A systemd script for running a Jupyter notebook server.

jupyter.service

# After Ubuntu 16.04, Systemd becomes the default. 
# It is simpler than https://gist.github.com/Doowon/38910829898a6624ce4ed554f082c4dd

[Unit]
Description=Jupyter Notebook

[Service]
Type=simple
PIDFile=/run/jupyter.pid
ExecStart=/home/phil/Enthought/Canopy_64bit/User/bin/jupyter-notebook --config=/home/phil/.jupyter/jupyter_notebook_config.py
User=phil
Group=phil
WorkingDirectory=/home/phil/Notebooks/
Restart=always
RestartSec=10
#KillMode=mixed

[Install]
WantedBy=multi-user.target

If encountering the following errors :

 Main PID: 24251 (code=exited, status=1/FAILURE)

Mar 21 13:48:21 ip-172-31-26-72 systemd[1]: jupyter.service: Main process exited, code=exited, status=1/FAILURE
Mar 21 13:48:21 ip-172-31-26-72 systemd[1]: jupyter.service: Unit entered failed state.
Mar 21 13:48:21 ip-172-31-26-72 systemd[1]: jupyter.service: Failed with result 'exit-code'.

Check /var/log/syslog for a more verbose message. It may be that

ValueError: certfile "certs/mycert.pem" does not exist

Solution : use absolute paths in your config file.

I suggest putting the service file in /etc/systemd/system/, not /usr/lib/systemd/system/. The latter is only for system units installed from the distribution package manager, while the one in /etc is for user-defined ones. It will make no difference in terms of functionality, but prevent conflicts with any package that might happen to install a jupyter.service in the future. Reference: systemd.unit(5)

Thanks @sandtler, that is a good point - I have updated the instructions in the first comment.

I suggest putting the service file in /etc/systemd/system/, not /usr/lib/systemd/system/. The latter is only for system units installed from the distribution package manager, while the one in /etc is for user-defined ones. It will make no difference in terms of functionality, but prevent conflicts with any package that might happen to install a jupyter.service in the future. Reference: systemd.unit(5)

If you need environment variables for Tensorflow GPU work, and .bashrc usually does the job for you, you need to somehow make those available to the Jupyter service, or else Notebooks that need the GPU won't be able to see it.

Not sure this is the best way, but it's the way I got it to work:

[Unit]
Description=Jupyter Notebook

[Service]
Type=simple
PIDFile=/run/jupyter.pid
# ------------------------------------
Environment="PATH=/usr/local/cuda-10.1/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
Environment="LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64:/usr/local/cuda-10.1/lib64:usr/local/cuda/lib64:/usr/local/cuda/extras/CUPTI/lib64"
Environment="CUDA_HOME=/usr/local/cuda-10.1"
Environment="NUMBAPRO_NVVM=/usr/local/cuda-10.1/nvvm/lib64/libnvvm.so"
Environment="NUMBAPRO_LIBDEVICE=/usr/local/cuda-10.1/nvvm/libdevice/"

ExecStart=/home/phil/.local/bin/jupyter-notebook
User=phil
Group=phil
WorkingDirectory=/home/phil/z_Notebooks
Restart=always
RestartSec=10
#KillMode=mixed

[Install]
WantedBy=multi-user.target