Add optional --app argument to apptainer instance start

[struktured]

Version of Apptainer

What version of Apptainer (or Singularity) are you using? Run

apptainer --version (or singularity --version).

apptainer version 1.1.5

Expected behavior

apptainer instance start --app app1 foo.sif foo_app

What did you expect to see when you do...?

INFO: app1 instance started successfully

With the idea that there was an %appstartscript app1 section which gets invoked from the definition file.

Actual behavior

Error for command "start": unknown flag: --app

[GodloveD]

Seems like a good idea to me. I believe that app support is added by playing fancy tricks with the environment at build time. Would you want to investigate further and perhaps submit a PR?

[struktured]

Seems like a good idea to me. I believe that app support is added by playing fancy tricks with the environment at build time. Would you want to investigate further and perhaps submit a PR?

I will investigate further and see if I can manage to implement it.

Thanks!

[struktured]

After adding some minor boilerplate in a few critical places I got it working!

However, I need to write a unit test and review the contribution guidelines before I open a pull request. The current diff is here.

[dtrudg]

Please could you loop in @vsoch on this proposal, or at least check if they wish to be involved?

The %app... directives are an implementation of a standard called SCIF (https://sci-f.github.io/)

The specification for SCIF defines the recipe / definition file sections, and there is currently no %appstart:

https://sci-f.github.io/spec-v1#sections

Elsewhere in the specification document there are defined apprun filenames and environment variable, that would also need appstart counterparts to be declared.

I believe that adding an %app... to the specification should be considered before it is implemented in apptainer.

I note that elsewhere, Apptainer is considering instance run or similar. It may be the case that an instance run --app with the existing SCIF app runscripts makes more sense then having separate run / start behavior per app. Likely some discussion needs to be had above the level of projects like Apptainer or SingularityCE that implement SCIF, to ensure that a consistent implementation of SCIF can be maintained w.r.t. any additional behaviour.

Thanks!

[vsoch]

Thanks for pinging me! I agree 100% this should be added to the spec first, and I’d be happy to do that. My 0.02 is that apprun corresponds to the runscript and appstart to the start script.

I’ll need a few days - here is the original paper if anyone is interested! https://academic.oup.com/gigascience/article/7/5/giy023/4931737

It’s truly an honor that the rebranded Singularity name was derived from this spec that I wrote and originally implemented.

@dtrudg when I’m done adding appstart we can add to Singularity too.

it’s 2am here so goodnight!

[vsoch]

I can get the spec PR in tonight after work - could someone from the Apptainer team please verify this plan is good?

[vsoch]

Please see sci-f/sci-f.github.io#13

[DrDaveD]

I can get the spec PR in tonight after work - could someone from the Apptainer team please verify this plan is good?

Sounds like a good plan to me.