I often use the "local" executor in my pipelines since I don't often require more than a few tens of cores for them to run efficiently.
However, I recently embarked in developing one that could (potentially) run more than 30k independent jobs at the same time.
Each of these jobs is very cheap: 1 cpu, little memory. The pipeline works fine if I run it with the "local" executor, but my nodes have only up to 48 CPUs, so it takes forever.
I am wondering if there is a way to trick nextflow into thinking that 4 nodes are a single node where to spawn jobs "locally". I am not an expert when it comes to mpi & co, so I would welcome any direction / tip.
Matteo