Kernel dying out repeatedly

[slicer]

Kernel dying out repeatedly

[lassoan]

Can you check if you can launch Slicer successfully on that system (without Jupyter) and maybe get a stack trace?

[tashrifbillah]

Okay, it looks like it isn't even starting properly ... dying right after connecting.

[lassoan]

Most common issue in a server environment is wrong X or OpenGL configuration. See all the questions on the Slicer forum about running Slicer on a headless system.

[tashrifbillah]

Can you check if you can launch Slicer successfully on that system

It does.

running Slicer on a headless system.

I know about compatibility mode we discussed on Slicer forum. Even if that is the issue, which is not supported by the above log, how may that help me solving this issue?

[lassoan]

@pieper I guess your experience is needed here.

[pieper]

@pieper I guess your experience is needed here.

Sorry, I've never used a Partners Jupyter server, but I probably have access if you need help debugging. @tashrifbillah can you describe exactly what system you are using?

[tashrifbillah]

• grx04.research.partners.org-- a CentOS 7 node through NoMachine
• module load libGLU/9.0.0-foss-2018b
• https://jupyterhub.partners.org

[tashrifbillah]

@pieper if you please hold, I am looking at ~/.local/share/jupyter/kernels/slicer-4.11/kernel.json and might have found the problem.

[tashrifbillah]

Never mind.

If I try the following on a Linux terminal:

[tb571@grx04 slicer-4.11]$ cat kernel.json
{
    "display_name": "Slicer 4.11",
    "language" : "python",
    "argv": [
        "/data/pnl/soft/pnlpipe3/Slicer-4.11.0-linux-amd64/bin/../SlicerWithExtensions.sh",
        "--no-splash",
        "--python-code",
        "connection_file=r'{connection_file}'; print('Jupyter connection file: ['+connection_file+']'); slicer.modules.jupyterkernel.startKernel(connection_file);slicer.util.mainWindow().showMinimized()"
    ]
}

I get this:

QStandardPaths: XDG_RUNTIME_DIR not set, defaulting to '/tmp/runtime-tb571'
Switch to module:  "Welcome"
Jupyter connection file: [{connection_file}]
startKernel connectionFile does not exist "{connection_file}"

Is this expected?

[lassoan]

Jupyter replaces {connection_file} in kernel.json with the actual connection file. You need to do the same if you want to test it by running the command yourself.

[tashrifbillah]

Hi @pieper , did you get a chance to look at it?
In the meantime, I found Partners VPN connection instruction which was easy to follow on my Windows laptop. After connecting through VPN, you can log into https://jupyterhub.partners.org/ from your laptop browser. Once you are in https://jupyterhub.partners.org/, you can open a terminal on the Notebook and follow SlicerJupyter installation instructions.

This is a way around NoMachine in case you don't have that setup. Happy to work with you over a call as well.

[tashrifbillah]

Hi @lassoan , is there a log where I may find the events during kernel initiation and failure that I described in the above?

[pieper]

Hi @tashrifbillah - I don't really use partner's jupyterhub or slicerjupyter much myself, so not sure I can do anything other than you will be able to do. Let's see what @lassoan suggests for debugging techniques and I can jump in if needed.

[lassoan]

Slicer logs everything into the Slicer log files.

[tashrifbillah]

I confirm that it is an OpenGL issue. https://jupyterhub.partners.org/ X environment is probably not set up properly.

[tashrifbillah]

Hi,

I am still trying to find a way out of this problem.

The following pops up Slicer-4.11 GUI and stays there:

(base) [tb571@grx05 slicer-4.11]$ jupyter console --kernel slicer-4.11
Switch to module:  "Welcome"
Jupyter connection file: [/PHShome/tb571/.local/share/jupyter/runtime/kernel-55305.json]
Run with XEUS 0.23.13
Starting Jupyter kernel server
Jupyter console 6.1.0

  __  _____ _   _ ___
  \ \/ / _ \ | | / __|
   >  <  __/ |_| \__ \
  /_/\_\___|\__,_|___/

  xeus-python: a Jupyter kernel for Python
  Python 3.6.7
In [1]:                                                                                

Is there a way to suppress that?

[tashrifbillah]

The following pops up Slicer-4.11 GUI and stays there

The same happens if I open the kernel from http://localhost:8889/tree .

[lassoan]

Can you post a screenshot?
Is the problem that the text is displayed or that something does not work correctly?

[lassoan]

The ascii word art is displayed the Jupyter client calls kernel_info_request. I don't remember any of the Jupyter clients calling this or at least it did not display the result. What Jupyter client do you use?

[tashrifbillah]

What Jupyter client do you use?

(base) [tb571@grx05 slicer-4.11]$ jupyter --version
jupyter core     : 4.6.3
jupyter-notebook : 6.0.3
qtconsole        : 4.7.4
ipython          : 7.8.0
ipykernel        : 5.3.0
jupyter client   : 6.1.3
jupyter lab      : not installed
nbconvert        : 5.6.1
ipywidgets       : 7.5.1
nbformat         : 5.0.7
traitlets        : 4.3.2

[tashrifbillah]

I guess my question is about the default behavior-- if I open a kernel, should Slicer GUI be open and stay minimized in the taskbar?

Are these the screenshots you asked for?

Minimized Slicer-4.11 on the taskbar

---

Maximized the minimized Slicer-4.11

---

[tashrifbillah]

Is the problem that the text is displayed or that something does not work correctly?

The problem is, we do not have any GUI access to the server other than Jupyter notebook where Slicer-4.11 Kernel would be running. And coming back to the issue I opened, the kernel fails to start attempting to launch Slicer-4.11 and failing.

[tashrifbillah]

Well, I think that is the default behavior as specified by "slicer.util.mainWindow().showMinimized()":

  "/PHShome/tb571/tmp/Slicer-4.11.0-linux-amd64/Slicer",
  "--no-splash",
  "--python-code",
  "connection_file=r'{connection_file}'; print('Jupyter connection file: ['+connection_file+']'); slicer.modules.jupyterkernel.startKernel(connection_file);slicer.util.mainWindow().showMinimized()"

[lassoan]

I don't remember ever seeing this XEUS graphics, but Slicer starting up minimized is the expected behavior. You don't need GUI access to the server, you should be able to do visualizations, processing, etc. all just using the Jupyter interface. The server has to provide minimal graphics resources, but if even the free binder service can meet these requirements it is hard to imagine how a dedicated, more powerful server would have any problem with this.

[pieper]

@tashrifbillah are you able to run Slicer on the remote server using Xvfb or Xdummy?