Deploy a container using Singularity

From ScientificComputing
Revision as of 08:09, 14 June 2021 by Jarunanp (talk | contribs)

Jump to: navigation, search

< Examples

In this tutorial, we show an example for how to develop a Docker image on you local computer and present three methods for how to deploy a container on the cluster by using Singularity:

Develop a docker image on your local computer

[user@local]$ mkdir test_container
[user@local]$ cd test_container
  • Open a file named Dockerfile with a text editor and add the following lines into the Dockerfile to create a container with Ubuntu as an OS and install Python3 with Numpy, Matplotlib and TensorFlow packages
FROM ubuntu:latest

WORKDIR /home/project

RUN apt-get -y update
RUN apt-get install -y lsb-release python3 python3-pip \
&& pip install numpy==1.19 matplotlib tensorflow
  • Build an image
[user@local]$ docker build -t dockeruser/ubuntu-python3:latest .
Sending build context to Docker daemon  3.072kB
Step 1/4 : FROM ubuntu:latest
 ---> 7e0aa2d69a15
Step 2/4 : WORKDIR /home/project
 ---> Using cache
 ---> d9947182f6a8
Step 3/4 : RUN apt-get -y update
 ---> Using cache
 ---> 89a634ac491b
Step 4/4 : RUN apt-get install -y lsb-release python3 python3-pip && pip install numpy==1.19 matplotlib tensorflow
 ---> Using cache
 ---> d639b6f22ee5
Successfully built d639b6f22ee5
Successfully tagged dockeruser/ubuntu-python3:latest
[user@local]$ docker images
REPOSITORY                   TAG       IMAGE ID       CREATED          SIZE
dockeruser/ubuntu-python3     latest    d639b6f22ee5   18 minutes ago   2.19GB

Method 1: Use the Docker image on Dockerhub

On your local computer

  • Log into your Docker account (replace dockeruser with your Docker username)
[user@local]$  docker login
username: dockeruser
password:
  • Push the image to Dockerhub
[user@local]$  docker push dockeruser/ubuntu-python3:latest
The push refers to repository [docker.io/dockeruser/ubuntu-python3]
4648e930d81f: Pushed 
2e38cc729d73: Pushed 
f281ab5d2fac: Pushed 
2f140462f3bc: Mounted from library/ubuntu 
63c99163f472: Mounted from library/ubuntu 
ccdbb80308cc: Mounted from library/ubuntu 
latest: digest: sha256:cd8a34b30aabe432232787d1e93844cd01027c2b235fb88f106297ed26c1f2ca size: 1576
  • Log into the Euler cluster (replace nethzuser with your NETHZ username)
[user@local]$ ssh nethzuser@euler.ethz.ch

On the cluster

  • Request a compute node with Singularity
[nethzuser@eu-login-01 ~]$ bsub -n 1 -R singularity -R light -Is bash
Generic job.
Job <174809810> is submitted to queue <light.5d>.
<<Waiting for dispatch ...>>
<<Starting on eu-ms-001-02>>
[nethzuser@eu-ms-001-02 ~]$
  • Load module eth_proxy to connect to the internet from a compute node
[nethzuser@eu-ms-001-02 ~]$ module load eth_proxy
  • Pull the Docker image with Singularity
[nethzuser@eu-ms-001-02 ~]$ cd $SCRATCH
[nethzuser@eu-ms-001-02 nethzuser]$ singularity pull docker://dockeruser/ubuntu-python3
INFO:    Converting OCI blobs to SIF format
INFO:    Starting build...
Getting image source signatures
Copying blob 345e3491a907 done  
...
INFO:    Creating SIF file...
[nethzuser@eu-ms-001-02 nethzuser]$ ls
ubuntu-python3_latest.sif
  • Run the container as shell
[nethzuser@eu-ms-001-02 nethzuser]$ singularity shell ubuntu-python3_latest.sif
Singularity> 
Singularity> lsb_release -a
No LSB modules are available.
Distributor ID:	Ubuntu
Description:	Ubuntu 20.04.2 LTS
Release:	20.04
Codename:	focal
Singularity> python3
Python 3.8.5 (default, May 27 2021, 13:30:53) 
[GCC 9.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
2021-06-09 14:08:24.919069: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library  'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /.singularity.d/libs
2021-06-09 14:08:24.919094: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not  have a GPU set up on your machine.
>>> tf.__version__
'2.5.0'
>>> 
Singularity> exit
exit

As you can see, the container contains GCC 9.3.0 with Python 3.8.5 and TensorFlow 2.5.0. At the time of writing tutorial, the latest version of GCC is 8.2.0 and TensorFlow 2.4.0

[nethzuser@eu-ms-001-02 nethzuser]$ module avail gcc
--------------------------- /cluster/apps/lmodules/Core --------------------------------
  gcc/4.8.2    gcc/4.8.5 (L)    gcc/5.4.0    gcc/6.3.0    gcc/7.3.0    gcc/8.2.0 (D)
[nethzuser@eu-ms-001-02 nethzuser]$ module load gcc/6.3.0 python/3.8.5
[nethzuser@eu-ms-001-02 nethzuser]$ python
Python 3.8.5 (default, Oct  6 2020, 10:04:29) 
[GCC 6.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
2021-06-09 14:56:47.778183: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library  'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /cluster/apps/gcc-6.3.0/openblas-0.2.20-cot3cawsqf4pkxjwzjexaykbwn2ch3ii/lib:/cluster/apps/nss/gcc-6.3.0/python/3.8.5/x86_64/lib64:/cluster/spack/apps/linux-centos7-x86_64/gcc-4.8.5/gcc-6.3.0-sqhtfh32p5gerbkvi5hih7cfvcpmewvj/lib64:/cluster/spack/apps/linux-centos7-x86_64/gcc-4.8.5/gcc-6.3.0-sqhtfh32p5gerbkvi5hih7cfvcpmewvj/lib:/cluster/apps/lsf/10.1/linux2.6-glibc2.3-x86_64/lib
2021-06-09 14:56:47.778214: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not  have a GPU set up on your machine.
>>> tf.__version__
'2.4.0'
>>>

Method 2: Convert a Docker image to a Singularity image

On your local computer

  • Create a tarball of the Docker image
[user@local]$ docker save d639b6f22ee5 -o ubuntu-python3.tar
[user@local]$ ls
ubuntu-python3.tar
  • Copy the tarball to $SCRATCH on the cluster (replace netzuser with your NETHZ username)
[user@local]$ scp ubuntu-python3.tar netzuser@euler.ethz.ch:/cluster/scratch/nethzuser
  • Log in to the Euler cluster (replace netzuser with your NETHZ username)
[user@local]$ ssh netzuser@euler.ethz.ch

On the cluster

  • Request a compute node with Singularity
[nethzuser@eu-login-01 ~]$ bsub -n 1 -R singularity -R light -Is bash
Generic job.
Job <174809810> is submitted to queue <light.5d>.
<<Waiting for dispatch ...>>
<<Starting on eu-ms-001-02>>
[nethzuser@eu-ms-001-02 ~]$
  • Go to $SCRATCH
[nethzuser@eu-ms-001-02 ~]$ cd $SCRATCH
[nethzuser@eu-ms-001-02 nethzuser]$ pwd
/cluster/scratch/nethzuser
  • Check if the tarball is in $SCRATCH
[nethzuser@eu-ms-001-02 nethzuser]$ ls ubuntu-python3.tar
ubuntu-python3.tar
  • Convert the tarball to a Singularity image. This shall create a sandbox directory called ubuntu-python3 in your working directory.
[nethzuser@eu-ms-001-02 nethzuser]$ singularity build --sandbox ubuntu-python3 docker-archive://ubuntu-python3.tar
INFO:    Starting build...
...
INFO:    Creating sandbox directory...
INFO:    Build complete: ubuntu-python3
  • Run the Singularity box
[nethzuser@eu-ms-001-02 nethzuser]$ singularity shell ubuntu-python3
Singularity> 
Singularity> gcc --version
gcc (Ubuntu 9.3.0-17ubuntu1~20.04) 9.3.0
Copyright (C) 2019 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Singularity> python3 --version
Python 3.8.5
ingularity> python3 -c "import tensorflow as tf; print(tf.__version__)"
2021-06-10 09:12:44.533833: W tensorflow/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libcudart.so.11.0'; dlerror: libcudart.so.11.0: cannot open shared object file: No such file or directory; LD_LIBRARY_PATH: /.singularity.d/libs
2021-06-10 09:12:44.533867: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine.
2.5.0
Singularity> exit
exit
[nethzuser@eu-ms-001-02 nethzuser]$

Method 3: Build a container image from Singularity definition file

Similar to Dockerfile for creating a Docker container, Singularity can build a container from a Singularity definition file as well.

Currently, Singularity on the cluster is not setup to build a container from a definition file.

< Examples