fusion predict write error
I routinely run into a disk write error when running the evaluation pipeline via geowatch schedule
that stalls the process, necessitating killing all the cmd_queue workers and restarting. The error occurs when extracting features (fusion/predict.py).
The error message indicates a trace that ends up in kwimage's cloud optimize geotiff write function. I usually I just set the writer queue to serial in fusion/predict.py to partially circumvent this (but it still happens).
I was able to replicate, trace now included below.
Edited by Scott Workman