for Manual Transmissions that are 2-door vs 4-door, have your index on (transmission_id, category_id).

Again, you want whatever will help narrow down the field of criteria based on some "minimum" condition. If you tack on an extra column to the index that might "commonly" be applied, that should only help the performance.