[gdal-dev] Missing exception when using multithread=True for gdal.Warp

[Tim Harris]

Hi, I'm having some trouble warping a VRT to a TIF where the VRT and its
constituent rasters are in S3. It stems from a random read error, but the
real problem is that the gdal.Warp call didn't throw an exception when it
was supposed to.

I saw there's this pretty old GitHub issue that may be related, but I'm not
sure:
https://github.com/OSGeo/gdal/issues/3372

In my situation what really happened was a random /vsicurl read failed...
sort of like an issue I reported in December that was fixed in GDAL 3.10.1:
https://github.com/OSGeo/gdal/issues/11552

Here's a section of my log file (filename sanitized):
ERROR 11: CURL error: Empty reply from server
0...10...20...30...40...50...60...70...80...90...100 - done.
ERROR 4: `/vsis3/path/to/raster.tif' does not exist in the file system, and
is not recognized as a supported dataset name.

Notice the CURL error but the progress still goes to 100%. The warp
completed successfully despite the read error.

I have a hopefully simple reproduce case here. The first few shell commands
will make two TIFs, put them in S3, then make a VRT of those TIFs and also
put that in S3. Then to simulate the CURL error above, it just deletes one
of the TIFs so that the VRT references a non-existent file:
gdal_create -of GTiff -outsize 1000 1000 -bands 1 -a_srs EPSG:4326 -a_ullr
0 1 1 0 input1.tif
gdal_create -of GTiff -outsize 1000 1000 -bands 1 -a_srs EPSG:4326 -a_ullr
1 1 2 0 input2.tif
aws s3 cp input1.tif s3://your-bucket/test/input1.tif
aws s3 cp input2.tif s3://your-bucket/test/input2.tif
gdalbuildvrt input.vrt /vsis3/your-bucket/test/input1.tif
/vsis3/your-bucket/test/input2.tif
aws s3 cp input.vrt s3://your-bucket/test/input.vrt
aws s3 rm s3://your-bucket/test/input2.tif

Then in Python:
from osgeo import gdal
gdal.UseExceptions()
gdal.Warp("output.tif", "/vsis3/tharris-bucket/test/input.vrt",
callback=gdal.TermProgress)

This throws a RuntimeError, as expected. But delete the partial file and
try with multithread=True:
rm output.tif
gdal.Warp("output.tif", "/vsis3/tharris-bucket/test/input.vrt",
callback=gdal.TermProgress, multithread=True)

This logs the error, but doesn't throw an exception. The end result is that
I have a TIF that is missing the contents of the file that suffered from
that CURL read error.

For what it's worth this also happens with the command line utility, it's
just not as obvious. Without -multi, it only prints the "0" for the
progress before the error halts execution:
gdalwarp /vsis3/your-bucket/test/input.vrt output.tif
Processing /vsis3/your-bucket/test/input.vrt [1/1] : 0ERROR 4:
`/vsis3/your-bucket/test/input2.tif' does not exist in the file system, and
is not recognized as a supported dataset name.

But with -multi, notice the progress goes to 100%:
gdalwarp /vsis3/your-bucket/test/input.vrt output.tif -multi
Processing /vsis3/your-bucket/test/input.vrt [1/1] : 0ERROR 4:
`/vsis3/your-bucket/test/input2.tif' does not exist in the file system, and
is not recognized as a supported dataset name.
...10...20...30...40...50...60...70...80...90...100 - done.

I guess my workaround is to just not use multithread=True. But is this
something that could be fixed?

Thanks
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20250228/5a9a6ade/attachment.htm>