You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
I want to add the instance to submit a job to the Fugaku job scheduler #525
I am postdoc in RIKEN R-CCS. I want to add the instance that use with the Fugaku supercomputer at RIKEN R-CCS. The code is already finished. Can you guys give me a permission to push a branch and I will create a pull request?
You can create a fork of PSI/J, create a branch in that fork, and then submit a PR to the main PSI/J repository.
Alternatively, you can distribute the package separately. If the PSI/J descriptor file is present in your package, PSI/J will pick it up automatically if it's installed. I would recommend this since that would allow you to maintain and release updates to the executor without having to go through the main PSI/J repository every time.
I think put the code in the main repository is better. As one of the requirement I receive from my superior, there are researchers from other laboratory from the US site or within Japan use the Fugaku too. They can use PSI/J instantly without instruction from us and we/I don't have to regularly maintain/update the code too.
They can use PSI/J instantly without instruction from us and
I don't think distributing it separately prevents that from happening. All they would need to do is pip install psij-fugaku where psij-fugaku is the executor package which should have the main PSI/J as a dependency.
we/I don't have to regularly maintain/update the code too.
You or somebody from your group would have to maintain the code, whether in the main repository or as a separate package. That is because we do not have access to Fugaku's scheduler and we are fairly short on manpower. And it is significantly easier to maintain a separate package in this case.
I will submit a PR soon.
Let's try a separate package first please. If that turns out to be a problem, we can talk about adding it to the main repository.
I need to submit the manuscript soon and I barely use git. Can you write the processes on what you want me to do? I will be back after I finish my manuscript.
One thing
we/I don't have to regularly maintain/update the code too.
You or somebody from your group would have to maintain the code, whether in the main repository or as a separate package. That is because we do not have access to Fugaku's scheduler and we are fairly short on manpower. And it is significantly easier to maintain a separate package in this case.
My reply was misleaded. I mean I still will responsible for the Fugaku's scheduler instance as long as I work in RIKEN. The part that I don't want to get into is the core of the PSI/J. Just from when I start the implementation of Fugaku's scheduler instance until finished, there are a few change already. I don't want to regularly fetch to update the my branch, by make it dependency is understandable but this will be the first time I doing it. If you kindly leave the instruction or process to do it, it would be ok.
Activity
hategan commentedon Jul 10, 2025
Thank you @kira42310.
You can create a fork of PSI/J, create a branch in that fork, and then submit a PR to the main PSI/J repository.
Alternatively, you can distribute the package separately. If the PSI/J descriptor file is present in your package, PSI/J will pick it up automatically if it's installed. I would recommend this since that would allow you to maintain and release updates to the executor without having to go through the main PSI/J repository every time.
kira42310 commentedon Jul 10, 2025
I think put the code in the main repository is better. As one of the requirement I receive from my superior, there are researchers from other laboratory from the US site or within Japan use the Fugaku too. They can use PSI/J instantly without instruction from us and we/I don't have to regularly maintain/update the code too.
I will submit a PR soon.
hategan commentedon Jul 10, 2025
Thank you Soratouch.
I don't think distributing it separately prevents that from happening. All they would need to do is
pip install psij-fugakuwherepsij-fugakuis the executor package which should have the main PSI/J as a dependency.You or somebody from your group would have to maintain the code, whether in the main repository or as a separate package. That is because we do not have access to Fugaku's scheduler and we are fairly short on manpower. And it is significantly easier to maintain a separate package in this case.
Let's try a separate package first please. If that turns out to be a problem, we can talk about adding it to the main repository.
@andre-merzky, thoughts?
andre-merzky commentedon Jul 10, 2025
I agree that a separate package seems a better fit. @kira42310 : let us know if you would like us to help with setting this up!
kira42310 commentedon Jul 11, 2025
I need to submit the manuscript soon and I barely use
git. Can you write the processes on what you want me to do? I will be back after I finish my manuscript.One thing
My reply was misleaded. I mean I still will responsible for the Fugaku's scheduler instance as long as I work in RIKEN. The part that I don't want to get into is the core of the PSI/J. Just from when I start the implementation of Fugaku's scheduler instance until finished, there are a few change already. I don't want to regularly fetch to update the my branch, by make it dependency is understandable but this will be the first time I doing it. If you kindly leave the instruction or process to do it, it would be ok.