Sys Admin Pocket Survival Guide - Singularity
Home Docker Kubernetes/Rancher CoreOS AWS BigData DevOps OpenStack Vagrant Container in HPC lsf/slurm/uge ansible

Singularity - Container for HPC, Science

Singularity 2.4 commands

see bottom of: Container in HPC

Singularity 2.6 commands

Ref: 2.6 doc at sylabs.io 
sudo singularity build pgi_netcdf.img pg_netcdf.def
sudo singularity build myimg Singularity 
sudo singularity build --writable metabolic.sif Singularity 2>&1  | tee singularity_build.log

# assuming Singularity is the .def file

Note that singularitypro26-runtime rpm does not support the "build" subcommand

Singularity 3.0 commands

Singularity 3.0 is a new rewrite in GO, it is not backward compatibile with 2.x.
3.0 doc at sylabs.io 

# read-only image
sudo singularity build ./ansys.sif ./Singularity

# sandbox would create writable directory
# cd /tmp (or other dir that is root writable)
sudo singularity build --sandbox ./ansys.sif ./Singularity  2>&1 | tee singularity_build.log



# pull has a centralized cache, but image still be written to `pwd` multiple times 
singularity pull shub://tin6150/r4eta    
singularity pull --name myR shub://tin6150/r4eta # store in pwd, 
singularity cache list -v

singularity exec myR 
singularity exec myR bash
singularity exec myR rstudio
singularity exec shub://tin6150/r4eta rstudio

# pull docker container, from github container repository
singularity pull  docker://ghcr.io/tin6150/r4eta:master
singularity shell docker://ghcr.io/tin6150/r4eta:master

singularity pull  docker://ghcr.io/tin6150/r4eta:master  
singularity shell docker://ghcr.io/tin6150/r4eta:master  # get bash prompt 
singularity run   docker://ghcr.io/tin6150/r4eta:master  # get R    prompt

singularity exec  docker://ghcr.io/tin6150/r4eta:master  Rscript ./drJin.R                    # R script in current working dir
singularity exec  docker://ghcr.io/tin6150/r4eta:master  Rscript /opt/gitrepo/r4eta/drJin.R   # R script inside container




running with gpu: --nv

singularity pull --name beast2.6.4-beagle.sif  docker://ghcr.io/tin6150/beast2:dock264-beagle
singularity exec --nv   beast2.6.4-beagle.sif /usr/bin/java -Dlauncher.wait.for.exit=true -Xms256m -Xmx8g -Duser.language=en -cp /opt/gitrepo/beast/lib/launcher.jar beast.app.beastapp.BeastLauncher -beagle_info

Singularity image format

  1. .simg - single file image from Singularity Hub prior to 3.0
  2. .sif - singularity 3.0 single file image, not directly writable
  3. [DIR] - 3.x --sandbox ./DIRNAME, writable 



hpl on gpu linpack from nvidia container






ref: https://catalog.ngc.nvidia.com/orgs/nvidia/containers/hpc-benchmarks


singularity pull --docker-login docker://nvcr.io/nvidia/hpc-benchmarks:21.4-hpl

export SIMG=$cacheDir/nvidia-benchmark/hpc-benchmarks_21.4-hpl.sif  
export DAT=~/gs/CF_BK/greta/benchmark_gpu/HPL-a10-Ns=20k.dat

singularity exec --nv $SIMG \
  mpirun --mca btl smcuda,self -x UCX_TLS=sm,cuda,cuda_copy,cuda_ipc  --bind-to none -np 2 \ 
  hpl.sh  --xhpl-ai  --cpu-affinity "0-7:8:15" --gpu-affinity "0:1" --cpu-cores-per-rank 8 \
  --dat $DAT

## above run, seems loaded on both gpu, but load is 100% & 0%    need to adjust PxQ?
















  [Doc URL: http://tin6150.github.io/psg/singularity.html] 


(cc) Tin Ho. See 
main page
 for copyright info. 









  hoti1

  bofh1