Development cycle using GOODS for persistence
Yanni Chiu
yanni at rogers.com
Tue Oct 19 15:21:50 UTC 2004
Chris Muller wrote:
> But in the three-years (two companies) in which I was priviledged to work with
> it, nobody wanted to take on the complexity of having different versions of
> objects and code running simultaneously in a production system.
>
> I can totally understand this. Imagine the headaches that could be created for
> a production system if subtle differences between the code in versions of a
> class caused "bad data" that went unnoticed for a long time. Instead, they
> tackled the migrations head-on. Sure, during development, you could relax a
> little, but part of the deployment process ensured one version of each class.
In a talk at Smalltalk Solution '04, on GemStone usage a WAMU, the
approach taken was a lazy conversion. So two class definitions/method
dictionaries were present in production, but the conversion took
place JIT, IIUC. It had to be done this way because the large volume
of data to be converted meant the downtime for conversion was too long.
--yanni
More information about the Squeak-dev
mailing list
|