Marek,

We are using Cayenne, MySQL with InnoDB storage engine, both configured for 
MySQL PK generation.  We had some problems which I am convinced was associated 
with the documented table locking bugs in early versions of MySQL 5.0.x.  It 
appears that there may have been multiple problems which do seem have been 
resolved in MySQL 5.1.x  (You can read up on this by googling the MySQL bug 
reports.)

With respect to these issues, the most current stable version (currently 
5.1.49) is recommended.  (Warning: there appears to be some sort of conceptual 
'holy-war' with 5.0.x vs the 5.1.x zealots; so I would just stick with the most 
recent stable version recommendation and the associated bug fixes as the 
argument)

I cannot speak to comparing the Cayenne PK generation strategy vs MySQL, but I 
believe both are very reliable.  I think that the only reason that you might 
consider using MySQL vs Cayenne PK-generation is if you are ever going to 
perform non-Cayenne inserts.  If you are always going to use an ORM, then you 
can use either method.

If you do any performance tests comparing the two, I would be interested in 
hearing the results.

Hope this helps,
Joe



On Jul 26, 2010, at 10:40 PM, Marek Stanisław Wawrzyczny wrote:

> Hi all,
> 
> It has been a while since I last posted in this forum.
> 
> First, a bit of a background. We have an existing MySQL schema that was 
> driven by a number of WebObjects applications. These obviously rely on EOF to 
> handle PK generation.
> We have decided to refactor the schema in a new database. The new web 
> applications are Cayenne based.
> 
> We now have a bit of discussion on how to handle PK generation in the new 
> schema. My personal preference is auto incrementing PK columns. It certainly 
> allows for a more generic approach to the handling of primary keys - it is 
> not necessary for people to be aware of any additional infrastructure should 
> records be inserted outside of any applications.
> 
> Others have expressed concern with MySQL handling of the LAST_INSERT_ID() 
> functionality where there used to be (still is?) a problem with MySQL's 
> threading implementation in high concurrency/volume environments. (I presume 
> this is precisely the mechanism used by Cayenne to populate IDs within the 
> Entities?)
> Another cited advantage being suggested is that this approach allows for a 
> vendor-independent PK generation strategy.
> 
> I guess my vested interest here is to dismiss arguments against the use of 
> the auto incrementing PK strategy :) but that said, I'm very interested to 
> hear whether anyone has done any performance analysis or has had particular 
> problems with either strategy.
> 
> 
> Cheers,
> 
> Marek Wawrzyczny
> ish group

Reply via email to