MySQL Multiple Case When Exists Statement - mysql

Question

I have two tables. Let's call it: SEATS and SEAT_ALLOCATION_RULE table.
Below are the table schema:
CREATE TABLE IF NOT EXISTS `SEATS` (
`SeatID` int(11) NOT NULL AUTO_INCREMENT,
`SeatName` varchar(255) NOT NULL DEFAULT '',
PRIMARY KEY (`SeatID`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1 AUTO_INCREMENT=9 ;
INSERT INTO `SEATS` (`SeatID`, `SeatName`) VALUES
(1, 'Super VIP'),
(2, 'VIP'),
(3, 'Business'),
(4, 'Economy'),
(5, 'Standing');
CREATE TABLE IF NOT EXISTS `SEAT_ALLOCATION_RULE` (
`SeatID` int(11) NOT NULL DEFAULT '0',
`Origin` varchar(50) NOT NULL DEFAULT '0',
`Destination` varchar(50) NOT NULL DEFAULT '',
`Passenger_Type` varchar(25) NOT NULL DEFAULT '',
PRIMARY KEY (`SeatID`,`Origin`,`Destination`,`Passenger_Type`)
) ENGINE=InnoDB DEFAULT CHARSET=latin1;
INSERT INTO `SEAT_ALLOCATION_RULE` (`SeatID`, `Origin`, `Destination, `Passenger_Type`) VALUES
(1, 'Malaysia','',''),
(2, 'Malaysia','Singapore',''),
(3, 'Malaysia','Singapore','Senior_Citizen'),
(4, 'Bangkok','Japan','Student'),
(5, 'Cambodia','China','Senior_Citizen');
SEAT_ALLOCATION_RULE table determines which seat should a passenger be assigned to based on the following order in priority:
1. Origin, destination, and passenger_type match
2. Origin and destination match
3. Origin match
It means that if all the fields (origin, destination, and passenger_type) match, it should take higher priority than if it is just two fields match and so on. If a column is empty, it is considered as unspecified and hence has lower priority. So, for example:
If the Origin is Malaysia, Destination is Singapore, and Passenger_Type is Senior_Citizen, it should return seatID 3
If the Origin is Malaysia, Destination is Singapore, and Passenger_Type is Student, it should return seatID 2 (since it only match Origin and Destination)
If the Origin is Malaysia, Destination is US, and Passenger_Type is Student, it should return seatID 1 (since it only match Origin).
Now, based on the rules above, if the origin is Malaysia, destination is Singapore, and Passenger_Type is student, the query to return seatID is as follow:
SELECT s.SeatID, s.SeatName
FROM SEATS s
WHERE
CASE WHEN EXISTS(
select 1
from SEAT_ALLOCATION_RULE r
where s.SeatID = r.SeatID
AND r.Origin = 'Malaysia'
AND r.Destination = 'Singapore'
AND r.Passenger_Type='Student') Then 1
WHEN EXISTS(
select 1
from SEAT_ALLOCATION_RULE r
where s.SeatID = r.SeatID
AND r.Origin = 'Malaysia'
AND r.Destination = 'Singapore'
AND r.Passenger_Type='') Then 1
WHEN EXISTS(
select 1
from SEAT_ALLOCATION_RULE r
where s.SeatID = r.SeatID
AND r.Origin = 'Malaysia'
AND r.Destination = ''
AND r.Passenger_Type='') Then 1 END
However, the query above does not work as it will return seatID 1 and 2, but the expected output is only seatID 2 (since origin and destination matches and it takes higher precedence). Can someone help to correct my SQL query?

Answer 1

This should do the trick:
select seatid
from seat_allocation_rule sar
order by ((sar.origin = :origin) << 2) + ((sar.destination = :destination) << 1) + (sar.passenger_type = :passenger_type) desc,
((sar.origin <> '') << 2) + ((sar.destination <> '') << 1) + (sar.passenger_type <> '') asc
limit 1
To understand how:
create table testcase (
origin varchar(255),
destination varchar(255),
passenger_type varchar(255),
expected_seat int(11)
);
insert into testcase values ('Malaysia','Singapore','Senior_Citizen',3),
('Malaysia','Singapore','Student',2),
('Malaysia','US','Student',1);
select * from (
select tc.*,
sar.seatid,
case when sar.seatid = tc.expected_seat then 'Y' else '-' end as pass,
((sar.origin = tc.origin) << 2)
+ ((sar.destination = tc.destination) << 1)
+ ((sar.passenger_type = tc.passenger_type) << 0) as score,
((sar.origin <> '') << 2)
+ ((sar.destination <> '') << 1)
+ ((sar.passenger_type <> '') << 0) as priority
from seat_allocation_rule sar
cross join testcase tc
) x order by expected_seat desc, score desc, priority asc;

Answer 2

This fixes the existing SQL:
SELECT DISTINCT s.SeatID, s.SeatName
FROM SEATS s
LEFT JOIN SEAT_ALLOCATION_RULE r ON r.SeatID = s.SeatID
AND r.Origin = 'Malaysia'
AND (
(r.Destination = 'Singapore' AND r.Passenger_Type IN ('Student', ''))
OR
(r.Destination = '' AND r.Passenger_Type = '')
)
WHERE r.SeatID IS NOT NULL
But it's only a partial solution, and it's hand-coding logic you really want to apply based solely on the data.
A complete solution will use hypothetical inputs for your passenger's ticket info to produce all eligible seats. This is a great use of lateral joins/apply, which are sadly lacking in MySql (all of their major competitors have had these for at least two release cycles, along with other gems that are absent from the current MySql release like windowing functions, ctes, full joins... I could go on). Here's how I'd do it in Sql Server:
SELECT p.PassengerID, s.SeatID, s.SeatName
FROM Passenger p
CROSS APPLY (
SELECT TOP 1 r.SeatID
FROM SEAT_ALLOCATION_RULE r
WHERE COALESCE(NULLIF(r.Origin, ''),p.Origin) = p.Origin
AND COALESCE(NULLIF(r.Destination,''), p.Destination) = p.Destination
AND COALESCE(NULLIF(r.Passenger_Type,''),p.Passenger_Type) = p.Passenger_Type
ORDER BY
CASE WHEN r.Origin <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Destination <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Passenger_Type <> '' THEN 1 ELSE 0 END DESC
) r
INNER JOIN SEATS s ON s.SeatID = r.SeatID
WHERE p.PassengerID = /* passenger criteria here */
I know the Sql Server solution isn't much immediate help to you, but perhaps it will suggest a better MySql solution.
Without APPLY, the only way I know to do this is to first compute the MAX() match count for your passengers (how many parts of the rules match):
SELECT p.PassengerID,
MAX(CASE WHEN r.Origin <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Destination <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Passenger_Type <> '' THEN 1 ELSE 0 END) AS MatchCount
FROM Passenger p
INNER JOIN SEAT_ALLOCATION_RULE r ON COALESCE(NULLIF(r.Origin, ''),p.Origin) = p.Origin
AND COALESCE(NULLIF(r.Destination,''), p.Destination) = p.Destination
AND COALESCE(NULLIF(r.Passenger_Type,''),p.Passenger_Type) = p.Passenger_Type
GROUP BY p.PassengerID
And then use that to filter down to results that have the same number of matches:
SELECT p
FROM Passenger p
INNER JOIN ( /* matchecounts */
SELECT p.PassengerID,
MAX(CASE WHEN r.Origin <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Destination <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Passenger_Type <> '' THEN 1 ELSE 0 END) AS MatchCount
FROM Passenger p
INNER JOIN SEAT_ALLOCATION_RULE r ON COALESCE(NULLIF(r.Origin, ''),p.Origin) = p.Origin
AND COALESCE(NULLIF(r.Destination,''), p.Destination) = p.Destination
AND COALESCE(NULLIF(r.Passenger_Type,''),p.Passenger_Type) = p.Passenger_Type
GROUP BY p.PassengerID
) m ON m.PassengerID = p.PassengerID
INNER JOIN SEAT_ALLOCATION_RULE r ON COALESCE(NULLIF(r.Origin, ''),p.Origin) = p.Origin
AND COALESCE(NULLIF(r.Destination,''), p.Destination) = p.Destination
AND COALESCE(NULLIF(r.Passenger_Type,''),p.Passenger_Type) = p.Passenger_Type
INNER JOIN SEATS s ON s.SeatID = r.SeatID
WHERE m.MatchCount =
(CASE WHEN r.Origin <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Destination <> '' THEN 1 ELSE 0 END
+ CASE WHEN r.Passenger_Type <> '' THEN 1 ELSE 0 END)
AND p.PassengerID = /* Passenger criteria here */
Which repeats a lot of code as well as effort in the DB, and is not very efficient. You can repeat the passenger criteria in the nested query, but that would only help a little. This option might also return multiple records for a passenger if they match two rules equally, though you can solve this easily enough with a GROUP BY expression.
In either case, note you can improve performance and simplify code by using actual NULL values instead of empty strings for missing parts of the SEAT_ALLOCATION_RULE table.