Unable to get interactive desktop to render


#1

We are using this doc to try enable interactive desktop. However, once we have put in place the suggested config it still does not appear.

Are there some gotcha’s that we are missing?


#2

it still does not appear

Is the problem that the desktop app itself doesn’t appear in the dropdown list? If so, there is likely an erb error. You should be able to direct your browser directly to the interactive app by entering the URL in the browser. If the form.yml is named “owens.yml” or “owens.yml.erb” then the URL would be https://your.ondemand.edu/pun/sys/dashboard/batch_connect/sys/bc_desktop/owens/session_contexts/new

Of course replace owens with whatever the name of the yaml file is under /etc/ood/config/apps/bc_desktop/


#3

We were able to get this to render. The issue we were running into was trying to display the desktop to test against a virtualized instance that did not have a job manager. Therefore, in the docs on setting up the jobs we skipped this line line.

  job:
    adapter: "..."

We looked into the code and saw the logic that looks to see if there is a job adapter defined and realized it was required. A.k.a our bad.


#4

So maybe we need to introduce an “install mode”, sort’ve like an ephemeral “debug mode” for the production environment. For instance, lets say that new rpm installs (not updates) created a /etc/ood/config/INSTALL file and as long as this file existed the mode was “install mode”.

Or maybe just a /install page on the Dashboard that shows all the apps that are found, any apps that the current user cannot launch, any clusters configured and what is properly configured and what configurations are missing, any cluster config files that have errors in them (or whose attempted parsing results in an error).

Any thoughts on what would have made your life a whole lot easier in this process?


#5

We are currently taking notes while we try get the interactive desktop running.

I will write a reply based on our observations.


#6

Apologies for the delay, had to shift focus.

After reviewing the process we realized that our Slurm configuration was 90% of the problem. Our Ansible Role to setup Slurm the incorrect permissions on the /var/spool folder.

Secondly, because we didn’t have much expertise with Slurm we tried to circumvent it and left it out of the cluster.d file.

Third, the regex that we had setup did not seem to like our config and was the final issue we encountered. If noVNC is not connecting to server this could be one of the issues.

We feel that the suggestions you are good ideas, personally I would prefer the second option it seems to cover debugging issues we encountered and would be the better long term configuration option for a web app in my opinion.