[OSM-dev] Data corruption :) II

[Stefan de Konink]

Andy Allan wrote:
> On Mon, Nov 24, 2008 at 11:06 AM, Stefan de Konink <stefan at konink.de> wrote:
> 
>> Foreign key constraints are part of the typical database; not an API
>> matter.
> 
> Yep, but here's where theory and real-life meet. The new API requires
> changes to the database schema (e.g. new tables for changesets,
> changing users columns to changesetids on many tables) and you only
> get a limited number of times you can take the database off-line to
> implement things before everyone starts getting pissed off. So API0.6
> includes more than just changes to the XML API.

Are you saying you are not more pissing people off with bad data than 
with fixing all constraint problems; and adding a foreign key constraints?

A maintaince window is very typical. So one way or the other. It would 
be good to start thinking about multiple api servers, (maybe take 
ROMA/HandlerOSM/etc. to compensate for your downtime), and for now 
leaving the others readonly while the read-write version is secured.

I don't really see a problem in this. It is not rocket science to put a 
reverse proxy in front of the API that will direct GET's to a bunch of 
slaves and PUT's to the main api. And take the main api down to add the 
constraints.


Stefan