I have been working with Docker as a way to create reproducible analysis images for bioinformatic sequence analysis. I was initially trying to make an uber-image with all imaginable tools but I ran into the 10GB image limit that there currently is for the devicemapper driver.
What I am exploring instead is using smaller analysis or pipeline specific images, with locally mounted home drives. This way we can mount local file locations and read from and write to the local file system . Her is a simple example of the Augustus gene modelling software in a Docker image that I wrote : https://index.docker.io/u/tlaurent/docker-augustus/ .
One other requirement is that we have a GUI desktop environment of some sort that can be accessed with xRDP . Luckily docker containers can be nested jpetazzo/dind -- So the idea is to have bare metal or cloud servers that act as Docker hosts on which we run the GUI desktop environment. This host will also run docker and be able to pull any of our Docker images.
I am using the [phusion/baseimage] (rep too low) which has its own init system and runs an ssh daemon mixed with the nested image build.
I was unable to get gnome working over xrdp , KDE-full started loading but then seg faulted, however LXDE installed and seems to be working well. https://index.docker.io/u/tlaurent/docker-base-lxde-xrdp/
A couple questions : - Can anyone think of problems with running a Desktop Environment in a Docker container ?-- It is strange that KDE segfaulted. - What would be the best practice for such a system -- multi user RDP environment?? Should all potentially large folders become volumes maybe host mounted? /home /var
Any thought or suggestions would be greatly appreciated. Anyone else able to get this sort of thing working?
