[Zope] When to use a relational database

[Jaroslav Lukesh]

----- Original Message ----- 
From: "Dieter Maurer" <dieter at handshake.de>
> quizzical wrote at 2008-2-6 16:03 -0800:
>> ...
>>Apologies if this is a dumb question, I have had a good look through the
>>archives and on the web. My question is basically about how to design a
>>website. I really like the ease with which you can build applications in
>>zope but I often see comments implying that in order to scale well one may
>>want to use a backend relational datbase. I have also seen several 
>>comments
>>to the effect that one should keep the state of a user session out of the
>>ZODB.
>>
>>I am slightly confused by this, I can't find any information which would
>>help me decide what data would be better off in a relational database.
>
> A relational database is good when you have large amounts
> of strictly structured data -- e.g. millions of user records
> where each user record has a fews fields with atomic datatypes.

For example pricelists etc. But you need to make join with ZODB - with 
hyperlink (URL1) or ZODB folder object property

> The ZODB is good when you have weakly structured or unstructured data
> and your write rate is not too high.

ZODB effectivity is very high at tree structured data, as catalogues etc

I use ZODB as catalogue container and RDB as pricelist at one travel agency 
site. If you want to know which, I will send you ULR to private mail and 
make some source view method for you. But it is written in pure DTML, few 
computing-intensive parts are in Python (tree structure, etc) and ZSQL to 
MSSQL. No sql-inject is possible thanks to DTML ;-)

Near all other my webs are pure ZODB and uses SQL to store forms data only.

Pure SQL driven web I use only for our own ERP, at ZODB are only DTML and 
bunch of ZSQL.

> The ZODB uses an almost self describing format. Therefore, it can
> store virtually everything, without any data model fixed beforehand.
> However, this leads to high redundancy and efficiency loss.


It is possible to use some products like linux symlink to avoid data 
duplicates.

Regards, JL.