Hi Galaxy people,
At the University of Oslo, we have an infrastructure for storage and computation on sensitive data, called TSD (presented in the attached image). from inside, the infrastructure is logically divided into projects. Each project has a set of shared VMs (among
its users) which are running over a file-system (HNAS). The shared project VMS has access to a computational SLURM cluster which is installed on another file-system (Colossus). There are directories in colossus which are mounted on HNAS for each project in
order to allow each project users to push files into colossus and run jobs on the cluster. Users of a particular project can access the project main VMs through user VMs (one for each user). Each project VMs (both shared and user VMs) are in a separate subnet.
All of this is inside the TSD. Now to access the TSD from outside, there is a complex authentication mechanism where a user can access his/her own VM. And to transfer data from/to the TSD is another complex story. The important thing is that there is NO
internet access in or out.
There are two issues here:
1 - What we need to do is to install one Galaxy VM inside each project area, so that it is accessible by all project users. But we cannot use mercurial to access your distribution server. We can though install a bitbucket server inside the TSD and have the
code-base there, so that It can be accessed by all project VMs, but I'm not sure what is the procedure here.
2- We are very concerned about the issue of regularly updating Galaxy instances in projects to the recent release. In many cases it causes many problems, e.g. tool versioning conflicts. So we have the idea of installing each of our tools together with all
of its dependencies in a separate docker container, and run those as images on each Galaxy project VM. Is this possible and tested? Should this permanently solve the upgrading problem, or do you suggested another alternative?