我需要在这里交易吗？

Question

I have a web application in which users create objects that get stored in a MySQL database. Each object has a globally unique identifier that is stored in a table.

DROP TABLE IF EXISTS `uidlist`;
CREATE TABLE IF NOT EXISTS `uidlist` (
`uid` varchar(9) CHARACTER SET ascii COLLATE ascii_bin DEFAULT NULL,
`chcs` varchar(16) DEFAULT NULL,
 UNIQUE KEY `uid` (`uid`)
) ENGINE=InnoDB DEFAULT CHARSET=ascii;

When a new object is to be created and stored I generate a new uid and ensure that it does not already exist in the uidlist table. (I should mention that collisions are rare since the potential range of UIDs I have is very large).

No issues here - it works just fine. However, with an increasing number of users wanting to simultaneously create + store objects the need to check uidlist is liable to become a bottleneck.

To circumvent the problem here is what I have done:

I have a secondary table

DROP TABLE IF EXISTS `uidbank`;
CREATE TABLE IF NOT EXISTS `uidbank` (
`uid` varchar(9) CHARACTER SET ascii COLLATE ascii_bin DEFAULT NULL,
`used` tinyint(1) NOT NULL DEFAULT '0'
 ) ENGINE=InnoDB DEFAULT CHARSET=ascii;

I pre-populate this table at short intervals via a CRON job - which ensures that it always has 1000 uid values that are tested for uniqueness.

When a live user requires a new UID I do the following:

function makeUID()
{
 global $dbh;
 $sql = "DELETE FROM `uidbank` WHERE used = '1';";
 //discard all "used" uids from previous hits on the bank 
 $sql .= "UPDATE `uidbank` SET used = '1' WHERE used = '0' LIMIT 1;";
 //setup a new hit
 $dbh->exec($sql);
 //be done with that

 $sql = "SELECT uid FROM `uidbank` WHERE used = '1'";
 $uid = $dbh->query($sql)->fetchColumn();
 //now pickup the uid hit we just setup
 return $uid;
 //return the "safe" uid ready for use
}

No issues here either. It works perfectly well in my single user test environment. However, my SQL skills are pretty basic so I am not 100% sure that

• this is the right way to handle the job
• that my "safe" UID pickup method will not return unsafe values because in the mean time another user has been assigned the same UID.

I'd much appreciate any tips on how this scheme might be improved.

Answer 1

Any reason you are not using a serial as your unique identifier? That would definitely be my first suggestion and would negate the necessity for the complicated setup you have.

Assuming that there is some reason then the biggest flaw I can see in your current makeUID call is that you are likely to get into a situation whereby the update sets 'used' to 1 and this row is deleted by a secondary call to makeUID before it has been able to successfully return the column meaning your second select (SELECT uid FROM uidbank WHERE used = '1') would return no rows

Could you explain why you dont use a serial so then I can try and get a better idea as to what is going on