Pulp

Can demonstrate the problem here, showing how the services attempt to be started and how they are incorrect. There is no virtualenv for pulp, package commands (like gunicorn) are in $PATH and don't need to be full paths.

[vagrant@localhost ~]$ cat /etc/systemd/system/multi-user.target.wants/pulpcore-api.service
[Unit]
Description=Pulp WSGI Server
After=network-online.target
Wants=network-online.target

[Service]
Environment="DJANGO_SETTINGS_MODULE=pulpcore.app.settings"
Environment="PULP_SETTINGS=/etc/pulp/settings.py"
Environment="PATH=/usr/local/lib/pulp/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin"
User=pulp
Group=pulp
PIDFile=/run/pulpcore-api.pid
RuntimeDirectory=pulpcore-api
ExecStart=/usr/local/lib/pulp/bin/gunicorn pulpcore.app.wsgi:application \
          --bind '127.0.0.1:24817' \
          --workers 4 \
          --access-logfile -
ProtectSystem=full
PrivateTmp=yes
PrivateDevices=yes


# This provides reconnect support for PostgreSQL and Redis. Without reconnect support, if either
# is not available at startup or becomes disconnected, this process will die and not respawn.
Restart=always
RestartSec=3

# This directive is set to an absolute path in other Pulp units. Using an
# absolute path is an abuse of the directive, as it should be a relative path,
# not an absolute path. PIDFile is now used to ensure that PID files are laid
# out in a standard way. If this directive had any other effects, it is better
# to use the correct directive than to uncomment this.
# WorkingDirectory=/var/run/pulpcore-api/

[Install]
WantedBy=multi-user.target
[vagrant@localhost ~]$ /usr/local/lib/pulp/bin/gunicorn pulpcore.app.wsgi:application --bind '127.0.0.1:24817' \
>           --workers 4 \
>           --access-logfile -
-bash: /usr/local/lib/pulp/bin/gunicorn: No such file or directory

The isssue was caused by roles like pulp_api having lines in their defaults/main.yml like: pulp_install_dir: "/usr/local/lib/pulp"

I fixed it by removing all of those defaults, incorrect or correct, that are now duplicate. They became duplicate because this change made those roles actually depend on pulp_common.

https://pulp.plan.io/issues/7005

As a user, I can run a single stably-named Ansible role to install Pulp