-
Notifications
You must be signed in to change notification settings - Fork 303
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Parts of satpy fail with the dask distributed scheduler (geotiff writer, nearest neighbor resampler) #1762
Comments
Many parts of satpy do not support the distributed scheduler. The geotiff writer is a big one. It can be improved if we switched to rioxarray: https://corteva.github.io/rioxarray/stable/examples/dask_read_write.html @mraspaud has played around with this idea in the past and may have other ideas. Other parts of this come down to how the data is opened in the reader. Also, the pykdtree (as your last error shows) is not serializable and can't be done over distributed. Not resampling or using the 'native' resampler are currently the only options right now. I believe we'd have to update pydktree to allow these serialized trees...or maybe the gradient search works. That's another thing @mraspaud has played with I think. |
Hmm, OK. Then the update to trollflow2 that @pnuu made in pytroll/trollflow2#83 is not currently useful until there are major updates to satpy, trollimage, and pykdtree? For some reason I thought some centres were using the distributed scheduler operationally, across multiple servers, but apparently not? |
I could of course be completely wrong, but that's how I've understood a lot of this stuff. I also think @pnuu sometimes "cheats" and generates his KDTree indexes in a threaded scheduler, caches them as zarr arrays, and then uses that cache when he runs them in a distributed manner. |
It works when using |
Describe the bug
When using
dask.distributed.Client()
,Scene.save_datasets()
fails withTypeError: self._hds cannot be converted to a Python object for pickling
when using the geotiff writer.To Reproduce
Expected behavior
I expect the datasets to be written without error messages.
Actual results
Full console output:
An image is written, but the image is entirely black. Here converted to PNG because GitHub won't let me upload TIF.
Environment Info:
Additional context
Other writers also seem to fail. With the
geotiff
writer, it fails as described above. With thesimple_image
writer, it gets stuck in an endless loop starting with (when interrupted):With the NetCDF writer, it gets stuck in a different endless loop:
The text was updated successfully, but these errors were encountered: