SSH with 2FA

[Aariq]

Recently my HPC turned on 2-factor authentication for SSH access. I had this working well for a while—running Q() from RStudio brought up a dialog box to enter an option for how to authenticate, then push went to the app on my phone. I'm not sure what changed, but now when I run Q(), the console hangs. Sometimes, after I try to terminate the stuck R process the 2FA dialog box pops up. Is there any chance this is something on the clustermq side, or is this likely to be an issue with the 2FA?

[Aariq]

I followed the debugging guide and got stuck with this step:

system("ssh -R 54709:localhost:6687 <user@host> R --vanilla")

It hangs, and when I interrupt the R session, the dialog box for 2FA comes up. If I add wait = FALSE it appears to work though

[mschubert]

Thanks, that likely also explains behavior in #206. I'm not sure what the best solution is yet, because if we set wait=FALSE a forward failure will not be caught.

There was no change on the clustermq side (CRAN) since 0.8.95 (July 2020).

[Aariq]

Ok, so my workflow with targets and tar_make_clustermq() is working again and I honestly don't know what changed or if something was actually broken. I wanted to comment though because the debugging step, system("ssh -R 54709:localhost:6687 <user@host> R --vanilla"), led me astray. Running this from R running in a terminal session works, but I need to enter an option for 2FA, like so:

> system("ssh -R 54709:localhost:6687 <user@host> R --vanilla")
(<user@host>) Duo two-factor login for <user>

Enter a passcode or select one of the following options:

 1. Duo Push to XXX-XXX-5555
 2. Phone call to XXX-XXX-5555

Passcode or option (1-2): 

I enter an option, then authenticate on my phone, and I'm in. But if I run this R code from R Studio either in the console pane, by clicking "Source" in the editor pane, or by running the line from the editor with command-enter, it hangs, then only after I terminate the R session does a dialog box pop up with the text above. If I save the code as a .R script and use "Source as a Local Job..." the dialog box pops up and the code runs. So the hanging I experienced with this debugging step wasn't actually related to clustermq working or not. It seems like it's maybe actually bug in RStudio.

[Aariq]

Oh, and tar_make_clustermq() works fine from the console, editor, or "source as a local job...", by the way.

[mschubert]

Thanks, this is helpful! Would it work for you if I change the recommendation to test system("ssh -R 54709:localhost:6687 <user@host> R --version")?

[Aariq]

This seems to have the same behavior. I'm guessing any shell command that prompts the user for something before it finishes does this in RStudio. I don't know enough about shell scripting to get a minimal reprex though.