Discussion on "docker image to cloud" tools

[behrica]

Dear Scott,

I was just reading #47 from you, and I am working on similar idea and just wanted to discuss with you.

My requirements are:

• assuming we distribute reproducible research papers or code as docker images to have easy sharing
• I want to be able to run a docker image, by ONLY using my web browser.So taking a docker image (by just knowing the name of the image) and no need to install anything on my PC except a web browser
• This means I need a cloud server somehow, digital ocean as an example, and have a web based tool running somewhere, which takes the docker image name and digitalocean credentials and spins up a new droplet, install and runs the docker image (Rstudio server based)
• After this I can login into Rstudio and work on the research article in exactly the same environment then the original author

I have a very basic application (done in clojure, hosted here: https://blooming-lake-3277.herokuapp.com), which does that. (absolute minimal features so far)

By looking at your API, I just realized that doing an R package which does that (instead of my web application), would work as well. So just using RStudio server as a "universal shell" in a browser window ...

So it would boil down to having a function in R, like

startNewDockerInCloud ( dockerImageName, digitalOceanToke,digitalOceanSshKey,....)

It would just do some simple calls to the digitalocean API, as you can see in my code: https://github.com/behrica/ropensciCloud

I use CoreOS as droplet base image, which allows to pass user_data to the droplet, to start automatically one docker image on boot.

What do you think ?

Maybe as a remark. I will concentrate on digitalocean now, but in the future I might want to extend it, so it can work with different cloud providers (Amazon EC2)

[hadley]

Duplicate of #47? What more do you want from analogsea?

[hadley]

(Sorry, didn't mean to close it without further discussion)

[sckott]

Hi @behrica - Thanks for the comment. I see now that you've been engaged in the Docker discussion on our mailing list, good to see you here.

Yeah, I think that's a neat idea that you are working on. Is there anything you are proposing for this package?

[cboettig]

@behrica Sounds like you're suggesting we take a look at CoreOS for our droplets? I'm not really familiar with coreOS (but reading this now, sounds cool!). Maybe you could highlight some of the advantages in using CoreOS here? (e.g. is the goal just a simple light set-up or more about networking multiple machines?)

(right, this package essentially lets the R user sidestep the need know a the shell commands to launch a docker container on a droplet, and monitor the droplet.)

[behrica]

@cboettig I quickly googled before and found the advantages of CoreOS to be:

1. more light wait (they claim to use 40% less memory then Linux)
2. More importantly : CoreOS on digitalocean is the only image which support passing of user_data into the image at boot time. I can use this very elegantly to start a docker container at boot time. See here: https://github.com/behrica/ropensciCloud/blob/master/resources/user_data.txt

So I do not need to ssh into the machine.

I knew the concept of user_data from Amazon EC2 and find it very powerful. Even though the same can be achieved with SSH.

[behrica]

Discussion continues on #17