GDALRasterSource Issue Tracking

These are known anomalies for using GDALRasterSource that need to be fixed before we can start using it in production:

• Fails with an error for the “Mona Lisa” Tiff with the following error: ERROR 1: The transformation is already "north up" or a transformation between pixel/line and georeferenced coordinates cannot be computed for /vsis3/rasterfoundry-production-data-us-east-1/user-uploads/auth0|57b744b09da7f9936f14fd5b/e16a297d-48e4-42b6-b301-18ddca0a8f9f/monalisa-cog.tiff. There is no affine transformation and no GCPs. Specify transformation option SRC_METHOD=NO_GEOTRANSFORM to bypass this check. (I’ve attached a zipped version of the tiff)

I’m not sure if this is a problem that can be solved with GDAL or if it’s just bad data that we handled naively before so didn’t throw an error.

• The GDAL bindings do not respect reading credentials from an AWS creds profile (though it does work when used on a machine with an IAM role) so this seems to primarily a local problem
• Servers seem to stop responding to requests after a period of time – it’s probably not worth looking into this further until we have better, reproducible steps

monalisa-cog.zip

[notthatbreezy]

Figured out how to correctly “georeference” non-georeferenced imagery to satisfy GDALRasterSource’s requirements (and GDAL more generally).

The strategy I used is to call gdalinfo and get the upper-left and lower-right coordinates and use that in gdal_translate when assigning a spatial reference system like:

gdal_translate -a_ullr 0.0 369.0 656.0 0.0 -a_srs epsg:3857 usa_joel_embiid_3_sixers.png fixed.tiff

Which I used to generate

[jisantuc]

Possible reproduction of the “server stops after a while” problem:

• add this tif as a COG to a project: s3://rasterfoundry-development-data-us-east-1/user-uploads/auth0%7C59318a9d2fbbca3e16bcfc92/9aa8e212-204c-4531-9a0b-0fb8d3faa2f0/cog.tif
• zoom in and out a bunch
• watch as new requests instantly trigger 499s (eventually – doesn’t take much)

~Note that the server does eventually recover, but the behavior while it’s cranky isn’t great~ it seems like maybe it doesn’t actually recover – my healthcheck is now timing out after waiting a bit for things to clear

EDIT: thinking back, my project with a huge number of scenes where I was trying to build a global basemap… was also MODIS imagery.