This is not specific to tunel / sregistry but to singularity, we need to update the Dockerfile here to use the fix that is provided by @dctrud, the original issue is described here apptainer/singularity#1091

Hello Vanessa,

I installed Tunel and I was able to lunch the GUI interface. But everything is super slow to the point that it's not usable. Any idea what I could be missing? Clicking any button takes more than 2 min to load.

Thanks

A user might start the container again, and need to re-authenticate and completely forget about the command line.

The interface should have the same living documentation (linked to Github / issues) as the other containertools.

and add settings page information, particularly about nvidia tokens

I just made it, and already I see too much unnecessary text, colors, etc. Clean this up!

Does Globus immediately take the file (upload) to some staging area, so we can confidently delete within some random of time when the transfer is submit?

See globus/globus-sdk-python#286 (comment)

singularity... is still broken

I should be able to...

• to transfer / share images (e.g., Google Drive share or to a Globus endpoint)
• to rename / move containers
• delete and inspect individual containers
• some kind of sync?

This is valid Dockerfile syntax to set 2 environment variables:

ENV A=B X=Y

When converting this from Docker to Singularity, the Singularity output is incorrect:

%post
A=B=X=Y
%environment
export A=B=X=Y

It is common in Docker to chain commands together into a single layer and break the command so it does not exceed 80 characters. For instance:

RUN apt-get update -y && \
  apt-get-install -y --no-install-recommends \
     python \
     python3 && \
  rm -rf /var/lib/apt/lists/* 

This is converted to Singularity as

%post
apt-get update -y && \
apt-get install -y --no-install-recommends \
python \
python3 && \
rm -rf /var/lig/apt/lists/*

The builder then interprets this as 5 effective commands rather than 1 long command split over 5 lines. The builder fails when trying to run python.

If I am a Stanford student, I would want to be able to open up the interface, pull a container from Nvidia GPU cloud or Docker Hub, and then transfer it easily to my endpoint on Sherlock.

We likely can integrate many of the Google Cloud endpoints via OAuth2 and skip over the application credentials. This will allow for use cases like sharing / transferring to and from Storage, and Drive.

Given the client's choice of session in the browser, I can see a bug where the token is not saved for Nvidia GPU. This cache should use the standard sregistry, and not rely on sessions.

I had first planned to add the Singularity <--> Dockerfile translation tool, but now with recipe maker we can consider this! https://github.com/NVIDIA/hpc-container-maker

As a user, I should be able to create a recipe, and then click build, and have a builder launch and (send me a message to my browser) to notify that the build is done. I should be able to select from the Builders repository of recipes, or my own. It also might include integration with Github for version control recipes, need to think about this. Another option it to have a link between the interface --> Github --> CI and building.

Each container, along with the entire list of containers, should be served via a restful API of the server. This API then might be able to be interacted with by other users (I wonder what cool things we could do with this!)

Each container in the table should be clickable, with a detailed view.

posted this in the wrong repo :) This is suggested by @samcmill

• do something once, it doesn't work
• spend all day debugging and not getting help, hate self
• get immediately closed ticket or otherwise unhelpful response back, give up and still hate self
• come back with redeemed motivation and figure it out, thank god
• but then come back to do the same thing again, and it's broken again
• 7 hours later, continue hating self
• question wtf am I doing this for
• jump off cliff

This is valid Dockerfile syntax to name a layer (for multistage builds):

FROM ubuntu:16.04 AS level

Singularity has no corresponding concept. The converter carries it through which results in an invalid Singularity recipe.

Bootstrap: docker
From: ubuntu:16.04 AS devel