Dockerfile if else condition with external arguments

Question

I have dockerfile

FROM centos:7
ENV foo=42

then I build it

docker build -t my_docker .

and run it.

docker run -it -d  my_docker

Is it possible to pass arguments from command line and use it with if else in Dockerfile? I mean something like

FROM centos:7
if (my_arg==42)
     {ENV=TRUE}
else:
     {ENV=FALSE}

and build with this argument.

 docker build -t my_docker . --my_arg=42

Answer 1

It might not look that clean but you can have your Dockerfile (conditional) as follow:

FROM centos:7
ARG arg
RUN if [[ -z "$arg" ]] ; then echo Argument not provided ; else echo Argument is $arg ; fi

and then build the image as:

docker build -t my_docker . --build-arg arg=45

or

docker build -t my_docker .

Answer 2

There is an interesting alternative to the proposed solutions, that works with a single Dockerfile, require only a single call to docker build per conditional build and avoids bash.

Solution:

The following Dockerfile solves that problem. Copy-paste it and try it yourself.

ARG my_arg

FROM centos:7 AS base
RUN echo "do stuff with the centos image"

FROM base AS branch-version-1
RUN echo "this is the stage that sets VAR=TRUE"
ENV VAR=TRUE

FROM base AS branch-version-2
RUN echo "this is the stage that sets VAR=FALSE"
ENV VAR=FALSE

FROM branch-version-${my_arg} AS final
RUN echo "VAR is equal to ${VAR}"

Explanation of Dockerfile:

We first get a base image (centos:7 in your case) and put it into its own stage. The base stage should contain things that you want to do before the condition. After that, we have two more stages, representing the branches of our condition: branch-version-1 and branch-version-2. We build both of them. The final stage than chooses one of these stages, based on my_arg. Conditional Dockerfile. There you go.

Output when running:

(I abbreviated this a little...)

my_arg==2

docker build --build-arg my_arg=2 .
Step 1/12 : ARG my_arg
Step 2/12 : ARG ENV
Step 3/12 : FROM centos:7 AS base
Step 4/12 : RUN echo "do stuff with the centos image"
do stuff with the centos image
Step 5/12 : FROM base AS branch-version-1
Step 6/12 : RUN echo "this is the stage that sets VAR=TRUE"
this is the stage that sets VAR=TRUE
Step 7/12 : ENV VAR=TRUE
Step 8/12 : FROM base AS branch-version-2
Step 9/12 : RUN echo "this is the stage that sets VAR=FALSE"
this is the stage that sets VAR=FALSE
Step 10/12 : ENV VAR=FALSE
Step 11/12 : FROM branch-version-${my_arg}
Step 12/12 : RUN echo "VAR is equal to ${VAR}"
VAR is equal to FALSE

my_arg==1

docker build --build-arg my_arg=1 .
...
Step 11/12 : FROM branch-version-${my_arg}
Step 12/12 : RUN echo "VAR is equal to ${VAR}"
VAR is equal to TRUE

Thanks to Tõnis for this amazing idea!

Answer 3

Do not use build args described in other answers where at all possible. This is an old messy solution. Docker's target property solves for this issue.

Target Example

Dockerfile

FROM foo as base

RUN ...

# Build dev image
FROM base as image-dev

RUN ...
COPY ...

# Build prod image
FROM base as image-prod

RUN ...
COPY ...

docker build --target image-dev -t foo .

version: '3.4'

services:

  dev:
    build:
      context: .
      dockerfile: Dockerfile
      target: image-dev

Real World

Dockerfiles get complex in the real world. Use buildkit & COPY --from for faster, more maintainable Dockerfiles:

• Docker builds every stage above the target, regardless of whether it is inherited or not. Use buildkit to build only inherited stages. Docker must by v19+. Hopefully this will be a default feature soon.
• Targets may share build stages. Use COPY --from to simplify inheritance.

FROM foo as base
RUN ...
WORKDIR /opt/my-proj

FROM base as npm-ci-dev
# invalidate cache
COPY --chown=www-data:www-data ./package.json /opt/my-proj/package.json
COPY --chown=www-data:www-data ./package-lock.json /opt/my-proj/package-lock.json
RUN npm ci

FROM base as npm-ci-prod
# invalidate cache
COPY --chown=www-data:www-data ./package.json /opt/my-proj/package.json
COPY --chown=www-data:www-data ./package-lock.json /opt/my-proj/package-lock.json
RUN npm ci --only=prod

FROM base as proj-files
COPY --chown=www-data:www-data ./ /opt/my-proj

FROM base as image-dev
# Will mount, not copy in dev environment
RUN ...

FROM base as image-ci
COPY --from=npm-ci-dev /opt/my-proj .
COPY --from=proj-files /opt/my-proj .
RUN ...

FROM base as image-stage
COPY --from=npm-ci-prod /opt/my-proj .
COPY --from=proj-files /opt/my-proj .
RUN ...

FROM base as image-prod
COPY --from=npm-ci-prod /opt/my-proj .
COPY --from=proj-files /opt/my-proj .
RUN ...

Enable experimental mode.

sudo echo '{"experimental": true}' | sudo tee /etc/docker/daemon.json

Build with buildkit enabled. Buildkit builds without cache by default - enable with --build-arg BUILDKIT_INLINE_CACHE=1

CI build job.

DOCKER_BUILDKIT=1 \
    docker build \
    --build-arg BUILDKIT_INLINE_CACHE=1 \
    --target image-ci\
    -t foo:ci
    .

Use cache from a pulled image with --cache-from

Prod build job

docker pull foo:ci
docker pull foo:stage

DOCKER_BUILDKIT=1 \
    docker build \
    --cache-from foo:ci,foo:stage \
    --target image-prod \
    -t prod
    .

Answer 4

From some reason most of the answers here didn't help me (maybe it's related to my FROM image in the Dockerfile)

So I preferred to create a bash script in my workspace combined with --build-arg in order to handle if statement while Docker build by checking if the argument is empty or not

Bash script:

#!/bin/bash -x

if test -z $1 ; then 
    echo "The arg is empty"
    ....do something....
else 
    echo "The arg is not empty: $1"
    ....do something else....
fi

Dockerfile:

FROM ...
....
ARG arg
COPY bash.sh /tmp/  
RUN chmod u+x /tmp/bash.sh && /tmp/bash.sh $arg
....

Docker Build:

docker build --pull -f "Dockerfile" -t $SERVICE_NAME --build-arg arg="yes" .

Remark: This will go to the else (false) in the bash script

docker build --pull -f "Dockerfile" -t $SERVICE_NAME .

Remark: This will go to the if (true)

Edit 1:

After several tries I have found the following article and this one which helped me to understand 2 things:

1) ARG before FROM is outside of the build

2) The default shell is /bin/sh which means that the if else is working a little bit different in the docker build. for example you need only one "=" instead of "==" to compare strings.

So you can do this inside the Dockerfile

ARG argname=false   #default argument when not provided in the --build-arg
RUN if [ "$argname" = "false" ] ; then echo 'false'; else echo 'true'; fi

and in the docker build:

docker build --pull -f "Dockerfile" --label "service_name=${SERVICE_NAME}" -t $SERVICE_NAME --build-arg argname=true .

Answer 5

The accepted answer may solve the question, but if you want multiline if conditions in the dockerfile, you can do that placing \ at the end of each line (similar to how you would do in a shell script) and ending each command with ;. You can even define someting like set -eux as the 1st command.

Example:

RUN set -eux; \
  if [ -f /path/to/file ]; then \
    mv /path/to/file /dest; \
  fi; \
  if [ -d /path/to/dir ]; then \
    mv /path/to/dir /dest; \
  fi

In your case:

FROM centos:7
ARG arg
RUN if [ -z "$arg" ] ; then \
    echo Argument not provided; \
  else \
    echo Argument is $arg; \
  fi

Then build with:

docker build -t my_docker . --build-arg arg=42

Answer 6

Just use the "test" binary directly to do this. You also should use the noop command ":" if you don't want to specify an "else" condition, so docker does not stop with a non zero return value error.

RUN test -z "$YOURVAR" || echo "var is set" && echo "var is not set"
RUN test -z "$YOURVAR" && echo "var is not set" || :
RUN test -z "$YOURVAR" || echo "var is set" && :

Answer 7

According to the doc for the docker build command, there is a parameter called --build-arg.

Example usage:

docker build --build-arg HTTP_PROXY=http://10.20.30.2:1234 .

IMO it's what you need :)

Answer 8

Exactly as others told, shell script would help.

Just an additional case, IMHO it's worth mentioning (for someone else who stumble upon here, looking for an easier case), that is Environment replacement.

Environment variables (declared with the ENV statement) can also be used in certain instructions as variables to be interpreted by the Dockerfile.

The ${variable_name} syntax also supports a few of the standard bash modifiers as specified below:

• ${variable:-word} indicates that if variable is set then the result will be that value. If variable is not set then word will be the result.

• ${variable:+word} indicates that if variable is set then word will be the result, otherwise the result is the empty string.

Answer 9

Using Bash script and Alpine/Centos

Dockerfile

FROM alpine  #just change this to centos 

ARG MYARG=""
ENV E_MYARG=$MYARG

ADD . /tmp
RUN chmod +x /tmp/script.sh && /tmp/script.sh

script.sh

#!/usr/bin/env sh

if [ -z "$E_MYARG" ]; then
    echo "NO PARAM PASSED"
else
    echo $E_MYARG
fi

Passing arg: docker build -t test --build-arg MYARG="this is a test" .

....
Step 5/5 : RUN chmod +x /tmp/script.sh && /tmp/script.sh
 ---> Running in 10b0e07e33fc
this is a test
Removing intermediate container 10b0e07e33fc
 ---> f6f085ffb284
Successfully built f6f085ffb284

Without arg: docker build -t test .

....
Step 5/5 : RUN chmod +x /tmp/script.sh && /tmp/script.sh
 ---> Running in b89210b0cac0
NO PARAM PASSED
Removing intermediate container b89210b0cac0
....

Answer 10

I saw a lot of possible solutions, but no one fits on the problem I faced today. So, I'm taking time to answer the question with one another possible solution that worked to me.

In my case I toke advantage of the well known if [ "$VAR" == "this" ]; then echo "do that"; fi. The caveat is that Docker, I don't know explain why, doesn't like the double equal on this case. So we need to write like that if [ "$VAR" = "this" ]; then echo "do that"; fi.

There is the full example that worked in my case:

FROM node:16

# Let's set args and envs
ARG APP_ENV="dev"
ARG NPM_CMD="install"
ARG USER="nodeuser"
ARG PORT=8080
ENV NPM_CONFIG_PREFIX=/home/node/.npm-global
ENV PATH=$PATH:/home/node/.npm-global/bin
ENV NODE_ENV=${APP_ENV}

# Let's set the starting point
WORKDIR /app

# Let's build a cache
COPY package*.json .
RUN date \
 # If the environment is production or staging, omit dev packages
 # If any other environment, install dev packages
 && if [ "$APP_ENV" = "production" ]; then NPM_CMD="ci --omit=dev"; fi \
 && if [ "$APP_ENV" = "staging" ]; then NPM_CMD="ci --omit=dev"; fi \
 && npm ${NPM_CMD} \
 && usermod -d /app -l ${USER} node

# Let's add the App
COPY . .

# Let's expose the App port
EXPOSE ${PORT}

# Let's set the user
USER ${USER}

# Let's set the start App command
CMD [ "node", "server.js" ]

So if the user pass the proper build argument, the docker build command will create an image of app for production. If not, it will create an image of the app with dev Node.js packages.

To make it works, you can call like this:

# docker build --build-arg APP_ENV=production -t app-node .

Answer 11

For any one trying to build Windows based image, you need to access argument with %% for cmd.

# Dockerfile Windows
# ...
ARG SAMPLE_ARG
RUN if %SAMPLE_ARG% == hello_world ( `
    echo hehe %SAMPLE_ARG% `
    ) else ( `
    echo haha %SAMPLE_ARG% `
)
# ...

BTW, ARG declaration must be placed after FROM, otherwise the argument will not be available.

Answer 12

# The ARGs in front of FROM is for image
ARG IMLABEL=xxxx \
    IMVERS=x.x

FROM ${IMLABEL}:${IMVERS}

# The ARGs after FROM is for parameters to be used in the script
ARG  condition-x

RUN if [ "$condition-x" = "condition-1" ]; then \
      echo "$condition-1"; \
    elif [ "$condition-x" = "condition-1" ]; then \
      echo "$condition-2"; \
    else
      echo "$condition-others"; \
    fi

build -t --build-arg IMLABEL --build-arg IMVERS --build-arg condition-x -f Dockerfile -t image:version .

Answer 13

I had a similar issue for setting proxy server on a container.

The solution I'm using is an entrypoint script, and another script for environment variables configuration. Using RUN, you assure the configuration script runs on build, and ENTRYPOINT when you run the container.

--build-arg is used on command line to set proxy user and password.

As I need the same environment variables on container startup, I used a file to "persist" it from build to run.

The entrypoint script looks like:

#!/bin/bash
# Load the script of environment variables
. /root/configproxy.sh
# Run the main container command
exec "$@"

configproxy.sh

#!/bin/bash

function start_config {
read u p < /root/proxy_credentials

export HTTP_PROXY=http://$u:$p@proxy.com:8080
export HTTPS_PROXY=https://$u:$p@proxy.com:8080

/bin/cat <<EOF > /etc/apt/apt.conf 
Acquire::http::proxy "http://$u:$p@proxy.com:8080";
Acquire::https::proxy "https://$u:$p@proxy.com:8080";
EOF
}

if [ -s "/root/proxy_credentials" ]
then
start_config
fi

And in the Dockerfile, configure:

# Base Image
FROM ubuntu:18.04

ARG user
ARG pass

USER root

# -z the length of STRING is zero
# [] are an alias for test command
# if $user is not empty, write credentials file
RUN if [ ! -z "$user" ]; then echo "${user} ${pass}">/root/proxy_credentials ; fi

#copy bash scripts
COPY configproxy.sh /root
COPY startup.sh .

RUN ["/bin/bash", "-c", ". /root/configproxy.sh"]

# Install dependencies and tools
#RUN apt-get update -y && \
#    apt-get install -yqq --no-install-recommends \
#    vim iputils-ping

ENTRYPOINT ["./startup.sh"]
CMD ["sh", "-c", "bash"]

Build without proxy settings

docker build -t img01 -f Dockerfile . 

Build with proxy settings

docker build -t img01 --build-arg user=<USER> --build-arg pass=<PASS> -f Dockerfile . 

Take a look here.

Answer 14

You can just add a simple check:

RUN [ -z "$ARG" ] \
    && echo "ARG argument not provided." \
    && exit 1 || exit 0

Answer 15

I ended up with writing my own framework for Dockerfile generation based on a python-like syntax.

It's open source now, you can try it here: https://github.com/gkpln3/sand

You can write your Sandfile like this:

# Sandfile
from sand import *

From("ubuntu", Tag="20.04")
Run([
    "apt-get update",
    "apt-get install ffmpeg python3"
])

# Install python debugger on debug images.
if config.DEBUG:
    Run("pip3 install pdb")

Copy("app", "/app")
Entrypoint("python3 /app/app.py")

And it will be converted to this

# Auto-generated by Sand, do not edit!
FROM ubuntu:20.04
RUN apt-get update
RUN apt-get install ffmpeg python3
RUN pip3 install pdb
COPY app /app
ENTRYPOINT python3 /app/app.py

You can pass the configuration values using the -D or --set flag.

$ sand config -DDEBUG=True

Or use a YAML file.

# sand.yaml
DEBUG: True

$ sand config --values sand.yaml