Re: Auto-Increment

[Jon Roberts <jon@mentata.com>]

Sorry to resurrect this stale thread, but I have news to report. In 
short, I have changed my tune and made progress in the area of 
auto-incrementing identifiers. I revisited my solution to the 
requirement based on Howard's feedback, and not only did I implement it 
as prescribed by the best practice of storing the counter in the 
directory, I made this method available to all objects in my framework 
with a one-liner.

http://www.mentata.com/ldaphttp/examples/gospel/increment.htm

This is the open source process in action. I offer my thanks to the 
OpenLDAP community.

http://www.mentata.com/news/22Mar03.htm

To see the new algorithm work, just add a comment to the weblog:

http://www.mentata.com/ldaphttp/examples/gospel/jesusblog.htm

Don't be afraid. If lightning was going to strike anyone for this 
exercise, it would've surely been me :)

> > Looking at the archives, I'm definitely not the only one doing it like
> this.
>
> This is an argument for mediocrity. In the midst of an excellent post, I find
> this position surprising.

I agree, and I wasn't implying that because others do something it's 
automatically a good idea. I'm just suggesting that this is going to be 
the natural path for people who aren't shown the alternative up front. 
I'm actually presenting both techniques just in case somebody still 
finds cause to use a memory-resident incrementor instead.

There is a motivation I didn't mention in my original posts. An implicit 
goal of my efforts was to make my code work with anyone's directory, 
regardless of what they have in it. I wanted to leave all knowledge of 
specific object classes, attributes, identities, and access control 
rules to the implementing organization (and to the M in MVC). I suspect 
a lot of other developers are responding to this same subconcious need, 
too. With this requirement, I skirted the issue.

To a small extent, there is also still an issue of performance and 
perhaps scalability with this best practice. In the general case, the 
database vs. memory approaches for determining the next identifier 
compare at roughly 20:1 in terms of speed and resources. However, with 
both under a second and still only a fraction of the typical cost of 
creating an entry, I would conclude this consideration is barely relevant.

Why did I use a country object for my counter? Why not; it's one of the 
smallest classes in the core schema.

If this issue keeps coming up, as no doubt it will, feel free to send 
people to me as well as the list archives. I consider myself indebted, 
and am willing to work it off.

Yours in all humility,
Jon Roberts
www.mentata.com