James Fellows Yates
05/20/2024, 8:53 AMnextflow run command does under the hood is to write sbatch scripts for each step of the pipelije and submits tl SLURM for you. That way you can get really good paralisation and efficiency (ime. rather than running everything in a single job where you request let's say 100GB even though half the steps just use 1GB). So it's highly recommended to do so. The singularity default config (if I understand your question correctly) just says to load tools from singularity containers, that's it - it doesn't do anything with SLURM)
2. See above, you don't need to, but you should. You can still have the runner job being submitted to the cluster with the parameters. Ideally the institutional config just goes on nf-core/configs, so the user only has to handle with their own prams-yaml file and that's it - i.e. if the config is on nf-core/cpnfigs the user just specifies -profile <hpc_name> and that's it for the infrastructure settings. They don't have to specify a second conf8g
3. You don't have to buy indeed you will get a huge amount of redundancy. You should definitely set a cache though even if it's just on a per user basis - because otherwise each run will download the containers every single time. So sharing across users is even better