Repository navigation
HPC and SLURM
Rule-level threads/memory/runtime defaults live in
workflow/default-config/resources.yaml. Override any of them by creating
input/resources.yaml with just the keys you want to change (partial
overrides are deep-merged). Unlisted rules fall back to conservative defaults
(1 thread / 4 GB / 60 min).
To submit to SLURM, run snakemake with the bundled workflow profile (from the repo root, like any snakemake invocation):
snakemake --workflow-profile workflow/profiles/slurm --configfile config/test.yaml # smoke test on SLURM
snakemake --workflow-profile workflow/profiles/slurm --cores 64 # limit total cores
snakemake --workflow-profile workflow/profiles/slurm -n # dry-runThe profile (workflow/profiles/slurm/config.yaml) submits every job with
sbatch and sizes each one from input/resources.yaml.
slurm_account in the profile is pre-set to the ICMM_DM group account; replace
it (and qos) with your cluster's values if you're not in that group.
slurm_partition is left unset on purpose, so sbatch uses the cluster's
default partition (if yours has none, add slurm_partition). Per-invocation
overrides are also possible, e.g.
--set-resources star_align:mem_mb=64000. Every tool runs from the shared
conda env, so make sure it's visible from the compute nodes — if nodes can't
read your home dir, install on shared storage
(conda env create -p /shared/path/env, or
./workflow/scripts/install-env.sh -o /shared/software/tetranscripts-pipe-env)
and put its bin directory on the nodes' PATH.