Yes, the problem should be mentioned in the doc. If I remember well, the Seafile Team prefer we mention issues here than on GitHub, @daniel.pan can you look at this?
I found a workaround for now, by compiling numpy on my machine using pip.
Here are some (not very detailed) instructions on how to do it:
-
modify the seafile-server.yml to add a new shared volume, like:
/opt/seafile-pip:/shared2 -
restart docker containers
-
open a shell in the seafile container:
$ docker exec -it seafile /bin/bash -
remove pre-compiled numpy and copy pip packages to the new shared volume:
$ pip3 uninstall numpy
$ cp -r /usr/local/lib/python3.12/dist-packages/* /shared2 -
modify again the seafile-server.yml to make the shared volume replace the pip directory:
/opt/seafile-pip: /usr/local/lib/python3.12/dist-packages/ -
restart docker containers
-
re-open a shell in the seafile container:
$ docker exec -it seafile /bin/bash -
compile numpy for your computer (this can take a lot of time):
$ pip install --no-binary numpy numpy -Csetup-args=-Dcpu-baseline=none -
restart docker
-
test
One drawback is that you must update yourself all pip packages. Maybe this can be improved by just mounting the numpy directory instead of all pip modules…
I decided to use this workaround during the time to find a more recent CPU, because I think this problem will append more and more with time, not only with Seafile…