[Tile-serving] Switch2osm rewrite review needed

[Paul Norman]

> From: Marc Tobias [mailto:mtm at opencagedata.com]
> Sent: Saturday, May 10, 2014 5:52 AM
> To: Paul Norman
> Subject: [Tile-serving] re: Switch2osm rewrite review needed
> 
> Hi Paul,
> 
> Just this week I spent two days going through 3 year old setup
> instructions to setup a EC2 machine and wish it had been the same
> quality as your guide.
> 
> You use https:// instead of git:// for checkout, apt-get upgrade,
> correctly split the various apt-get install into sections so there is
> less guessing what is needed for what software, you use md5sum after
> download, you use long parameter switches ... Exactly how it should be
> written
> 
> My feedback:
> - add a link in 'found on the planet.osm wiki page' and 'tilemill'

I'm planning on the tilemill link being to another guide that I'm still 
testing. Good catch on the wiki link, and on my to-do is cleaning up 
that wiki page.

> - the section 'loading with osm2pgsql' is twice in the document and you
> use almost the same wording to introduce the section so my first
> reaction was it's a copy&paste.

To a large extent it is. The issue is loading an extract and loading the 
full planet is different, and I'm really not sure how to best make the 
differences clear. There's more than just the osm2pgsql command line 
differences, with the planet you want to osmupdate the file before loading,
but that's not necessary or an easy option with geofabrik extracts.

> I tend to use http://aria2.sourceforge.net/ for downloading the planet
> file multi-threaded and recoverable, but I think it's a rare case.

Ya - it's an extra complication. Someone familiar with other download tools
can easily use them. Personally I always do the download in one curl line
with curl -OOL 'http://planet.openstreetmap.org/pbf/planet-latest.osm.pbf{.md5,}'
but if I throw that in the guide it will confuse people.

> I also edit the
> sudo sysctl -w kernel.shmmax=25968543300 for tuning. My setup is usually
> Nominatim, not tileserving, and I'm running that on SSD servers without
> swap file. Not sure if that's a common enough case.

This isn't necessary with Postgres 9.3. It's also not necessary with default 
Postgres settings, although default settings suck, which I note in the guide.