This bug is currently preventing the bfast_gpu from executing on both test and prod servers.
description of the bug
When a new requirements.txt is set in a dashboard app the following script is updating/creating the venv used to launch the app. The proble m is if one of the lib is removed as the venv environment is not flush, it is not removed from the venv. It is painless is most of the case but for c++ based libs it can become critical (GDAL, CUDA, PROJ).
In the specific case of bfast_gpu the cupylib is dependant on the CUDA version. The previous installation was thus using cupy-cuda112 but as we updated CUDA in SEPAL we now request cupy-cuda114.
If you look in the current venv in the test server, launching the following script in this ven will raise the following error:
import cupy as cp
CuPy may not function correctly because multiple CuPy packages are installed
in your environment:
cupy-cuda112, cupy-cuda114
potential solution
here, instead of directly install we could do it in 2 steps:
This bug is currently preventing the
bfast_gpu
from executing on both test and prod servers.description of the bug
When a new
requirements.txt
is set in a dashboard app the following script is updating/creating the venv used to launch the app. The proble m is if one of the lib is removed as the venv environment is not flush, it is not removed from the venv. It is painless is most of the case but for c++ based libs it can become critical (GDAL, CUDA, PROJ).In the specific case of
bfast_gpu
thecupy
lib is dependant on the CUDA version. The previous installation was thus usingcupy-cuda112
but as we updated CUDA in SEPAL we now requestcupy-cuda114
.If you look in the current venv in the test server, launching the following script in this ven will raise the following error:
potential solution
here, instead of directly install we could do it in 2 steps: