[OSM-talk] Mapnik rendering update

[Jon Burgess]

On Fri, 2008-01-25 at 23:28 +0000, Jon Burgess wrote:
> On Fri, 2008-01-25 at 22:35 +0000, Robert (Jamie) Munro wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> > 
> > Jon Burgess wrote:
> > | Once this first pass of the rendering is complete the mod_tile setup
> > | will have all the existing tiles fully rendered with the current planet
> > | dump and current map style. At that point I will switch over the URLs on
> > | the server to make the new tiles live.
> > |
> > | The new system has already rendered 5M tiles in just under 24 hours.
> > | This covers all of zooms 0 - 12 and about halfway through zoom 13. At
> > | the current rate all the old 30M tiles will be rendered by about
> > | Wednesday next week.
> > 

I've made the new tiles from mod_tile live on the main OSM tile server
(on the default Mapnik layer). The initial impression I get is that the
tiles are being served to more responsively and the on-access rendering
seems to be working well too. Hopefully this will all continue smoothly.


There are a couple of issues that I've noticed so far. I don't think any
of these are serious enough to force us to go back to the old rendering
setup. They should all be resolved in due course.

* The new colour reduction algorithm causes some odd colour changes in a
few tiles. 99% are fine but a few have a background colour or road fill
which is slightly wrong.

* Some of the z16/z17 tiles from the old system have not been rendered
yet. They should get rendered and appear shortly after you access them.
The new system uses a 8x8 meta tile rendering approach so overall we
actually have more tiles than the old system (40M vs 30M previously).

* The 'Expires' HTTP header has been broken causing tiles not to be
cached as well as they should be in your browser (or other proxy/cache).

* The munin stats for Apache volume have broken, you can see the real
throughput is still OK by looking at the eth0 stats.

	Jon