Plots not showing for docker image build on Ubuntu 24.04 LTS

[Alex-J-Brown]

I've built the docker image version of the pipeline on my laptop running Ubuntu 24.04 LTS. The pipeline itself installs fine but I haven't been able to get plots to show (i.e. setaper or hplot). I guess this is more to do with docker than it is the pipeline but I still thought it was worth mentioning since it obviously affects the ability to use the pipeline for people who've installed with docker. I've tried various things from altering the docker file to use different matplotlib backends to setting "DISPLAY=host.docker.internal:0" and/or "xhost +" and a few other straws that I clutched at with no luck.

Any ideas?

[Alex-J-Brown]

Please note that this is not urgent at all from my perspective, I installed it via docker to test it so that I can help with other people's installs. I have the pipeline installed natively.

[AstroTeutloff]

I second this, with on M3 Mac architecture. (Though this issue seems to be more inherent to docker window forwarding instead of this particular install.)

[AstroTeutloff]

Okay, I figured it out:

On Mac/ Apple Silicon:

You need XQuartz installed, for the X11 windowing system.

1. Open the XQuartz settings and navigate to the "Security" tab.
   1. Enable "Allow connections from network clients"
   2. Close and restart XQuartz
2. In a terminal, enter ps aux | grep Xquartz.
   It should display something similar to
   felix 94303 0.0 0.1 410676368 11856 ?? S 11:33AM 0:00.01 /opt/X11/bin/Xquartz :1 -listen tcp -iglx -auth /Users/felix/.serverauth.94141

Make note of 2 things here: The :1, for the displaynumber, and the -listen tcp flag. I'm guessing, but the :1 could be any other integer as well, depending on your display setup.
If the flag states -nolisten tcp instead, step 1 didn't work correctly. If this happens again after trying to disable (restart) and enable (restart), this stackexchange post has helped me. (TLDR: you can enable it by running defaults write org.xquartz.X11.plist nolisten_tcp -bool false)

3. If step 1 and 2 were successful, it should now be possible to pass a window through the docker process by:

   xhost +localhost
   docker run -it --rm -v ./:/home/hiperuser/data -e DISPLAY=host.docker.internal:1 --platform linux/amd64 hipercam:latest

Here, the :1 needs to be the displaynumber as shown in step 2. I also added xhost +localhost instead of xhost + as the latter would allow any network connection to access and I am not sure about the security concerns of that.

I got my solution from this page:
https://gist.github.com/sorny/969fe55d85c9b0035b0109a31cbcb088

On Linux:

I haven't tested this, but I am going out on a limb and assume that Wayland is (as usual) the problem here (Or rather the nonexistence of a running X session).

The first thing to try is to run the session though a X11 based desktop session (i.e. reboot the computer and change the session upon first login to e.g. GNOME on Xorg) and try again.

Other than that, the w command can/should show your displayname in the "FROM" tab.
See this stackexchange post for "clarification"(?).
With this you can maybe find the correct display number under which X is running and then this can be added to the docker run call.

Edit: Small typo in the command for setting the defaults. Its nolisten_tcp not notlisten_tcp.

[StuartLittlefair]

@AstroTeutloff Brilliant! Thanks for working this out.

@Alex-J-Brown can you test this suggestion (use w to get your display number and replace DISPLAY=host.docker.internal:0 to DISPLAY=host.docker.internal:#NUM where #NUM is the correct display number?

If it works I'll update the README

[Alex-J-Brown]

I'm booted into Xorg.

Display seems to be :0 although the command w doesn't give me that.

Running the docker image as mentioned in the README doesn't give an error message but no plot appears (the original issue)

Running the docker image with -e DISPLAY=host.docker.internal:0 does give an error message and no plot appears

This behaviour is the same after running xhost +localhost

[StuartLittlefair]

Do you have another X-window open? On MacOS I need XQuartz open first.

[Alex-J-Brown]

I'm not particularly sure what that means but attempting the same commands using an xterm terminal or in the normal gnome terminal after opening an xterm session makes no difference to outcome of the docker command. I have no idea if it's relevant but the result of echo "$XDG_SESSION_TYPE" is "xcb" instead of "X11"

[AstroTeutloff]

Can you run ps aux | grep "X"? This should give you your X window session

[StuartLittlefair]

It's possible that you don't have the authority for X11 forwarding. Do you have a directory /tmp/.X11-unix? If so this is the location of the X11 display socket and you can add -v "/tmp/.X11-unix:/tmp/.X11-unix:rw" to mount the X11 Unix socket from your laptop to the container.

Similarly, try adding -v "$HOME/.Xauthority:/root/.Xauthority:rw"

[StuartLittlefair]

Finally - can I ask how you installed docker? According to this thread, Docker installed with snap will not work.

[Alex-J-Brown]

adding either or both of those two -v paths don't change anything. I'm pretty sure I didn't install with snap, instead using the .deb DockerDesktop package on the docker website

EDIT: apt list --installed shows docker and snap list doesn't so it was definitely not installed as a snap

[Alex-J-Brown]

I notice that -nolisten tcp flag is set which your mac instructions say should be changed
alex 3589 8.4 0.8 817508 133992 tty2 Sl+ 10:44 4:17 /usr/lib/xorg/Xorg vt2 -displayfd 3 -auth /run/user/1000/gdm/Xauthority -nolisten tcp -background none -noreset -keeptty -novtswitch -verbose 3 alex 15847 0.0 0.0 9152 2316 pts/2 S+ 11:36 0:00 grep --color=auto X