But depending on your zoom levels and map size, ideally you don’t even have a lambda service and just precache tiles. Then you’ve reduced your service to a dead simple file host problem.
If you don’t need that, you can just pre-generate all PNGs - then you don’t even need a server anymore, just a hash function to identify the right file and a big disk.
Surely these are examples of things that a client could do with raster data independent of the capabilities of a server?
I get why rasterio is preferred, its API is less annoying, but I'm not sure putting a leaking abstraction over gdal helps much. It just creates an aluring veneer over gdal, without really solving the main problems of gdal, being its monolithic monstrosity of hardly portable yet highly necessary GIS tool functionality.
Doesn't matter too much with gdal's warts, as this server should be deployed with ansible or in a container anyways, so portability is handled in a somewhat sane manner.
Thanks for open sourcing this!
Maybe, but I think mapbox is doing a great job with rasterio, and so far we didn’t run into any unsolvable problems (although we got close sometimes). I think for the ease of installation alone rasterio is a blessing.
> I think for the ease of installation alone rasterio is a blessing.
How does rasterio simplify installation, when you still have to deal with gdal's C extension dependencies?
Whenever I see a GIS application, I'm always curious to see how they solve the gdal problem.
The best is always no GDAL, but that gets pretty hard to do quite quickly as GDAL contains some really great algorithms for dealing with non-trivial raster processing.
Any chance you might consider adding NetCDF support? Used in so many places, contains time, but currently requires a behemoth like geoserver to share as a service
It's not :-)