Efficient select on huge table of ranges - mysql

Question

I have a mysql table which contains the fields: rangeFrom and rangeTo.
I want to request rows with a condition like: rangeFrom >= ? AND rangeTo <=? within a join.
EXPLAIN SELECT *
FROM Version
JOIN Contract FORCE INDEX FOR JOIN (versionRangeFrom)
ON Version.id >= Contract.versionRangeFrom
AND Version.id <= Contract.versionRangeTo
WHERE Version.completedAt = '2016-06-06 10:00:01';
Which mysql explains like this:
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE Version ref PRIMARY,completedAt completedAt 6 const 1 NULL
1 SIMPLE Contract ALL versionRangeFrom NULL NULL NULL 640744 Range checked for each record (index map: 0x8)
So it has to work though 640744 rows which takes about 1-2 seconds.
However inserting the version id in the queryworks fine
EXPLAIN SELECT *
FROM Contract
WHERE 5 >= Contract.versionRangeFrom AND 5 <= Contract.versionRangeTo;
This is then explained like this:
id select_type table type possible_keys key key_len ref rows Extra
1 SIMPLE Contract range versionRangeFrom versionRangeFrom 4 NULL 534 Using index condition; Using where
So in this case mysql only goes though 534 rows and that only takes about 30ms.
So how do I prepare for such a range check correctly. It seems that mysql is unable to use Indexes in those cases. I can work around it by using 2 queries but i'd rather have one.
Here more schemas:
CREATE TABLE `Version` (
`id` int(10) unsigned NOT NULL AUTO_INCREMENT,
`completedAt` datetime DEFAULT NULL,
`createdAt` datetime NOT NULL,
PRIMARY KEY (`id`),
KEY `completedAt` (`completedAt`)
)
CREATE TABLE `Contract` (
`id` int(10) unsigned NOT NULL AUTO_INCREMENT,
`contractId` bigint(20) unsigned NOT NULL,
`startAt` bigint(20) NOT NULL DEFAULT '0',
`endAt` bigint(20) NOT NULL DEFAULT '0',
`tradeStartAt` bigint(20) NOT NULL DEFAULT '0',
`tradeEndAt` bigint(20) NOT NULL DEFAULT '0',
`latestAiId` bigint(20) NOT NULL DEFAULT '0',
`type` varchar(4) COLLATE utf8_unicode_ci NOT NULL,
`daPreis` int(11) NOT NULL DEFAULT '0',
`lastTradePreis` int(11) NOT NULL DEFAULT '0',
`lastTradeVol` int(11) NOT NULL DEFAULT '0',
`VWAID` double NOT NULL DEFAULT '0',
`versionRangeFrom` int(10) unsigned NOT NULL,
`versionRangeTo` int(10) unsigned DEFAULT NULL,
PRIMARY KEY (`id`),
KEY `tradeStartAt` (`tradeStartAt`),
KEY `contractId` (`contractId`),
KEY `versionRangeFrom` (`versionRangeFrom`)
)

Answer 1

A different value than '5' would not need to look at only 534 rows.
The problem (x < from AND x > to)is non-trivial, and for which there is no simple answer.
Shrinking the table size would help a tiny bit. Don't use BIGINT (8 bytes) when some smaller datatype would suffice.
Perhaps the only real solution involves a major revamp of the schema and code. See my blog.
Edit
In certain situations, this subquery may be efficiently used to find the row in question:
( SELECT Contract.id FROM ...
WHERE Version.id >= Contract.versionRangeFrom
ORDER BY versionRangeFrom
LIMIT 1 )