I often run jupyter IPython notebooks on a linux cluster compute node without X11 with no problems. However, running the R kernel on the same setup is not working very well.
The machine details are:
Just starting the notebook causes the kernel to crash after the first cell is run with this in the log:
unable to open connection to X11 display ''
I can get it to work by starting the notebook with xvfb-run jupyter notebook. This lets me run R commands in the cells but when I try to produce plots, I get the following
Error in grid.Call(L_textBounds, as.graphicsAnnot(x$label), x$x, x$y, : X11 font -adobe-helvetica-%s-%s-*-*-%d-*-*-*-*-*-*-*, face 1 at size 9 could not be loaded
I suppose that if I could install the x11 fonts it would work, but this is on a compute node of a cluster and I don't have the administrative privileges to install them.
What is the correct way to configure jupyter notebook with the R kernel to produce graphics on a linux machine without X11?
I think that I've found the minimal configuration necessary for the irkernel when running the notebook on a linux machine with no display hardware and no physical input devices.
Run the notebook under the virtual framebuffer X server, xvfb:
xvfb-run jupyter notebook
Use cairo instead of X11:
# Run this in a notebook cell, or put in .Rprofile
options(bitmapType="cairo")
Set the jupyter.plot_mimetype. SVG looks much better and worked fine for me. PNG was ok too:
# Run this in a notebook cell, or put in .Rprofile
# svg much clearer, but won't rescale (scrolling works though)
options(jupyter.plot_mimetypes = "image/svg+xml")
# png had some artifacts, but had the nice feature that it would
# resize when the browser window changes size
#options(jupyter.plot_mimetypes = 'image/png')
# can easily resize plots (have to re-plot) with this:
#options(repr.plot.width=14, repr.plot.height=4)