Re: Identity modelling
Date: 30 Aug 2005 20:08:42 -0700
Message-ID: <1125457721.981045.102740_at_z14g2000cwz.googlegroups.com>
paul c wrote:
>
> Some people say a key is an "address" because keys have been described
> as the RM's "addressing mechanism".
Yeah, I've heard that kind of talk, too, and it drives me nuts. It totally glosses over the differences between location-based addressing and content based addressing.
> Guaranteeing sequence without gaps is a
> bigger problem than many people realize in fact I believe even though I
> can't prove it except anecdotally that there are always situations when
> it won't work.
Transaction A inserts a row in a table with an autoincrement column, but doesn't commit. Transaction B inserts a row in the same table, and gets the successor id as transaction A. Transaction B commits. Transaction A rolls back. Now you have a hole.
The only way out of this that I can see is full serialization of inserts into tables with autoincrement columns, which is a solution much worse than the problem!
Marshall Received on Wed Aug 31 2005 - 05:08:42 CEST