[OSM-talk] Live Data - all new Data in OSM

[Tom Hughes]

Frederik Ramm wrote:

> Tom Hughes wrote:
>> It's a completely insane solution though. It we want to do it we 
>> should just do it properly in the database not fart around with stupid 
>> hacks in the rails code that break as soon as any updates are not done 
>> via rails.
> 
> Assuming for a moment that the database was our bottleneck, something 
> that can be done by "farting around" on a number of easily scalable API 
> servers would of course compare favourably to burdening the 
> not-so-scalable database with triggers and extra write operations, would 
> it not?

The fact that the servers are easily scalable is part of the problem as 
it means that any such logging system involves merging the actions of 
some 80 or so processes spread over 4 separate machines (at present).

That either means some complicated and fragile locking scheme to control 
who is writing to the log at any given time or some scheme for merging a 
whole load of separate logs.

> Now I don't know how often you manually modify database contents, but I 
> would think that any operation of a scale that would lead us to bypass 
> the rails API would also be very likely to blow apart anyone who listens 
> for edits downstream, so in my eyes there's not much to be gained by 
> streaming these "manual override" kinds of edits as well.

I'm not thinking about manual modifications. I'm thinking about things 
like the gpx import that are no longer in rails. I think that is only 
likely to spread to include much of the API in the not too distant future.

Tom

-- 
Tom Hughes (tom at compton.nu)
http://www.compton.nu/