Arch Linux User Repository

Ah makes sense. Yeah, it's hard to balance support for old distros, new distros, constantly chasing deps, and the likes. We are a sticking with the django lts for now.

You probably want: community/uvicorn

Apparently I cannot read the output of pacman -Ss, how did I miss this.

Also I managed to get nginx-unit-python going. Turns out, do not mess with PYTHONHOME if you do not use a virtual environment or generally if you do not know what you are doing. It seems to find the Python modules if I remove the "home":-line from the configuration and set "path": to /usr/lib/etebase-server.

Still, I will be changing to a virtual environment because etebase-server-0.8.3 seems to require Django 3, and Arch is currently packaging Django 4, causing this error:

nginx-unitd-log[1363675]:   File "/usr/lib/etebase-server/django_etebase/signals.py", line 3, in <module>
nginx-unitd-log[1363675]:     user_signed_up = Signal(providing_args=["request", "user"])
nginx-unitd-log[1363675]: TypeError: Signal.__init__() got an unexpected keyword argument 'providing_args'

Similarly, /usr/lib/etebase-server/manage.py migrate is giving me a Django-related stack dump, too:

  File "/usr/lib/etebase-server/etebase_server/urls.py", line 4, in <module>
    from django.conf.urls import url
ImportError: cannot import name 'url' from 'django.conf.urls' (/usr/lib/python3.10/site-packages/django/conf/urls/__init__.py)

You probably want: community/uvicorn

Apparently this Django application switched from WSGI to ASGI. How do you guys run it? I have been using uwsgi-plugin-python, but that does not work anymore, obviously.

There is no guvicorn package available, so I went and tried with nginx-unit-python, but I cannot make it work for the life of me. I always get this error:

Fatal Python error: init_fs_encoding: failed to get the Python codec of the filesystem encoding
Python runtime state: core initialized
ModuleNotFoundError: No module named 'encodings'

This error is somewhat common when you google it, but solutions are scarce, and if they have one, it is based on having a virtual environment, so not applicable to this package. The point of having a package is not needing a venv, right?

I remember having the same error with uwsgi-plugin-python, but it weirdly went away when I switched to launching /usr/bin/uwsgi with all required options given on the command line instead of using an .ini file. Unfortunately this workaround is not applicable to nginx-unit-python.

Here is my configuration for nginx-unit-python:

{
    "listeners":{
        "unix:/srv/etebase/socket/etebase_server.sock": {
            "pass": "applications/etebase_server"
        }
    },
    "applications": {
        "etebase_server": {
            "type": "python 3.10",
            "processes": 1,
            "working_directory": "/srv/etebase",
            "path": "/usr/lib/python3.10/site-packages",
            "home": "/usr/lib/etebase-server",
            "module": "etebase_server.asgi",
            "callable": "application"
        }
    }
}

nginx-unitd is launched by systemd like this:

[Unit]
Description=nginx-unit running EteBase

[Service]
PrivateDevices=yes
PrivateTmp=yes
User=etebase
WorkingDirectory=/usr/lib/etebase-server
ExecStart=/usr/bin/unitd \
    --no-daemon \
    --control unix:/srv/etebase/nginx-unitd/control.sock \
    --state        /srv/etebase/nginx-unitd/state \
    --pid          /srv/etebase/nginx-unitd/nginx-unitd.pid \
    --log          /srv/etebase/nginx-unitd/nginx-unitd.log

[Install]
WantedBy=multi-user.target

Oops, done. Thanks!

Please add as dependancy : python-aioredis python-aiofiles

Yes, or change that path to something else.

Did you have to change ownership on /usr/lib/etebase-server/media to make it work?

Yeah, I think it's too specific. There are semi-scripts out there, but you really want to choose what you want and set it up accordingly.

Thanks. Managed to get it done, but idk if I'd be able to script the whole process to automate it.