← Back to team overview

maria-developers team mailing list archive

Re: GSoC 2016:Unique indexes for blobs


Hello Sergei,

         I understood most of the internals of long unique constraints in
MyISAM. I am still going through the code in InnoDB. I will soon reply to


On 21 March 2016 at 16:37, Sergei Golubchik <serg@xxxxxxxxxxx> wrote:

> Hi, Shubham!
> On Mar 21, Shubham Barai wrote:
> >
> > I am currently looking into InnoDB codebase to see if it is possible
> > for me to extend this feature to InnoDB storage engine. As  InnoDB
> > doesn't support this feature internally, it would require more time
> > than MyISAM.  Any suggestions regarding this would be helpful.
> Heh, that's very good (and ambitious) :)
> Do you already know how MyISAM supports arbitrary long UNIQUE
> constraints internally? It stores only the hash of the value (of the
> blob, for example) in the non-unique index, and on INSERT it checks if
> there are identical hashes in the index. If there are (hash collision)
> it will retrieve the rows and compare actual blob values.
> It seems that InnoDB should use the same approach as MyISAM. It'll need
> some care for a case when two concurrent transactions insert conflicting
> rows - as transaction changes are not visible until commit, you won't
> see the conflict until it's too late. But gap locks [1] should be able
> to prevent that.
> Regards,
> Sergei
> Chief Architect MariaDB
> and security@xxxxxxxxxxx
> [1] https://dev.mysql.com/doc/refman/5.7/en/innodb-record-level-locks.html

Follow ups