On Mon, Oct 8, 2012 at 9:33 PM, Paweł Paprota <[email protected]> wrote: > Hi Ian, > > Thanks for the response. What you wrote about partitioning the data etc. > is exactly what concerns me with the Changeset Activity Publisher > implementation - it will get too big with time.
what Ian says is exactly right - the current status is that work needs to be done to either denormalise the data or figure out a clever way to handle non-spatial queries over this spatial data structure. the least horrific option seems to be the denormalisation, but i've also been playing around with trying to write an sp-gist index and toying with the question of whether some other form of database might be a better fit. > On the other hand, for "social" purposes I guess there is no need to > hold on to every single change since forever... Does anyone read > Facebook posts from few months ago? Guess not... depending on how fast the generation process is, could it potentially be done on-the-fly and cached? > Slightly different use cases I think but still it would be good to learn > from OWL devs knowledge about scaling something like this. i can definitely tell you what doesn't work: sticking it all in one table and hoping the database will magically sort it out ;-) as yet i haven't seen a solution which doesn't incur significant drawbacks. cheers, matt _______________________________________________ dev mailing list [email protected] http://lists.openstreetmap.org/listinfo/dev
