The Wynton HPC cluster supports Linux containers via the Singularity software. A Linux container is an efficient and powerful virtualization method for running isolated Linux systems (“containers”) on any Linux system including CentOS used by our cluster.
Singularity, and Linux containers in general, provides solutions to common problems and hurdles that HPC users often face, e.g.
Because you can create and customize your own containers, and because Singularity also supports Docker containers, you have immediate access to a very large number of Singularity and Docker containers available via repositories such as:
All tasks for using Linux containers, such as downloading, building, and running containers, is done via the
singularity client and supported on Wynton HPC. The most common command calls are:
singularity run <image>- run predefined script within container
singularity exec <image>- execute any command within container
singularity shell <image>- run bash shell within container
singularity build <url>- import an existing Singularity or Docker container available online
singularity build --remote <image> <def file>- build a Singularity from local specifications in definition file
For full details, see
man singularity, and the Singularity website.
As an illustration on how to use Linux containers with Singularity, we will use the Docker container rocker/r-base available on Docker Hub. This particular container provides the latest release of the R software in an Ubuntu OS environment. Containers available from Singularity Hub, Biocontainers, and elsewhere, can be downloaded and used analogously.
To use this rocker/r-base container, we first pull it down to a Singularity image file
[alice@dev2 ~]$ mkdir lxc [alice@dev2 ~]$ cd lxc/ [alice@dev2 lxc]$ singularity build rocker_r-base.sif docker://rocker/r-base INFO: Starting build... Getting image source signatures Copying blob sha256:420047682034904224708ffcbc863130ba6bce56f9d642605dccd2d375af6969 Copying blob sha256:e603e00f9478df8f9435c78d26e7ae28f5af4bdcb4132435141c8b515c6c8e18 Copying blob sha256:c6ea85836d4d314e5a036814df370488276d0cf327b6ceb9a0af6fe6a4bf6166 Copying blob sha256:d585e1cbeb2fd3ecc65c3200c0de6af1f3124b4a5679e08477d262919e880bc7 Copying blob sha256:5ac5c0d19fbb6c61f4e7d5b08f66fc659d3530ae2f01c874e154268a7648776c Copying blob sha256:f833cf076231770cf7616d55ce466dd3e303eefff0185007978aff5a11709ac7 Copying blob sha256:52321ce7fbaf0726407f41ec0d8240bf3db1730a7d8b257e7ee66b311e168bc8 Copying config sha256:f1b04a6cdfec8d38e91f60af908c7e93d91b60b836f0075ab718c17271dd3f9c Writing manifest to image destination Storing signatures 2021/07/30 15:08:56 info unpack layer: sha256:420047682034904224708ffcbc863130ba6bce56f9d642605dccd2d375af6969 2021/07/30 15:09:00 info unpack layer: sha256:e603e00f9478df8f9435c78d26e7ae28f5af4bdcb4132435141c8b515c6c8e18 2021/07/30 15:09:00 info unpack layer: sha256:c6ea85836d4d314e5a036814df370488276d0cf327b6ceb9a0af6fe6a4bf6166 2021/07/30 15:09:01 info unpack layer: sha256:d585e1cbeb2fd3ecc65c3200c0de6af1f3124b4a5679e08477d262919e880bc7 2021/07/30 15:09:01 info unpack layer: sha256:5ac5c0d19fbb6c61f4e7d5b08f66fc659d3530ae2f01c874e154268a7648776c 2021/07/30 15:09:01 info unpack layer: sha256:f833cf076231770cf7616d55ce466dd3e303eefff0185007978aff5a11709ac7 2021/07/30 15:09:01 info unpack layer: sha256:52321ce7fbaf0726407f41ec0d8240bf3db1730a7d8b257e7ee66b311e168bc8 INFO: Creating SIF file... INFO: Build complete: rocker_r-base.sif [alice@dev2 lxc]$ ls -l rocker_r-base.sif -rwxr-xr-x. 1 alice boblab 308523008 Jul 30 15:09 rocker_r-base.sif
The above may take a minute or two to complete.
After this, we can run R within this container using:
[alice@dev2 lxc]$ singularity run rocker_r-base.sif R version 4.1.0 (2021-05-18) -- "Camp Pontanezen" Copyright (C) 2021 The R Foundation for Statistical Computing Platform: x86_64-pc-linux-gnu (64-bit) R is free software and comes with ABSOLUTELY NO WARRANTY. You are welcome to redistribute it under certain conditions. Type 'license()' or 'licence()' for distribution details. Natural language support but running in an English locale R is a collaborative project with many contributors. Type 'contributors()' for more information and 'citation()' on how to cite R or R packages in publications. Type 'demo()' for some demos, 'help()' for on-line help, or 'help.start()' for an HTML browser interface to help. Type 'q()' to quit R. > sum(1:10)  55 > q() Save workspace image? [y/n/c]: n [alice@dev2 lxc]$
Exactly what is “run” is defined by the so called “runscript” of the Singularity container, or the “CMD” if imported from a Docker container. An alternative way to launch R within this container is by explicitly executing R, e.g.
[alice@dev2 lxc]$ singularity exec rocker_r-base.sif R --quiet > sum(1:10)  55 > q("no") [alice@dev2 lxc]$
Note that, the Singularity image is marked as an executable, which means you can run it as any other executable, e.g.
[alice@dev2 lxc]$ ./rocker_r-base.sif R version 4.0.5 (2021-03-31) -- "Shake and Throw" Copyright (C) 2021 The R Foundation for Statistical Computing Platform: x86_64-pc-linux-gnu (64-bit) R is free software and comes with ABSOLUTELY NO WARRANTY. You are welcome to redistribute it under certain conditions. Type 'license()' or 'licence()' for distribution details. Natural language support but running in an English locale R is a collaborative project with many contributors. Type 'contributors()' for more information and 'citation()' on how to cite R or R packages in publications. Type 'demo()' for some demos, 'help()' for on-line help, or 'help.start()' for an HTML browser interface to help. Type 'q()' to quit R. > sum(1:10)  55 > q("no") [alice@dev2 lxc]$
To launch a shell within this container, and to also convince yourselves that the container runs Ubuntu (and not CentOS as on the Wynton HPC host system), do:
[alice@dev2 lxc]$ singularity shell rocker_r-base.sif Singularity> head -3 /etc/os-release PRETTY_NAME="Debian GNU/Linux buster/sid" NAME="Debian GNU/Linux" ID=debian Singularity> Rscript --version R scripting front-end version 4.1.0 (2021-05-18) Singularity> exit [alice@dev2 lxc]$ head -3 /etc/os-release NAME="CentOS Linux" VERSION="7 (Core)" ID="centos"
When running a container, only a few of the folders available “outside” are available “inside” the container. By default, you have access to the current working directory (=
$PWD) and your home folder (=
$HOME). In contrast, without further specifications, you will not have access to standard folders such as local
/scratch and global
/c4/scratch. Similarly, lab folders such as
/wynton/group/boblab are not available from inside the container.
[alice@dev2 lxc]$ singularity shell rocker_r-base.sif Singularity> ls /scratch ls: cannot access '/scratch': No such file or directory Singularity> ls /c4/scratch ls: cannot access '/c4/scratch': No such file or directory Singularity> ls /wynton/group/boblab ls: cannot access '/wynton/group/boblab': No such file or directory Singularity> echo $TMPDIR /scratch/alice Singularity> ls "$TMPDIR" ls: cannot access '/scratch/alice': No such file or directory
To make also these folders available within the container, we can use
--bind. In its simplest form, we can just list the folders we want to make available, e.g.
[alice@dev2 lxc]$ singularity shell --bind /scratch,/c4/scratch,/wynton/group/boblab rocker_r-base.sif Singularity> ls /scratch alice Singularity> ls /c4/scratch alice Singularity> ls /wynton/group/boblab data1 data2
singularity help instance start for more details and other ways to mount and rename folders within the container.
When it comes to the scheduler, there is nothing special about Singularity per se - the Singularity software can be used as any other software on the cluster. As a proof of concept, here is how to calculate the sum of one to ten using R within the above Linux container at the command line:
[alice@dev2 lxc]$ singularity exec rocker_r-base.sif Rscript -e "sum(1:10)"  55 [alice@dev2 lxc]$
To run this as a batch job, we need to create a job script.
[alice@dev2 lxc]$ cat demo-singularity.sh #!/usr/bin/bash #$ -N demo-singularity #$ -l mem_free=100M #$ -l h_rt=00:05:00 #$ -j y ./rocker_r-base.sif Rscript -e "sum(1:10)"
And now submit with
[alice@dev2 lxc]$ qsub demo-singularity.sh Your job 1657 ("hello_world") has been submitted
[alice@dev2 lxc]$ cat demo-singularity.o5987  55
singularity build command requires sudo privileges to build from a local definition file, however one option is to use the Sylabs.io Remote Builder which does not require sudo.
To use the remote build option one first needs a Sylabs.io access token, which can be generated after logging in to Sylabs.io.
Then add the
--remote option to the
singularity build command:
singularity build –remote <image> <def file> where <image> is the desired name of the Singularity image resulting from the build and <def file> is your Singularity definition file.
The official Singularity documentation has more information on definition files.
This is an example of building a Singularity container using the Remote Builder to run the IsoSeq3 application:
Bootstrap: docker From: continuumio/miniconda3 %post /opt/conda/bin/conda config --add channels bioconda /opt/conda/bin/conda install isoseq3 %runscript /opt/conda/bin/isoseq3
singularity build --remotecommand
[alice@dev2]$ singularity build --remote isoseq3 isoseq3.def INFO: Remote "default" added. INFO: Access Token Verified! INFO: Token stored in /root/.singularity/remote.yaml INFO: Remote "default" now in use. INFO: Starting build... Getting image source signatures ... INFO: Adding runscript INFO: Creating SIF file... INFO: Build complete: /tmp/image-4018701584 WARNING: Skipping container verification 148.5MiB / 148.5MiB [======================================] 100 % 68.0 MiB/s 0s Library storage: using 297.00 MiB out of 11.00 GiB quota (2.6% used) Container URL: https://cloud.sylabs.io/library/... INFO: Build complete: isoseq3
isoseq3should now exist along with the definition file
[alice@dev2]$ ls -1 isoseq3 isoseq3.def
%runscriptin the definition file specifies to run the isoseq3 command, you can use it like this
[alice@dev2]$ isoseq3 --version isoseq3 3.4.0 (commit v3.4.0)
You can install Singularity on a Linux host where you have root access an and build the container image there using
sudo singularity build
Then the finished container image can be transferred to Wynton for use.
If you do not have a Linux workstation, a Linux virtual machine is easy to install on Windows and macOS.
Similarity to installing Singularity on a Linux workstation, you can install Docker on a workstation you have access to and upload the Docker image to a registry such as Docker Hub or quay.io. Docker images can be converted to a Singularity container.
Q. Why not Docker?
A. Docker is one of the most popular and well-known software solutions for using Linux Containers. However, contrary to Singularity, it turns out that it is hard to get Docker to play well with multi-tenant HPC environments.
Q. What’s the filename extension
First of all, the filename extension is optional, and some prefer to drop them, e.g.
rocker_r-base instead of
rocker_r-base.sif. SIF, which is short for the Singularity Container Image Format, is a file format that can hold a Linux container environments in a single file. You might also see Singularity images named
*.simg, which are legacy file formats that Singularity used in the past, where
*.img indicates a writable (ext3) images whereas
*.simg indicates a read-only (squashfs) image.](https://groups.google.com/a/lbl.gov/d/msg/singularity/Cq7kIbN_L68/2mOdkwx2BAAJ)