Singularity on HPC¶
One of the architecturally defined features in Singularity is that it can execute containers like they are native programs or scripts on a host computer. As a result, integration with schedulers is simple and runs exactly as you would expect. All standard input, output, error, pipes, IPC, and other communication pathways that locally running programs employ are synchronized with the applications running locally within the container. Additionally, because Singularity is not emulating a full hardware level virtualization paradigm, there is no need to separate out any sandboxed networks or file systems because there is no concept of user-escalation within a container. Users can run Singularity containers just as they run any other program on the HPC resource.
We are in the process of developing Singularity Hub, which will allow for generation of workflows using Singularity containers in an online interface, and easy deployment on standard research clusters (e.g., SLURM, SGE). Currently, the Singularity core software is installed on the following research clusters, meaning you can run Singularity containers as part of your jobs:
- The Sherlock cluster at Stanford University
- SDSC Comet and Gordon (XSEDE)
- MASSIVE M1 M2 and M3 (Monash University and Australian National Merit Allocation Scheme)
Integration with MPI¶
Another result of the Singularity architecture is the ability to properly integrate with the Message Passing Interface (MPI). Work has already been done for out of the box compatibility with Open MPI (both in Open MPI v2.1.x as well as part of Singularity). The Open MPI/Singularity workflow works as follows:
- mpirun is called by the resource manager or the user directly from a shell
- Open MPI then calls the process management daemon (ORTED)
- The ORTED process launches the Singularity container requested by the mpirun command
- Singularity builds the container and namespace environment
- Singularity then launches the MPI application within the container
- The MPI application launches and loads the Open MPI libraries
- The Open MPI libraries connect back to the ORTED process via the Process Management Interface (PMI)
- At this point the processes within the container run as they would normally directly on the host.
This entire process happens behind the scenes, and from the user’s perspective running via MPI is as simple as just calling mpirun on the host as they would normally. Below are example snippets of building and installing OpenMPI into a container and then running an example MPI program through Singularity.
MPI Development Example¶
What are supported Open MPI Version(s)? To achieve proper container’ized Open MPI support, you should use Open MPI version 2.1. There are however three caveats:
- Open MPI 1.10.x may work but we expect you will need exactly matching version of PMI and Open MPI on both host and container (the 2.1 series should relax this requirement)
- Open MPI 2.1.0 has a bug affecting compilation of libraries for some interfaces (particularly Mellanox interfaces using libmxm are known to fail). If your in this situation you should use the master branch of Open MPI rather than the release.
- Using Open MPI 2.1 does not magically allow your container to connect to networking fabric libraries in the host. If your cluster has, for example, an infiniband network you still need to install OFED libraries into the container. Alternatively you could bind mount both Open MPI and networking libraries into the container, but this could run afoul of glib compatibility issues (its generally OK if the container glibc is more recent than the host, but not the other way around)
Code Example using Open MPI 2.1.0 Stable¶
$ # Include the appropriate development tools into the container (notice we are calling $ # singularity as root and the container is writable) $ sudo singularity exec -w /tmp/Centos-7.img yum groupinstall "Development Tools" $ $ # Obtain the development version of Open MPI $ wget https://www.open-mpi.org/software/ompi/v2.1/downloads/openmpi-2.1.0.tar.bz2 $ tar jtf openmpi-2.1.0.tar.bz2 $ cd openmpi-2.1.0 $ $ singularity exec /tmp/Centos-7.img ./configure --prefix=/usr/local $ singularity exec /tmp/Centos-7.img make $ $ # Install OpenMPI into the container (notice now running as root and container is writable) $ sudo singularity exec -w -B /home /tmp/Centos-7.img make install $ $ # Build the OpenMPI ring example and place the binary in this directory $ singularity exec /tmp/Centos-7.img mpicc examples/ring_c.c -o ring $ $ # Install the MPI binary into the container at /usr/bin/ring $ sudo singularity copy /tmp/Centos-7.img ./ring /usr/bin/ $ $ # Run the MPI program within the container by calling the MPIRUN on the host $ mpirun -np 20 singularity exec /tmp/Centos-7.img /usr/bin/ring
Code Example using Open MPI git master¶
The previous example (using the Open MPI 2.1.0 stable release) should work fine on most hardware but if you have an issue, try running the example below (using the Open MPI Master branch):
$ # Include the appropriate development tools into the container (notice we are calling $ # singularity as root and the container is writable) $ sudo singularity exec -w /tmp/Centos-7.img yum groupinstall "Development Tools" $ $ # Clone the OpenMPI GitHub master branch in current directory (on host) $ git clone https://github.com/open-mpi/ompi.git $ cd ompi $ $ # Build OpenMPI in the working directory, using the tool chain within the container $ singularity exec /tmp/Centos-7.img ./autogen.pl $ singularity exec /tmp/Centos-7.img ./configure --prefix=/usr/local $ singularity exec /tmp/Centos-7.img make $ $ # Install OpenMPI into the container (notice now running as root and container is writable) $ sudo singularity exec -w -B /home /tmp/Centos-7.img make install $ $ # Build the OpenMPI ring example and place the binary in this directory $ singularity exec /tmp/Centos-7.img mpicc examples/ring_c.c -o ring $ $ # Install the MPI binary into the container at /usr/bin/ring $ sudo singularity copy /tmp/Centos-7.img ./ring /usr/bin/ $ $ # Run the MPI program within the container by calling the MPIRUN on the host $ mpirun -np 20 singularity exec /tmp/Centos-7.img /usr/bin/ring Process 0 sending 10 to 1, tag 201 (20 processes in ring) Process 0 sent to 1 Process 0 decremented value: 9 Process 0 decremented value: 8 Process 0 decremented value: 7 Process 0 decremented value: 6 Process 0 decremented value: 5 Process 0 decremented value: 4 Process 0 decremented value: 3 Process 0 decremented value: 2 Process 0 decremented value: 1 Process 0 decremented value: 0 Process 0 exiting Process 1 exiting Process 2 exiting Process 3 exiting Process 4 exiting Process 5 exiting Process 6 exiting Process 7 exiting Process 8 exiting Process 9 exiting Process 10 exiting Process 11 exiting Process 12 exiting Process 13 exiting Process 14 exiting Process 15 exiting Process 16 exiting Process 17 exiting Process 18 exiting Process 19 exiting
By default Singularity tries to create a seamless user experience between the host and the container. To do this, Singularity makes various locations accessible within the container automatically. For example, the user’s home directory is always bound into the container as is /tmp and /var/tmp. Additionally your current working directory (cwd/pwd) is also bound into the container iff it is not an operating system directory or already accessible via another mount. For almost all cases, this will work flawlessly as follows:
$ pwd /home/gmk/demo $ singularity shell container.img Singularity/container.img> pwd /home/gmk/demo Singularity/container.img> ls -l debian.def -rw-rw-r--. 1 gmk gmk 125 May 28 10:35 debian.def Singularity/container.img> exit $
For directory binds to function properly, there must be an existing
target endpoint within the container (just like a mount point). This
means that if your home directory exists in a non-standard base
directory like “/foobar/username” then the base directory “/foobar”
must already exist within the container.
Singularity will not create these base directories! You must enter the
container with the option
--writable being set, and create the directory
Current Working Directory¶
Singularity will try to replicate your current working directory within the container. Sometimes this is straight forward and possible, other times it is not (e.g. if the base dir of your current working directory does not exist). In that case, Singularity will retain the file descriptor to your current directory and change you back to it. If you do a ‘pwd’ within the container, you may see some weird things. For example:
$ pwd /foobar $ ls -l total 0 -rw-r--r--. 1 root root 0 Jun 1 11:32 mooooo $ singularity shell ~/demo/container.img WARNING: CWD bind directory not present: /foobar Singularity/container.img> pwd (unreachable)/foobar Singularity/container.img> ls -l total 0 -rw-r--r--. 1 root root 0 Jun 1 18:32 mooooo Singularity/container.img> exit $
But notice how even though the directory location is not resolvable, the directory contents are available.
Standard IO and pipes¶
Singularity automatically sends and receives all standard IO from the host to the applications within the container to facilitate expected behavior from the interaction between the host and the container. For example:
$ cat debian.def | singularity exec container.img grep 'MirrorURL' MirrorURL "http://ftp.us.debian.org/debian/" $ Making changes to the container (writable) By default, containers are accessed as read only. This is both to enable parallel container execution (e.g. MPI). To enter a container using exec, run, or shell you must pass the --writable flag in order to open the image as read/writable.
Containing the container¶
By providing the argument
shell you will find that shared directories
are no longer shared. For example, the user’s home directory is
writable, but it is non-persistent between non-overlapping runs.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: (1) Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. (2) Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. (3) Neither the name of the University of California, Lawrence Berkeley National Laboratory, U.S. Dept. of Energy nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission. THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. You are under no obligation whatsoever to provide any bug fixes, patches, or upgrades to the features, functionality or performance of the source code ("Enhancements") to anyone; however, if you choose to make your Enhancements available either publicly, or directly to Lawrence Berkeley National Laboratory, without imposing a separate written license agreement for such Enhancements, then you hereby grant the following license: a non-exclusive, royalty-free perpetual license to install, use, modify, prepare derivative works, incorporate into other computer software, distribute, and sublicense such enhancements or derivative works thereof, in binary and source code form. If you have questions about your rights to use or distribute this software, please contact Berkeley Lab's Innovation & Partnerships Office at IPO@lbl.gov. NOTICE. This Software was developed under funding from the U.S. Department of Energy and the U.S. Government consequently retains certain rights. As such, the U.S. Government has been granted for itself and others acting on its behalf a paid-up, nonexclusive, irrevocable, worldwide license in the Software to reproduce, distribute copies to the public, prepare derivative works, and perform publicly and display publicly, and to permit other to do so.
In layman terms…¶
In addition to the (already widely used and very free open source) standard BSD 3 clause license, there is also wording specific to contributors which ensures that we have permission to release, distribute and include a particular contribution, enhancement, or fix as part of Singularity proper. For example any contributions submitted will have the standard BSD 3 clause terms (unless specifically and otherwise stated) and that the contribution is comprised of original new code that the contributor has authority to contribute.