Closed pat-s closed 5 years ago
You are right, this is currently not possible.
One way to support this would be to add an SSH template instead of hardcoding the commands, where you could supply the R_PROFILE_USER
environment variable and thus set any option you want.
Another option might be to pass the clustermq.defaults
variable via SSH.
Can you expand a bit on your use case? Like, are there different queues per project?
I think option 2) might be more intuitive for the user.
Send jobs of different projects via SSH.
Each project has it's packrat
library and (potentially) a different R version.
So to use a project specific packrat
library, one needs to start R from the root of the project directory with the correct R version.
These commands (usually a simple cd
and module load <R interpreter>
) are set in the slurm_clustermq.tmpl
file. However, currenltly always the template specified in the default R interpreter is chosen. I would like to be able to point to a specific template file when using the SSH approach.
This is already possible when directly initializing from the frontend because then a project specific .Rprofile
can be used which again points to a custom template file. This custom .Rprofile
also loads packrat and a specific R version.
@pat-s I now moved the SSH command to a template with the latest develop
.
You should be able to
#5
"cd {{ project_dir }} && R --no-save --no-restore -e [...]"
I'm not sure if I want to pass defaults via SSH, because the user also may want different defaults for different machines (and this would override it).
Similarly, a custom template can be used with options set from the environment variable R_PROFILE
pointing to an R profile file.
@mschubert Thanks a lot for this feature!
When using the SSH way, the scheduler template specified in the default remote R interpreter is used. Is there a way to set a custom path (preferred locally) to a scheduler template which should be used on the remote machine?
Use case: Deploy jobs of different projects (with different scheduler settings) using the SSH way. Currently the only way to do this is to manually update/edit the path to the template in
.Rprofile
of the default R interpreter on the HPC (unless I overlooked something :horse: ).