Is this a familiar conversation?

• "I've got some really cool code, want a go?"


• "Yes sure, send it to me!"


• "Err, it's giving me an error?"


• "Well, it worked on my machine..."

• Docker, Singularity
• Open source tool for building and managing virtual environments in a single workflow
• Using a single file you can specify the exact environment required for your workflow
• Doesn't require an underlying Hypervisor (Virtualisation platform)
• Runs on Windows, macOS and Linux

• Virtual Machines vs Containers
• Containers vs Images vs Recipes
• Registries/Hubs

Basic Singularity commands

• Running a simple preexisting container

      
        $ singularity run shub://GodloveD/lolcow
      
    

      
        $ singularity run docker://godlovedc/lolcow
      
    

Singularity pull

• You're likely to want an actual copy of an image, as repeatedly requesting a remote image causes problems:

      
        $ singularity pull shub://GodloveD/lolcow
        $ singularity run lolcow_latest.sif
      
    

Singularity Definition File for lolcow(Recipe)

A simple definition for creating an image

      
        # Basic Singularity.def
        BootStrap: docker
        From: ubuntu:16.04

        %post
          apt-get -y update
          apt-get -y install fortune cowsay lolcat

        %environment
          export LC_ALL=C
          export PATH=/usr/games:$PATH

        %runscript
          fortune | cowsay | lolcat
      
    

Cleaning up Singularity

Cleaning up what you've done

      
        $ singularity cache list

        $ singularity cache clean

        $ rm lolcow_latest.sif 
      
    

Test all this code out yourself using the containers_examples directory in the GitHub Repo for this presentation

• Before I do anything, setup a vagrant VM as learnt in the previous TechTalk

        $ vagrant up
      

• Use a simple gnuplot container

      
        $ singularity pull shub://icaoberg/singularity-gnuplot
        $ singularity run --app gnuplot singularity-gnuplot_latest.sif
      
    

• Let's actually build a container

      
        $ singularity build bastet Singularity.def
      
    

• Using GPU - tensorflow

      
        $ singularity run --nv tensorflow_1.14.0-gpu-py3.sif ./bench.py
      
    

• Accessing host machine files on Containers

      
        $ singularity shell singularity-gnuplot_latest.sif
      
    

• Home directory and /tmp are mounted by default
• Additional paths can be mounted using -B

      
        $ singularity shell -B /nobackup:/nobackup singularity-gnuplot_latest.sif