I'm facing a strange problem (or better: two different, weird problems) trying to pass build-args to my Dockerfile through docker-compose up.
My files - initial setup
Dockerfile:
ARG NODE_VERSION
FROM node:${NODE_VERSION}
ARG NPM_REGISTRY_TOKEN
RUN echo "=====> token ${NPM_REGISTRY_TOKEN}"
... ... ...
docker-compose.yml:
version: '3'
services:
myservice:
build:
context: ./myservice
dockerfile: ../Dockerfile
args:
- NODE_VERSION=10.15.1-alpine
- NPM_REGISTRY_TOKEN
With this initial setup in place, I have the following behaviour (on Linux Mint 20, docker-compose version 1.26.2, build eefe0d31):
running docker build --build-arg NPM_REGISTRY_TOKEN=xyz123 produces in output =====> token xyz123: the NPM_REGISTRY_TOKEN arg flows to the Dockerfile
running docker-compose build --build-arg NPM_REGISTRY_TOKEN=xyz123 myservice produces in output =====> token xyz123: the NPM_REGISTRY_TOKEN arg flows to the Dockerfile
running NPM_REGISTRY_TOKEN=xyz123 docker-compose up myservice produces in output =====> token : the NPM_REGISTRY_TOKEN env arg should flow to the Dockerfile due to - NPM_REGISTRY_TOKEN (according to https://docs.docker.com/compose/compose-file/#args: You can omit the value when specifying a build argument, in which case its value at build time is the value in the environment where Compose is running) but it seems to not be available during build
My files - reloaded
Simply changing my docker-compose.yml file to
version: '3'
services:
myservice:
build:
context: ./myservice
args:
- NODE_VERSION=10.15.1-alpine
- NPM_REGISTRY_TOKEN
dockerfile: ../Dockerfile
seems to solve the problem: switching args and dockerfile entries in yml file unlocks the capability to pass environment variables to Dockerfile as build-args through docker-compose up, too. Problem solved. Or not?
Changing OS, getting new problem
So, developers in my team use a bunch of different operating systems: Linux, Mac Os, and Windows, too.
Running the same commands on the same version (1.26.2) of docker-compose on Windows 10 Professional 1909 we're getting the same problem we faced initially, both using the initial version of the docker-compose.yml file and using the version that works on Linux.
We tried passing env var from command line, setting them in the command prompt, setting them as system variables through GUI... we tried launching docker-compose up for git-bash, too, but we're not able to get the variable value in Dockerfile.
I googled a bit aaround but I've not found any reference to known bugs or limitation of the Windows version of docker-compose.
Anyone have any idea what the problem might be? Thank you very much in advance!
So, finally, after some try-and-fail on different OSs and with different configurations, I ended up with an explanation of my problem - and therefore with a viable workaround, which allowed me to reach a satisfactory configuration for my docker-compose-yml file.
Short answer: it wasn't a matter of OSs nor env var passing nor order of context / dockerfile sections - it was a matter of clash between different services in my compose file.
More in detail: my docker-compose.yml file contained an additional service, too, whose job was to initialize the database the application was pointing to:
version: '3'
services:
myservice:
build:
context: ./myservice
dockerfile: ../Dockerfile
args:
- NODE_VERSION=10.15.1-alpine
- NPM_REGISTRY_TOKEN
depends_on:
- persistence
- db_initializer
command: sh -c './wait-for localhost:5432 -- ./wait-for localhost:15672 -- npm run start:dev'
persistence:
# Setting up the DBMS here
db_initializer:
build:
context: ./myservice
dockerfile: ../Dockerfile
args:
- NODE_VERSION=10.15.1-alpine
depends_on:
- persistence
command: sh -c './wait-for localhost:5432 -- ./wait-for localhost:15672 -- npm run db:migrate'
So, the problem was that I was configuring two services based on the same, self-build image, launching it with different commands (npm run db:migrate for the db_initializer service, npm run start:dev for the application service). Apparently compose took the configuration provided for the first initialized service (db_initializer, because myservice was dependant on it) and used that configuration for both services, ignoring the (different) args section I was providing for the second container: so I was able to solve (this time really!) the problem simply merging services declaration, including all args I needed:
version: '3'
services:
myservice:
build:
context: ./myservice
dockerfile: ../Dockerfile
args:
- NODE_VERSION=10.15.1-alpine
- NPM_REGISTRY_TOKEN
depends_on:
- persistence
- db_initializer
command: sh -c './wait-for localhost:5432 -- ./wait-for localhost:15672 -- run db:migrate && npm run start:dev'
persistence:
# Setting up the DBMS here
So, after a bunch of months without collecting answers, I think it's time to share my experience, hoping it can help someone encountering this weird behaviour.
Related
Actually I have checked some questions like this
What I do not understand is; if I change my docker-compose.yml and add profile to it then should I leave the Dockerfile without profile ?
For example my docker-compose file:
backend:
container_name: backend
image: backend
build: ./backend
restart: always
deploy:
restart_policy:
condition: on-failure
max_attempts: 15
ports:
- '8080:8080'
environment:
- MYSQL_ROOT_PASSWORD=DbPass3008
- MYSQL_PASSWORD=DbPass3008
- MYSQL_USER=DbUser
- MYSQL_DATABASE=db
depends_on:
- mysql
And I will add:
environment:
- "SPRING_PROFILES_ACTIVE=test
As far as I understand I need to put 3 different compose file and run them with -f parameter for different environments like:
docker-compose -f docker-compose-local/test/prod up -d
But my question is that my Dockerfile is already specifying profile as:
FROM openjdk:17-oracle
ADD ./target/backend-0.0.1-SNAPSHOT.jar backend.jar
EXPOSE 8080
ENTRYPOINT ["java","-jar", "-Dspring.profiles.active=TEST", "backend.jar"]
So how should I change this Dockerfile? Even if I create 3-4 different compose file, they are all using same Dockerfile. Should I create different Dockerfiles too (seems ridicilous) but what is the correct way ?
There's no need to add a java -Dspring.profiles.active=... command-line option; Spring will recognize the runtime SPRING_PROFILES_ACTIVE environment variable on its own. That means all of your environments can use the same image (which is generally a good practice).
Compose can also expand host environment variables in some contexts, so you may be able to use a single Compose file with environment-variable references
version: '3.8'
services:
backend:
environment:
- SPRING_PROFILES_ACTIVE=${ENVIRONMENT:-dev}
ENVIRONMENT=test docker-compose up -d
I tend to discourage putting environment-specific settings in a src/main/resources/*.yml file, since it means you need to recompile the application jar file whenever you deploy to a new environment. Another possibility is to set most Spring properties as environment variables, and then use multiple Compose files to include environment-specific settings. The one downside here is that you need multiple docker-compose -f options and you need to repeat them on every docker-compose invocation.
I am trying Teamcity on-premise solution using docker-compose.
I have a project which needs a docker compose runner type.
Here is the docker compose override yaml file:
version: '3'
services:
nextjs:
build:
context: ./frontend
dockerfile: ${NEXTJS_DOCKERFILE:-Dockerfile}
volumes:
- ./frontend:/app
- /app/node_modules
- /app/.next
nginx:
build: ./nginx
ports:
- 80:80
I have added the env variable using parameter on build configuration too:
The docker compose step gives the following error:
It seems like the variable substitution is not happening, I am not sure why.
Could anyone tell me what is happening?
Team city version : 2022.04.2
Thanks
So I know that there are a lot of tutorials on the topics, both docker and maven, but I'm having some confusion in combining them alltogether.
I created a multi-module Maven project with 2 modules, 2 spring applications, let's call them application 1 and application 2.
Starting each other via IntelliJ IDEA green "run" button works fine, now I'd like to automate things and run via docker.
I have Dockerfiles that looks the same in both cases:
(in both modules it's the same, only JAR name's different)
FROM adoptopenjdk:11-jre-hotspot
MAINTAINER *my name here lol*
ADD https://github.com/ufoscout/docker-compose-wait/releases/download/2.9.0/wait /wait
RUN chmod +x /wait
ARG JAR_FILE=target/*.jar
COPY ${JAR_FILE} application1-0.0.1-SNAPSHOT-jar-with-dependencies.jar
ENTRYPOINT ["java","-jar","/application1-0.0.1-SNAPSHOT-jar-with-dependencies.jar"]
CMD /wait && /*.jar
I also have docker-compose:
version: '2.1'
services:
application1:
container_name: app1
build:
context: ../app1
image: docker.io/myname/app1:latest
hostname: app1
ports:
- "8080:8080"
networks:
- spring-cloud-network-app1
application2:
container_name: app2
build:
context: ../app2
depends_on:
application1:
condition: service_started
links:
- application1
image: docker.io/myname/app2:latest
environment:
WAIT_HOSTS: application1:8080
ports:
- "8070:8070"
networks:
- spring-cloud-network-app2
networks:
spring-cloud-network-app1:
driver: bridge
spring-cloud-network-app2:
driver: bridge
What I do currently is:
I run maven package for each module and receive files like "application1(-2)-0.0.1-SNAPSHOT-jar-with-dependencies.jar" in both target folders.
"docker build -t springio/app1 ."
"docker-compose up --build"
And it works, but I feel I do some extra steps.
How can I do the project so that I ONLY have to run docker compose?
(after each time I change things in the code)
Again, I know it's a quite simple thing but I kinda lost the logic.
Thanks!
P.S
Ah, and about the "...docker-compose-wait/releases/download/2.9.0/wait /wait"
It's important that app start one after another, tried different solutions, unfortunately, doesn't really work as good as I would like to. But I guess I'll leave it as is.
So, again, if anyone ever wonders how to do the things I asked, here's the answer: you need multi-stage build Dockerfile.
It'll look like this:
#
# Build stage
#
FROM maven:3.6.0-jdk-11-slim AS build
COPY src /home/app/src
COPY pom.xml /home/app
RUN mvn -f /home/app/pom.xml clean package
#
# Package stage
#
FROM openjdk:11-jre-slim
COPY --from=build /home/app/target/demo-0.0.1-SNAPSHOT.jar /usr/local/lib/demo.jar
EXPOSE 8080
ENTRYPOINT ["java","-jar","/usr/local/lib/demo.jar"]
What it does is it basically first creates a jar file, copies it into package stage and eventually runs.
That's allow you to run your app in docker by running only docker compose.
I want to test ELK.
It works fine
BUt when I want to do a
docker-compose up
behind a proxy
docker-compose up --no-recreate
Building kibana
Step 1 : FROM kibana:latest
---> 544887fbfa30
Step 2 : RUN apt-get update && apt-get install -y netcat
---> Running in 794342b9d807
It failed
W: Some index files failed to download. They have been ignored, or old ones used instead.
Is' OK with
docker build --build-arg http_proxy=http://proxy:3128 --build-arg https_proxy=http://proxy:3128 kibana
But when I redo a docker-compose up, il tries to re-build, and failed to pass through proxy
Any help ?
You will need docker-compose 1.6.0-rc1 in order to pass the proxy to your build through docker-compose.
See commit 47e53b4 from PR 2653 for issue 2163.
Move all build related configuration into a build: section in the service.
Example:
web:
build:
context: .
dockerfile: Dockerfile.name
args:
key: value
As mkjeldsen points out in the comments
If key should assume the value of an environment variable of the same name, value can be omitted (docker-compose ARGS):
Especially useful for https_proxy: if the envvar is unset or empty, the builder will not apply proxy, otherwise it will.
I ran into the same problem. What helped me was using the explicit version 2.2 and then build - args and - network as described in the documentation.
VonC is right, it works for me by adding args section under the build lines in docker-compose file:
original:
ssh:
build: ssh/.
container_name: ssh
ports:
- "3000:22"
networks:
vault_net:
ipv4_address: 172.16.238.20
Modified:
ssh:
build:
context: "ssh/."
args:
HTTP_PROXY: http://X.X.X.X:XXXX
HTTPS_PROXY: http://X.X.X.X:XXXX
NO_PROXY: .domain.ltd,127.0.0.1
container_name: ssh
ports:
- "3000:22"
networks:
vault_net:
ipv4_address: 172.16.238.20
Note that I have to add quotes for context since it needs to be formatted as string.
Thanks a lot.
did you try it on clean machine?
docker-machine stop default
docker-machine create -d virtualbox test
docker-machine start test
eval $(docker-machine env test)
docker-compose up
I have a Dockerised application which I would like to run in both proxy and non-proxy host environments. I'm trying to resolve this problem by copying the normal environment variables, such as http_proxy, into the containers if and only if they exist in the host.
I can get 90% of the way there by running
set | grep -i _proxy=>proxies.env
in a top-level script, and then having, in my docker-compose.yml:
myserver:
build: ./myserver
env_file:
- proxies.env
This copies the host's environmental proxy variables, if any, into the server container, and it works in the sense that these variables are available at container run time, in other words by the stage that the Dockerfile CMD or ENTRYPOINT executes.
However I have one container which needs to run npm as a build step, ie from a RUN command in the Dockerfile, and these variables appear not to be present at this stage, so npm can't find the proxy and hangs. In other works, if I have
RUN set
in my Dockerfile, I can't see any variables from proxies.env, but if I do
docker exec -it myserver /bin/bash
and then run set, I can see everything from proxies.env.
Can anyone recommend a way to make these variables visible at container build time, without having to hard-code them, so that my docker-compose.yml and Dockerfile will still work both for hosts with proxies and hosts without proxies?
(Running with centos 7, docker-compose 1.3.1 and docker 1.7.0)
Update 2016, docker-compose 1.6.2, docker 1.10+, with a docker-compose.yml version 2:
You now have the args: sub-section of the build: section, which includes that very interesting possibility:
Build arguments with only a key are resolved to their environment value on the machine Compose is running on.
See PR 2653 (January 2016)
As a result, a way to introduce the proxy variables without hard-coding them in the docker-compose.yml file itself is with that precise syntax:
version: '2'
services:
myservice:
build:
context: .
args:
- http_proxy
- https_proxy
- no_proxy
Before calling docker-compose up, you need to make sure your proxy environment variables are set:
export http_proxy=http://username:password#proxy.com:port
export https_proxy=http://username:password#proxy.com:port
export no_proxy=localhost,127.0.0.1,company.com
docker-compose up
Then your Dockerfile built by the docker-compose process will pick up automatically the proxy variable values, even though the docker-compose.yml does not include any hard-coded specific values.
May be you the "environment" option solves your problem. In your docker compose file would looks like:
myserver:
build: ./myserver
environment:
- HTTP_PROXY=192.168.1.8
- VARIABLE=value
- ...
Maybe you can try this:
Before you call RUN, ADD the .env file into the image
ADD proxies.env proxies.env
then prefix your RUN statement:
RUN export `cat proxies.env` && echo "FOO is $FOO and BAR is $BAR"
This produces the following output:
root#armenubuntudev:~/Dockers/set-env# docker build -t ashimoon/envtest .
Sending build context to Docker daemon 3.584 kB
Sending build context to Docker daemon
Step 0 : FROM ubuntu
---> 91e54dfb1179
Step 1 : ADD proxies.env proxies.env
---> Using cache
---> 181d0e082e65
Step 2 : RUN export `cat proxies.env` && echo "FOO is $FOO and BAR is $BAR"
---> Running in 30426910a450
FOO is 1 and BAR is 2
---> 5d88fcac522c
Removing intermediate container 30426910a450
Successfully built 5d88fcac522c
docker-compose.yml
...
server:
build: .
args:
env: $ENV
...
Dockerfile
ARG env
ENV NODE_ENV $env
This example fixes YUM.
version: '2'
services:
example-service:
build:
context: .
args:
http_proxy: proxy.example.com:80