Environment variables are values you can set in a session, which can be used to influence the behavior of programs. It's often considered best practice to use environment variables to pass settings to a program in a container, because they are easily set and don't rely on writing and binding in program-specific configuration files. When building a container you may need to set fixed or default environment variables. When running containers you may need to set or override environment variables.
The metadata <sec:metadata>
of a container is information that describes the container. {Project} automatically records important information such as the definition file used to build a container. Other details such as the version of {Project} used are present as labels <sec:labels>
on a container. You can also specify your own to be recorded against your container.
When you run a program in a container with {Project}, the environment variables that the program sees are a combination of:
- The environment variables set in the base image (e.g. Docker image) used to build the container.
- The environment variables set in the
%environment
section of the definition file used to build the container.- Most of the environment variables set on your host, which are passed into the container.
- Any variables you set specifically for the container at runtime, using the
--env
,--env-file
options, or by setting{ENVPREFIX}ENV_
variables outside of the container.- The
PATH
variable can be manipulated to add entries.- Runtime variables
{ENVPREFIX}_xxx
set by {Project} to provide information about the container.
The environment variables from the base image or definition file used to build a container always apply, but can be overridden.
You can choose to exclude passing environment variables from the host into the container with the -e
or --cleanenv
option.
We'll go through each place environment variables can be defined, so that you can understand how the final environment in a container is created, and can be manipulated.
If you are interested in variables available when you are building a container, rather than when running a container, see build
environment section <build-environment>
.
When you build a container with {Project} you might bootstrap from a library or Docker image, or using Linux distribution bootstrap tools such as debootstrap
, yum
etc.
When using debootstrap
, yum
etc. you are starting from a fresh install of a Linux distribution into your container. No specific environment variables will be set. If you are using a library
or Docker
source then you may inherit environment variables from your base image.
If I build {aProject} container from the image docker://python:3.7
then when I run the container I can see that the PYTHON_VERSION
variable is set in the container:
$ {command} exec python.sif env | grep PYTHON_VERSION
PYTHON_VERSION=3.7.7
This happens because the Dockerfile
used to build that container has ENV PYTHON_VERSION 3.7.7
set inside it.
You can override the inherited environment with {ENVPREFIX}ENV_
vars, or the --env / --env-file
flags (see below), but Dockerfile
ENV
vars will not be overridden by host environment variables of the same name.
Environment variables can be included in your container by adding them to your definition file. Use export
in the %environment
section of a definition file to set a container environment variable:
Bootstrap: docker
From: alpine
%environment
export MYVAR="Hello"
%runscript
echo $MYVAR
Now the value of MYVAR
is Hello
when the container is launched. The %runscript
is set to echo the value.
$ {command} run env.sif
Hello
Warning
{Project} uses an embedded shell interpreter to evaluate and setup container environments, therefore all commands executed from the %environment
section have an execution timeout of 1 minute. While it is possible to source a script from there, it is not recommended to use this section to run potentially long initialization tasks because this would impact users running the image and the execution could abort due to timeout.
In some circumstances the value that needs to be assigned to an environment variable may only be known after e.g. software installation, in %post
. For situations like this, the ${ENVPREFIX}_ENVIRONMENT
variable is provided. Redirecting text to this variable will cause it to be written to a file called /.singularity.d/env/91-environment.sh
that will be sourced at runtime.
Variables set in the %post
section through ${ENVPREFIX}_ENVIRONMENT
take precedence over those added via %environment
.
If you have environment variables set outside of your container, on the host, then by default they will be available inside the container. Except that:
- An environment variable set on the host will be overridden by a variable of the same name that has been set either inside the container image, or via
{ENVPREFIX}ENV_
environment variables, or the--env
and--env-file
flags.- The
PS1
shell prompt is reset for a container specific prompt.- The
PATH
environment variable will be modified to contain default values.- The
LD_LIBRARY_PATH
is modified to a default/.singularity.d/libs
, that will include NVIDIA / ROCm libraries if applicable.
Note
See compatibility documentation for SINGULARITYENV_
prefixed environment variable support here <singularity_environment_variable_compatibility>
.
To override an environment variable that is already set in the container with the value from the host, use {ENVPREFIX}ENV_
or the --env
flag. For example, to force MYVAR
in the container to take the value of MYVAR
on the host:
$ export {ENVPREFIX}ENV_MYVAR="$MYVAR"
$ {command} run mycontainer.sif
# or
$ {command} run --env "MYVAR=$MYVAR"
If you do not want the host environment variables to pass into the container you can use the -e
or --cleanenv
option. This gives a clean environment inside the container, with a minimal set of environment variables for correct operation of most software.
$ {command} exec --cleanenv env.sif env
HOME=/home/dave
LANG=C
LD_LIBRARY_PATH=/.singularity.d/libs
PATH=/startpath:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
PROMPT_COMMAND=PS1="{Project}> "; unset PROMPT_COMMAND
PS1={Project}>
PWD=/home/dave/doc-tesrts
{ENVPREFIX}_COMMAND=exec
{ENVPREFIX}_CONTAINER=/home/dave/doc-tesrts/env.sif
{ENVPREFIX}_ENVIRONMENT=/.singularity.d/env/91-environment.sh
{ENVPREFIX}_NAME=env.sif
TERM=xterm-256color
Warning
If you work on a host system that sets a lot of environment variables, e.g. because you use software made available through environment modules / lmod, you may see strange behavior in your container. Check your host environment with env
for variables such as PYTHONPATH
that can change the way code runs, and consider using --cleanenv
.
It can be useful for a program to know when it is running in a {Project} container, and some basic information about the container environment. {Project} will automatically set a number of environment variables in a container that can be inspected by any program running in the container.
{ENVPREFIX}_COMMAND
- how the container was started, e.g.exec
/run
/shell
.{ENVPREFIX}_CONTAINER
- the full path to the container image.{ENVPREFIX}_ENVIRONMENT
- path inside the container to the shell script holding the container image environment settings.{ENVPREFIX}_NAME
- name of the container image, e.g.myfile.sif
ordocker://ubuntu
.{ENVPREFIX}_BIND
- a list of bind paths that the user requested, via flags or environment variables, when running the container.
Note
See compatibility documentation for SINGULARITY_
prefixed environment variable support here <singularity_environment_variable_compatibility>
.
You can override variables that have been set in the container image, or define additional variables, in various ways as appropriate for your workflow.
The --env
option on the run/exec/shell
commands allows you to specify environment variables as NAME=VALUE
pairs:
$ {command} run env.sif
Hello
$ {command} run --env MYVAR=Goodbye env.sif
Goodbye
Separate multiple variables with commas, e.g. --env MYVAR=A,MYVAR2=B
, and use shell quoting / shell escape if your variables include special characters.
The --env-file
option lets you provide a file that contains environment variables as NAME=VALUE
pairs, e.g.:
$ cat myenvs
MYVAR="Hello from a file"
$ {command} run --env-file myenvs env.sif
Hello from a file
If you export an environment variable on your host called {ENVPREFIX}ENV_xxx
before you run a container, then it will set the environment variable xxx
inside the container:
$ {command} run env.sif
Hello
$ export {ENVPREFIX}ENV_MYVAR="Overridden"
$ {command} run env.sif
Overridden
PATH
is a special environment variable that tells a system where to look for programs that can be run. PATH
contains multiple filesystem
locations (paths) separated by colons. When you ask to run a program myprog
, the system looks through these locations one by one, until it finds myprog
.
To ensure containers work correctly, when a host PATH
might contain a lot of host-specific locations that are not present in the container, {Project} will ensure PATH
in the container is set to a default.
/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
This covers the standard locations for software installed using a system package manager in most Linux distributions. If you have software installed elsewhere in the container, then you can override this by setting PATH
in the container definition %environment
block.
If your container depends on things that are bind mounted into it, or you have another need to modify the PATH
variable when starting a container, you can do so with {ENVPREFIX}ENV_APPEND_PATH
or {ENVPREFIX}ENV_PREPEND_PATH
.
If you set a variable on your host called {ENVPREFIX}ENV_APPEND_PATH
then its value will be appended (added to the end) of the PATH
variable in the container.
$ {command} exec env.sif sh -c 'echo $PATH'
/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
$ export {ENVPREFIX}ENV_APPEND_PATH="/endpath"
$ {command} exec env.sif sh -c 'echo $PATH'
/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/endpath
Alternatively you could use the --env
option to set a APPEND_PATH
variable, e.g. --env APPEND_PATH=/endpath
.
If you set a variable on your host called {ENVPREFIX}ENV_PREPEND_PATH
then its value will be prepended (added to the start) of the PATH
variable in the container.
$ {command} exec env.sif sh -c 'echo $PATH'
/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
$ export {ENVPREFIX}ENV_PREPEND_PATH="/startpath"
$ {command} exec env.sif sh -c 'echo $PATH'
/startpath:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
Alternatively you could use the --env
option to set a PREPEND_PATH
variable, e.g. --env PREPEND_PATH=/startpath
.
{Project} uses an embedded shell interpreter to process the container startup scripts and environment. When this processing is performed, by default a single step of shell evaluation happens in the container context. The shell from which you are running {Project} may also evaluate variables on your command line before passing them to {Project}.
This default behavior of {Project} differs from Docker/OCI handling of environment variables / ENV
directives. To avoid the extra evaluation of variables that {Project} performs you can:
- Follow the instructions about escaping in the sections below, to add additional escape characters and/or quoting.
- Use the
--no-eval
or--compat
flags.
--no-eval
prevents {Project} from evaluating environment variables on container startup, so that they will take the same value as with a Docker/OCI runtime:
# Set an environment variable that would run `date` if evaluated
$ export {ENVPREFIX}_MYVAR='$(date)'
# Default behavior
# MYVAR was evaluated in the container, and is set to the output of `date`
$ {command} run ~/ubuntu_latest.sif env | grep MYVAR
MYVAR=Tue Apr 26 14:37:07 CDT 2022
# --no-eval / --compat behavior
# MYVAR was not evaluated and is a literal `$(date)`
$ {command} run --no-eval ~/ubuntu_latest.sif env | grep MYVAR
MYVAR=$(date)
The --compat
flag is a short-hand flag to activate --no-eval
along with other Docker/OCI compatibility flags. See compat-flag
for more details.
To set a container environment variable to the value of a variable on the host, use double quotes around the variable, so that it is processed by the host shell before the value is passed to {Project}. For example:
{command} run --env "MYHOST=$HOSTNAME" mycontainer.sif
This will set the MYHOST
environment variable inside the container to the value of the HOSTNAME
on the host system. $HOSTNAME
is substituted before the host shell runs {command}
.
Note
You can often use no quotes, but it is good practice to use quotes consistently so that variables containing e.g. spaces are handled correctly.
To set an environment variable to a value that references another variable inside the container, you should escape the $
sign to \$
. This prevents the host shell from substituting the value. Instead it will be substituted inside the container.
For example, to create an environment variable MYPATH
, with the same value as PATH
in the container (not the host's PATH
):
{command} run --env "MYPATH=\$PATH" mycontainer.sif
You can also use this approach to append or prepend to variables that are already set in the container. For example, --env PATH="\$PATH:/endpath"
would have the same effect as --env APPEND_PATH="/endpath"
, which uses the special APPEND/PREPEND
handling for PATH
discussed above.
If you need to pass an environment variable into the container verbatim, it must be quoted and escaped appropriately. For example, if you need to set a path containing a literal $LIB
for the LD_PRELOAD
environment variable:
{command} run --env="LD_PRELOAD=/foo/bar/\\\$LIB/baz.so" mycontainer.sif
This will result in LD_PRELOAD
having the value /foo/bar/$LIB/baz.so
inside the container.
The host shell consumes the double \\
, and then environment processing within {Project} will consume the third \
that escapes the literal $
.
You can also use single quotes on the command line, to avoid one level of escaping:
{command} run --env='LD_PRELOAD=/foo/bar/\$LIB/baz.so' mycontainer.sif
When a container is run with {Project}, the container environment is constructed in the following order:
- Clear the environment, keeping just
HOME
and{ENVPREFIX}_APPNAME
.- Set Docker/OCI defined environment variables, where a Docker or OCI image was used as the base for the container build.
- If
PATH
is not defined set the {Project} defaultPATH
or- If
PATH
is defined, add any missing path parts from {Project} defaults- Set environment variables defined explicitly in the
%environment
section of the definition file. These can override any previously set values.- Set environment variables that were defined in the
%post
section of the build, by addition to the${ENVPREFIX}_ENVIRONMENT
file.- Set SCIF (
--app
) environment variables- Set base environment essential vars (
PS1
andLD_LIBRARY_PATH
)- Inject
{ENVPREFIX}ENV_
/--env
/--env-file
variables so they can override or modify any previous values.- Apply special
APPEND_PATH
/PREPEND_PATH
handling.- Restore environment variables from the host, if they have not already been set in the container, and the
--cleanenv
/--containall
options were not specified.
Warning
While {Project} will process additional scripts found under /.singularity.d/env
inside the container, it is strongly recommended to avoid manipulating the container environment by directly adding or modifying scripts in this directory. Please use the %environment
section of the definition file, and the ${ENVPREFIX}_ENVIRONMENT
file from %post
if required.
A future version of {Project} may move container scripts, environment, and metadata outside of the container's root filesystem. This will permit further reproducibility and compatibility improvements, but will preclude environment manipulation via arbitrary scripts.
The umask
value on a Linux system controls the default permissions for newly created files. It is not an environment variable, but influences the behavior of programs in the container when they create new files.
Note
A detailed description of what the umask
is, and how it works can be found at Wikipedia.
{Project} sets the umask
in the container to match the value outside, unless:
- The
--fakeroot
option is used, in which case a0022
umask is set so thatroot
owned newly created files have expected 'system default' permissions, and can be accessed by other non-root users who may use the same container later.- The
--no-umask
option is used, in which case a0022
umask is set.
Each {Project} container has metadata describing the container, how it was built, etc. This metadata includes the definition file used to build the container and labels, which are specific pieces of information set automatically or explicitly when the container is built.
{Project} container default labels are represented using the rc1 Label Schema.
When building a container from an existing image, either directly from a URI or with a definition file, your container will inherit the labels that are set in that base image. For example the LABEL
a Docker container sets in its Dockerfile
, or a SIF container that sets labels in its definition file as described below.
Inherited labels can only be overwritten during a build when the build is performed using the --force
option. {Project} will warn that it is not modifying an existing label when --force
is not used:
$ {command} build test2.sif test2.def
...
INFO: Adding labels
WARNING: Label: OWNER already exists and force option is false, not overwriting
You can add custom labels to your container using the %labels
section in a definition file:
Bootstrap: docker
From: ubuntu:latest
%labels
OWNER Joana
You may wish to set a label to a value that is not known in advance, when you are writing the definition file, but can be obtained in the %post
section of your definition file while the container is building.
{Project} allows this, through adding labels to the file defined by the {ENVPREFIX}_LABELS
environment variable in the %post
section:
Bootstrap: docker
From: ubuntu:latest
# These labels take a fixed value in the definition
%labels
OWNER Joana
# We can now also set labels to a value at build time
%post
VAL="$(myprog --version)"
echo "my.label $VAL" >> "${ENVPREFIX}_LABELS"
Labels must be added to the file one per line, in a NAME VALUE
format, where the name and value are separated by a space.
The inspect
command gives you the ability to view the labels and/or other metadata that were added to your container when it was built.
Running inspect without any options, or with the -l
or --labels
options will display any labels set on the container
$ {command} inspect ubuntu.sif
my.label: version 1.2.3
OWNER: Joana
org.label-schema.build-arch: amd64
org.label-schema.build-date: Thursday_12_November_2020_10:51:59_CST
org.label-schema.schema-version: 1.0
org.label-schema.usage.singularity.deffile.bootstrap: docker
org.label-schema.usage.singularity.deffile.from: ubuntu:latest
org.label-schema.usage.singularity.version: 3.7.0-rc.1
We can easily see when the container was built, the source of the base image, and the exact version of {Project} that was used to build it.
The custom label OWNER
that we set in our definition file is also visible.
The -d
or -deffile
flag shows the definition file(s) that were used to build the container.
$ {command} inspect --deffile jupyter.sif
And the output would look like:
Bootstrap: docker
From: debian:9
%help
Container with Anaconda 2 (Conda 4.5.11 Canary) and Jupyter Notebook 5.6.0 for Debian 9.x (Stretch).
This installation is based on Python 2.7.15
%environment
JUP_PORT=8888
JUP_IPNAME=localhost
export JUP_PORT JUP_IPNAME
%startscript
PORT=""
if [ -n "$JUP_PORT" ]; then
PORT="--port=${JUP_PORT}"
fi
IPNAME=""
if [ -n "$JUP_IPNAME" ]; then
IPNAME="--ip=${JUP_IPNAME}"
fi
exec jupyter notebook --allow-root ${PORT} ${IPNAME}
%setup
#Create the .condarc file where the environments/channels from conda are specified, these are pulled with preference to root
cd /
touch .condarc
%post
echo 'export RANDOM=123456' >>${ENVPREFIX}_ENVIRONMENT
#Installing all dependencies
apt-get update && apt-get -y upgrade
apt-get -y install \
build-essential \
wget \
bzip2 \
ca-certificates \
libglib2.0-0 \
libxext6 \
libsm6 \
libxrender1 \
git
rm -rf /var/lib/apt/lists/*
apt-get clean
#Installing Anaconda 2 and Conda 4.5.11
wget -c https://repo.continuum.io/archive/Anaconda2-5.3.0-Linux-x86_64.sh
/bin/bash Anaconda2-5.3.0-Linux-x86_64.sh -bfp /usr/local
#Conda configuration of channels from .condarc file
conda config --file /.condarc --add channels defaults
conda config --file /.condarc --add channels conda-forge
conda update conda
#List installed environments
conda list
Which is the definition file for the jupyter.sif
container.
The -r
or --runscript
option shows the runscript for the image.
$ {command} inspect --runscript jupyter.sif
And the output would look like:
#!/bin/sh
OCI_ENTRYPOINT=""
OCI_CMD="bash"
# ENTRYPOINT only - run entrypoint plus args
if [ -z "$OCI_CMD" ] && [ -n "$OCI_ENTRYPOINT" ]; then
SINGULARITY_OCI_RUN="${OCI_ENTRYPOINT} $@"
fi
# CMD only - run CMD or override with args
if [ -n "$OCI_CMD" ] && [ -z "$OCI_ENTRYPOINT" ]; then
if [ $# -gt 0 ]; then
SINGULARITY_OCI_RUN="$@"
else
SINGULARITY_OCI_RUN="${OCI_CMD}"
fi
fi
# ENTRYPOINT and CMD - run ENTRYPOINT with CMD as default args
# override with user provided args
if [ $# -gt 0 ]; then
SINGULARITY_OCI_RUN="${OCI_ENTRYPOINT} $@"
else
SINGULARITY_OCI_RUN="${OCI_ENTRYPOINT} ${OCI_CMD}"
fi
exec $SINGULARITY_OCI_RUN
The -t
or --test
flag shows the test script for the image.
$ {command} inspect --test jupyter.sif
This will output the corresponding %test
section from the definition file.
The -e
or --environment
flag shows the environment variables that are defined in the container image. These may be set from one or more environment files, depending on how the container was built.
$ {command} inspect --environment jupyter.sif
And the output would look like:
==90-environment.sh==
#!/bin/sh
JUP_PORT=8888
JUP_IPNAME=localhost
export JUP_PORT JUP_IPNAME
The -H
or -helpfile
flag will show the container's description in the %help
section of its definition file.
You can call it this way:
$ {command} inspect --helpfile jupyter.sif
And the output would look like:
Container with Anaconda 2 (Conda 4.5.11 Canary) and Jupyter Notebook 5.6.0 for Debian 9.x (Stretch).
This installation is based on Python 2.7.15
This flag gives you the possibility to output your labels in a JSON format.
You can call it this way:
$ {command} inspect --json ubuntu.sif
And the output would look like:
{
"data": {
"attributes": {
"labels": {
"my.label": "version 1.2.3",
"OWNER": "Joana",
"org.label-schema.build-arch": "amd64",
"org.label-schema.build-date": "Thursday_12_November_2020_10:51:59_CST",
"org.label-schema.schema-version": "1.0",
"org.label-schema.usage.singularity.deffile.bootstrap": "docker",
"org.label-schema.usage.singularity.deffile.from": "ubuntu:latest",
"org.label-schema.usage.singularity.version": "3.7.0-rc.1"
}
}
},
"type": "container"
}
The /.singularity.d
directory in a container contains scripts and environment files that are used when a container is executed.
You should not manually modify files under /.singularity.d
, from your definition file during builds, or directly within your container image. {Project} replaces older action scripts dynamically, at runtime, to support new features. In the longer term, metadata will be moved outside of the container, and stored only in the SIF file metadata descriptor.
/.singularity.d/
├── actions
│ ├── exec
│ ├── run
│ ├── shell
│ ├── start
│ └── test
├── env
│ ├── 01-base.sh
| ├── 10-docker2singularity.sh
│ ├── 90-environment.sh
│ ├── 91-environment.sh
| ├── 94-appsbase.sh
│ ├── 95-apps.sh
│ └── 99-base.sh
├── labels.json
├── libs
├── runscript
├── runscript.help
├── {Project}
└── startscript
- actions: This directory contains helper scripts to allow the container to carry out the action commands. (e.g.
exec
,run
orshell
). In later versions of {Project}, these files may be dynamically written at runtime, and should not be modified in the container. - env: All
*.sh
files in this directory are sourced in alphanumeric order when the container is started. For legacy purposes there is a symbolic link called/environment
that points to/.singularity.d/env/90-environment.sh
. Whenever possible, avoid modifying or creating environment files manually to prevent potential issues building & running containers with future versions of {Project}. Additional facilities such as--env
and--env-file
are available to allow manipulation of the container environment at runtime. - labels.json: The json file that stores a containers labels described above.
- libs: At runtime the user may request some host-system libraries to be mapped into the container (with the
--nv
option for example). If so, this is their destination. - runscript: The commands in this file will be executed when the container is invoked with the
run
command or called as an executable. For legacy purposes there is a symbolic link called/singularity
that points to this file. - runscript.help: Contains the description that was added in the
%help
section. - {Project}: This is the definition file that was used to generate the container. If more than 1 definition file was used to generate the container additional {Project} files will appear in numeric order in a sub-directory called
bootstrap_history
. - startscript: The commands in this file will be executed when the container is invoked with the
instance start
command.