Need some help with custom submission and custom script files together with bc_desktop app

[novosirj (Ryan Novosielski)]

Basically all we need to do is to change the following line in the bc_desktop/template/script.sh.erb:

source "<%= session.staged_root.join("desktops", "#{context.desktop}.sh") %>"

…and instead do something like this:

singularity exec ... image.img "<%= session.staged_root.join("desktops", "#{context.desktop}.sh") %>"

script_wrapper seems maybe to be a little too far upstream, but there may be some solution I’m not seeing that one of you might.

@jeff.ohrstrom

I can’t tell if you want to boot a desktop (or different types of desktops like mate or xfce) or you want to boot an application. The bc_desktop app is a convenience application quickly deploying desktops with different configurations (resoure allocations or into different clusters and so on). Where we’ve setup the submit.sh.erb scripts because they’re fairly generic and all we want is to start an xfce session for example.

We want to run a desktop, and by and large the existing options are fine, but need some changes because we don’t have some of the dependencies on all of our nodes, and therefore need to point to alternate locations for some of the dependencies (like VNC and websockify) and use Singularity to run an image that contains the desktop group, etc. The alternate location stuff should all be doable via environment variables set in the various YAML files, but the one command I’m not sure how to handle.

@jeff.ohrstrom

If I make some of those bits variable, it would be this. UUID is generated at runtime and completely dynamic for every submission, and APP_NAME is what you call your app.
$HOME/ondemand/data/sys/dashboard/batch_connect/sys/bc_desktop/$APP_NAME/output/$UUID . But that’s for the actual job submission, I’m guessing you want logs for the actual job creation which could be in /var/log/ondemand-nginx/$USER , though there may or may not be anything helpful here.

Had discovered both of those locations already; unfortunately nothing related to the job dying during creation in either one of them, but it sounds like what I was trying to do is not achievable via the method I was trying, so that might be moot.