If you are experimenting with scheduling changes, you should only need to perform the install of changed components on the head node (where the slurmctld runs). You can just do a "make install" of individual components as needed (e.g. individual plugins).
More information about building and installing Slurm is available here: http://slurm.schedmd.com/quickstart_admin.html Quoting Abhishek Gupta <[email protected]>:
Hello, I am extending some of the job scheduling policies in SLURM, but I am having a hard time testing it. Currently, I am using rpmbuild and install to get my changes to the cluster and then test them those. However, this whole cycles takes long time, and even for small changes I need to do this whole process. Can I do something different to avoid having to do this? I think unit testing would be one option, but I also want to run and test on an actual deployment, because I am doing some interaction between job scheduler and applications, which cannot be done by simulation or unit testing. Thanks, Abhishek CS-Ph.D student, UIUC
