What index should I use when using JOIN on PRIMARY KEY

问题

I'm trying to optimise the following MySQL query

SELECT Hotel.HotelId, Hotel.Name, Hotel.Enabled, Hotel.IsClosed, 
HotelRoom.HotelId, HotelRoom.RoomId, HotelRoom.Name AS RoomName
FROM Hotel
INNER JOIN 
   HotelRoom ON Hotel.HotelId = HotelRoom.HotelId    
WHERE Hotel.IsClosed = 0
AND Hotel.Enabled = 1
AND HotelRoom.Deleted = 0
AND HotelRoom.Enabled = 1
AND IF(LENGTH(TRIM(sAuxiliaryIds)) > 0 AND sAuxiliaryIds IS NOT NULL, 
FIND_IN_SET(Hotel.AuxiliaryId, sAuxiliaryIds), 1=1) > 0 
ORDER BY Hotel.HotelId ASC, HotelRoom.RoomId ASC

The PRIMARY KEYS are Hotel.Hotel and HotelRoom.RoomId, and I've got a FOREIGN KEY from HotelRoom.HotelId to Hotel.HotelId.

Should I be creating a INDEX for (Hotel.IsClosed, Hotel.Enabled) and (HotelRoom.Deleted, HotelRoom.Enabled) which is used in the WHERE clause, and should this index include the PRIMARY key so for example I should create a INDEX for (Hotel.HotelId, Hotel.IsClosed, Hotel.Enabled)

EDIT 1 I've added the following in the WHERE statement AND IF(LENGTH(TRIM(sAuxiliaryIds)) > 0 AND sAuxiliaryIds IS NOT NULL, FIND_IN_SET(Hotel.AuxiliaryId, sAuxiliaryIds), 1=1) > 0 Should these also be included in INDEX

This is what the EXPLAIN statement is showing for this query

I added both INDEX suggestions but when I ran the EXPLAIN statement they both showed that no key was going to be used

回答1:

MySQL's Optimizer does not care which table comes first in a JOIN. It will look at statistics (etc) to decide for itself whether to start with Hotel or HotelRoom. So, you should write indexes for both cases, so as not to restrict the Optimizer.

MySQL almost always performs a JOIN by scanning one table. Then, for each row in that table, look up the necessary row(s) in the other table. See "Nested Loop Join" or "NLJ". This implies that the optimal indexes are (often) thus: For the 'first' table, columns of the WHERE clause involving the first table. For the second table, the columns from both the WHERE and ON clauses involving the second table.

Assuming that the Optimizer started with Hotel:

Hotel: INDEX(IsClosed, Enabled)   -- in either order
HotelRoom:  INDEX(Deleted, Enabled, HotelId)  -- in any order

If it started with HotelRoom:

HotelRoom:  INDEX(Deleted, Enabled)  -- in either order
Hotel: PRIMARY KEY(HotelId)  -- which you already have?

If there are a lot of closed/disabled hotels, then this may be beneficial:

Hotel: INDEX(IsClosed, Enabled, HotelId)

As Tim mentioned, it may be beneficial to augment an index to include the rest of the columns mentioned, thereby making the index "covering". (But don't do this with the PRIMARY KEY or any UNIQUE key.)

If you provide SHOW CREATE TABLE and the sizes of the tables, we might have further suggestions.

回答2:

There are two potential indexing strategies here, depending on which of the two tables appears on the left side of the inner join (either table could potentially appear on either side of the join). Given that the HotelRoom table likely contains many more records than the Hotel table, I would suggest placing the Hotel table on the left side of the join. This would imply that the Hotel table would be scanned, and the index used for the join to HotelRoom. Then, we can try using the following index on HotelRoom:

CREATE INDEX hotel_room_idx ON HotelRoom (HotelId, Deleted, Enabled, Name, RoomId);

This should speed up the join substantially, covers the WHERE clause, and also covers all columns in the select on HotelRoom. Note that the following simplified index might also be very effective:

CREATE INDEX hotel_room_idx ON HotelRoom (HotelId, Deleted, Enabled);

This just covers the join and WHERE clause, but MySQL might still choose to use it.

来源：https://stackoverflow.com/questions/57425866/what-index-should-i-use-when-using-join-on-primary-key