headless displays on remote server with port-forwarding

How to properly use vtki on remote servers?

On a remote server with port-forwarding (i.e. ssh ... -L 8888:localhost:8888), the Jupyter kernel always dies when I call plot() on vtki objects (The kernel appears to have died. It will restart automatically.) . This happens on both AWS EC2 and a local cluster.

A minimum reproducible code is:

import vtki
from vtki import examples
ds = examples.load_airplane() 

ds.plot(use_pane=True)  # this kills the kernel
# use_panel=False gives the same error

Similar in ipython:

In [1]: import vtki
   ...: from vtki import examples
   ...: ds = examples.load_airplane()

In [2]: ds.plot()
ERROR:root:bad X server connection. DISPLAY=
Aborted (core dumped)

In principle it should work on remote servers, because the Binder demos are working properly. But Binder is connected directly via HTTP, while on most remote servers people would use SSH+port-forwarding.

Environment setup

I am using EC2 Ubuntu 18.04 LTS (ami-0a313d6098716f372 in us-east-1). The complete steps to install the packages are:

wget https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
bash miniconda.sh -b -p miniconda
echo '. $HOME/miniconda/etc/profile.d/conda.sh' >> ~/.bashrc && source ~/.bashrc
conda create --name vtk python=3.7
conda activate vtk
conda install nodejs  # required when importing vtki in Jupyter
pip install jupyter vtki panel

# To avoid "ModuleNotFoundError: No module named 'vtkOpenGLKitPython' " when importing vtk
# https://stackoverflow.com/q/32389599
# https://askubuntu.com/q/629692
sudo apt update && sudo apt install python-qt4 libgl1-mesa-glx

python -c "import vtki" # it should work

The notebook is started with:

ssh ... -L 8888:localhost:8888
conda activate vtk
jupyter notebook --NotebookApp.token='' --no-browser --port=8888

[JiaweiZhuang]

No problem! Being a grad student who also does a bunch of geoscience and software development, I’d love to see this sort of software projects taking off.

Not sure how relevant it is, but you might also want to take a look at Xarray (http://xarray.pydata.org) and Pangeo (https://pangeo.io/) that are widely used in cloud environments. There are some important similarities between Xarray and VTK/VTKI – both represent multi-dimensional gridded data; both are used for geoscience but also for other fields, etc. The two packages might be able to learn from each other. For example:

• The Pangeo community are somewhat struggling with the data model for unstructured grids (pangeo-data/pangeo#356). VTK seems to have very powerful unstructured-grid-related APIs (plotting, slicing, subsetting) that we can learn from (I am speaking of the Xarray community here).
• On the other hand, I think Xarray has the right data model for multi-variable datasets (the Dataset vs DataArray abstraction). For vtki, a non-ideal feature to me is that switching between variables in a vtk.vtkDataSet requires a somewhat implicit set_active_scalar() call. Following Xarray’s data model, a dataset['variable_name'] call will extract a single variable while keeping the metadata (e.g. grid coordinates that can be used for plotting). In vtki, dataset['variable_name'] gives a pure numpy array without metadata. If it can instead track grid coordinates, then plotting a variable will be as simple as dataset['variable_name'].plot(), like in Xarray.

[banesullivan]

Ah, cool! We may have a number of users that would like that. Thanks for sharing!