I use docker and docker compose to package scientific tools into easily/universally executable modules. One example is a docker that packages a rather complicated python library into a container that runs a jupyter notebook server; the idea is that other scientists who are not terribly tech-savvy can clone a github repository, run docker-compose up then do their analyses without having to install the library, configure various plugins and other dependencies, etc.
I have this all working fine except that I'm having issues getting the volume mounts to work in a coherent fashion. The reason for this is that the library inside the docker container handles multiple kinds of datasets, which users will store in several separate directories that are conventionally tracked through shell environment variables. (Please don't tell me this is a bad way to do this--it's the way things are done in the field, not the way I've chosen to do things.) So, for example, if the user stores FreeSurfer data, they will have an environment variable named SUBJECTS_DIR that points to the directory containing the data; if they store HCP data, they will have an environment variable HCP_SUBJECTS_DIR. However, they may have both, either, or neither of these set (as well as a few others).
I would like to be able to put something like this in my docker-compose.yml file in order to handle these cases:
version: '3'
services:
my_fancy_library:
build: .
ports:
- "8080:8888"
environment:
- HCP_SUBJECTS_DIR="/hcp_subjects"
- SUBJECTS_DIR="/freesurfer_subjects"
volumes:
- "$SUBJECTS_DIR:/freesurfer_subjects"
- "$HCP_SUBJECTS_DIR:/hcp_subjects"
In testing this, if the user has both environment variables set, everything works swimmingly. However, if they don't have one of these set, I get an error about not mounting directories that are fewer than 2 characters long (which I interpret to be a complaint about mounting a volume specified by ":/hcp_subjects").
This question asks basically the same thing, and the answer points to here, which, if I'm understanding it right, basically explains how to have multiple docker-compose files that are resolved in some fashion. This isn't really a viable solution for my case for a few reasons:
This tool is designed for use by people who don't necessarily know anything about docker, docker-compose, or related utilities, so expecting them to write/edit their own docker-compose.yml file is a problem
There are more than just two of these directories (I have shown two as an example) and I can't realistically make a docker-compose file for every possible combination of these paths being declared or not declared
Honestly, this solution seems really clunky given that the information needed is right there in the variables that docker-compose is already reading.
The only decent solution I've been able to come up with is to ask the users to run a script ./run.sh instead of docker-compose up; the script examines the environment variables, writes out its own docker-compose.yml file with the appropriate volumes, and runs docker-compose up itself. This also seems somewhat clunky, but it works.
Does anyone know of a way to conditionally mount a set of volumes based on the state of the environment variables when docker-compose up is run?
You can set defaults for environment variable in a .env-file shipped alongside with a docker-compose.yml [1].
By setting your environment variables to /dev/null by default and then handling this case in the containerized application, you should be able to achieve what you need.
Example
$ tree -a
.
├── docker-compose.yml
├── Dockerfile
├── .env
└── run.sh
docker-compose.yml
version: "3"
services:
test:
build: .
environment:
- VOL_DST=${VOL_DST}
volumes:
- "${VOL_SRC}:${VOL_DST}"
Dockerfile
FROM alpine
COPY run.sh /run.sh
ENTRYPOINT ["/run.sh"]
.env
VOL_SRC=/dev/null
VOL_DST=/volume
run.sh
#!/usr/bin/env sh
set -euo pipefail
if [ ! -d ${VOL_DST} ]; then
echo "${VOL_DST} not mounted"
else
echo "${VOL_DST} mounted"
fi
Testing
Environment variable VOL_SRC not defined:
$ docker-compose up
Starting test_test_1 ... done
Attaching to test_test_1
test_1 | /volume not mounted
test_test_1 exited with code 0
Environment variable VOL_SRC defined:
$ VOL_SRC="./" docker-compose up
Recreating test_test_1 ... done
Attaching to test_test_1
test_1 | /volume mounted
[1] https://docs.docker.com/compose/environment-variables/#the-env-file
Even though #Ente's answer solves the problem, here is an alternative solution when you have more complex differences between environments.
Docker compose supports multiple docker-compose files for configuration overriding in different environments.
This is useful if you have different named volumes you need to potentially mount on the same path depending on the environment.
You can modify existing services or even add new ones, for instance:
# docker-compose.yml
version: '3.3'
services:
service-a:
image: "image-name"
volumes:
- type: volume
source: vprod
target: /data
ports:
- "80:8080"
volumes:
vprod:
vdev:
And then you have the override file to change the volume mapping:
# docker-compose.override.yml
services:
service-a:
volumes:
- type: volume
source: vdev
target: /data
When running docker-compose up -d both configurations will be merged with the override file taking precedence.
Docker compose picks up docker-compose.yml and docker-compose.override.yml by default, if you have more files, or files with different names, you need to specify them in order:
docker-compose -f docker-compose.yml -f docker-compose.custon.yml -f docker-compose.dev.yml up -d
Related
I have a few repos that I want to deploy using docker-compose within the same project, for each I have a docker-compose.yaml in the root of the directory that defines the list of services. The docker files are located in docker/{service}/Dockerfile relative to the repo root, and hence the docker-compose.yaml looks something like
version: "3"
services:
service1:
build:
context: .
dockerfile: ./docker/service1/Dockerfile
networks:
- default
service2:
build:
context: .
dockerfile: ./docker/service2/Dockerfile
networks:
- default
service3:
build:
context: .
dockerfile: ./docker/service3/Dockerfile
networks:
- default
networks:
default:
However when I run the command from the directory above the repo roots
docker-compose -p project -f repo1/docker-compose.yaml -f repo2/docker-compose.yaml build
I get the error ERROR: Cannot locate specified Dockerfile: ./docker/service1/Dockerfile
Which is a service from repo2
These commands run fine with just one file specified and only one set of services are "not found" (repo2's). I assume that the first file is therefore setting the context of compose, is there a way I can tell compose not to do this?
The documentation for the docker-compose -f option notes:
When you use multiple Compose files, all paths in the files are relative to the first configuration file specified with -f.
Practically, this probably means all of the Compose files need to be in the same directory. If you have a file that really has no host paths at all (neither a build: context directory nor volumes: bind mounts) it could in principle be somewhere else, but IME this is pretty unusual.
For the setup you describe, it might be more practical to launch the separate projects separately:
(cd repo1 && docker-compose up --build -d)
(cd repo2 && docker-compose up --build -d)
A more typical use of multiple Compose files is to provide options split across several files; for example, a base file that specifies image names and an override file that provides developer-oriented features like publishing database ports and building images from source. Share Compose configurations between files and projects describes this use case a little more. That page similarly comments:
Tracking which fragment of a service is relative to which path is difficult and confusing, so to keep paths easier to understand, all paths must be defined relative to the base file.
In short: it's an intentional feature of Compose that all path references in all Compose files are relative to the location of the first file, regardless of the location of the file that contains the path reference, and there's not an override option for it.
I am very (read very) new to Docker so experimenting. I have created a very basic Dockerfile to pull in Laravel:
FROM composer:latest
RUN composer_version="$(composer --version)" && echo $composer_version
RUN composer global require laravel/installer
WORKDIR /var/www
RUN composer create-project --prefer-dist laravel/laravel site
My docker-compose.yml file looks like:
version: '3.7'
services:
laravel:
build:
context: .
dockerfile: laravel.dockerfile
container_name: my_laravel
network_mode: host
restart: on-failure
volumes:
- ./site:/var/www/site
When I run docker-compose up, the ./site directory is created but the contents are empty. I've put this in docker-compose as I plan on on including other things like nginx, mysql, php etc
The command:
docker run -v "/where/i/want/data/site:/var/www/site" my_laravel
Results in the same behaviour.
I know the install is successful as I modified my dockerfile with the follwing two lines appended to it:
WORKDIR /var/www/site
RUN ls -la
Which gives me the correct listing.
Clearly misunderstanding something here. Any help appreciated.
EDIT: So, I was able to get this to work... although, it slightly more difficult than just specifying a path..
You can accomplish this by specifying a volume in docker-compose.yml.. The path to the directory (on the host) is labeled as device in the compose file.. It appears that the root of the path has to be an actual volume (possibly a share would work) but the 'destination' of the path can be a directory on the specified volume..
I created a new volume called docker on my machine but I suppose you could do this with your existing disk/volume..
I am on a Mac and this docker-compose.yml file worked for me:
version: '3.7'
services:
nodemon-test:
container_name: my-nodemon-test
image: oze4/nodemon-docker-test
ports:
- "1337:1337"
volumes:
- docker_test_app:/app # see comment below on which name to use here
volumes:
docker_test_app: # use this name under `volumes:` for the service
name: docker_test_app
driver: local
driver_opts:
o: bind
type: none
device: /Volumes/docker/docker_test_app
The container specified exists in my DockerHub.. this is the source code for it, just in case you are worried about anything malicious. I created it like two weeks ago to help someone else on StackOverflow.
Shows files from the container on my machine (the host)..
You can read more about Docker Volume configs here if you would like.
ORIGINAL ANSWER:
It looks like you are trying to share the build directory with your host machine.. After some testing, it appears Docker will overwrite the specified path on the container with the contents of the path on the host.
If you run docker logs my_laravel you should see an error about missing files at /var/www/site.. So, even though the build is successful - once Docker mounts the directory from your machine (./site) onto the container (/var/www/site) it overwrites the path within the container (/var/www/site) with the contents of the path on your host (./site) - which is empty.
To test and make sure the contents of /var/www/site are in fact being overwritten, you can run docker exec -it /bin/bash (you may need to replace /bin/bash with /bash).. This will give you command line access inside of the container. From there you can do ls -a /var/www/site..
Furthermore, you can also pre-stage ./site to have a random test file in it (test.txt or whatever), then docker-compose up -d, then run the same commands from the step above docker exec -it ... and see if the staged test.txt file is now inside the container - this gives you definitive evidence that when you run volumes, the data on your host overwrites data in the container.
With that being said, doing something like this and sharing a log directory will work... the volume path specified on the container is still overwritten, the difference is the container is writing to that path.. it doesn't rely on it for config files/app files.
Hope this helps.
Letzt imagine i have 3 compose files (only focus on the mysql service)
docker-compose.yml
docker-compose.staging.yml
docker-compose.prod.yml
In my docker compose.yml i have my basic mysql stuff with dev als build target
version: "3.4"
services:
mysql:
build:
target: dev
...
And start it with
docker-compose up -d
In my staging environment i would like to expose port 3306, but also want another build target so i would create the docker-compose.staging.yml with the following content.
version: "3.4"
services:
mysql:
build
target: prod
ports:
- 3306:3306
And combine it with
docker-compose -f docker-compose.yml -f docker-compose.staging.yml up -d
So the build target is overwritten and the port 3306 is now exposed to the outside.
Now i want the same in the docker-compose.prod.yml, just without having the port 3306 exposed to the outside ... How can i override the ports directive to not having ports exposed?
I tried to put an empty array in the prod.yml without success (port is still exposed):
version: "3.4"
services:
mysql:
ports: []
In the end i would like to stack the up command like this:
docker-compose -f docker-compose.yml -f docker-compose.staging.yml -f docker-compose.prod.yml up -d
I also know the docs says
For the multi-value options ports, expose, external_links, dns, dns_search, and tmpfs, Compose concatenates both sets of values
But how can i reach my goal anyway without duplicating configuration?
Yes for sure, i could omit the docker-compose.staging.yml but in the staging.yml are build steps defined, which should also be used for the prod stage to not have any differences between the built container.
So duplicating things isn't really an option.
Thanks
I would actually strongly suggest just not using the "target" command in your compose files. I find it to be extremely beneficial to build a single image for local/staging/production - build once, test it, and deploy it in each environment. In this case, you change things using environment variables or mounted secrets/config files.
Further, using compose to build the images is... fragile. I would recommend building the images in a CI system, pushing them to a registry, and then using the image version tags in your compose file- it is a much more reproducible system.
You might consider using extends key in your compose files like this:
mysql:
extends:
file: docker-compose.yml
service: mysql
ports:
- 3306:3306
# other definitions
Although you'd have to change your compose version from 3.4 to < 3 ( like 2.3 ) because v3 doesn't support this feature ref as there is a open feature request hanging for a long time now.
Important note here is that you shouldn't expose any ports in your base docker-compose.yml file, only on the specific composes.
Oficial docs ref for extends
edit
target clause is not supported in v2.0 so I've adjusted the answer to match the extends and target requirement. That's compose v2.3.
edit from comments
As there is a deploy keyword requirement, then there is compose v3 requirement. And as for now, there is no possibility to extend composes. I've read in some official doc (can't find it now for ref) that they encourage us to use flat composes specific for environment so that it's always clear. Also Docker states that's hard to implement in v3 (ref in the above issue) and it's not going to be implemented anywhere soon. You have to use separate compose files per environment.
I really did not know how to word the title.
In my system I have two instances:
a Prod server
a Dev server
Dev used mostly for testing. In each case I have two versions of AMQP both having different hostnames.
To avoid duplication or unnecessary time rewriting the same code in multiple projects I wanted to use the env file that docker compose has, though everywhere I read, no one discusses this case. That case being that depending on where a stack is deployed is which env file it would use and that env file existing on the swarm itself rather than the individual projects.
Hopefully, I didn't miss anything when explaining this. Summary being two swarms each having their own env file that the containers deployed to it can use. Also if I need to reword anything, I will do so.
you can have multiple .env files and assign them to services in docker-compose.yml like
web:
env_file:
- web-variables.env
nginx:
env_file:
- nginx-variables.env
and if you want to change them for development environemnt you could override the docker-compose.yml with docker-compose.development.ymlfile and then start it with
docker-compose -f docker-compose.yml -f docker-compose.development.yml up -d
I'm trying to wrap my head around Docker, but I'm having a hard time figuring it out. I tried to implement it in my small project (MERN stack), and I was thinking how do you distinct between development, (maybe staging), and production environments.
I saw one example where they used 2 Docker files, and 2 docker-compose files, (each pair for one env, so Dockerfile + docker-compose.yml for prod, Dockerfile-dev + docker-compose-dev.yml for dev).
But this just seems like a bit of an overkill for me. I would prefer to have it only in two files.
Also one of the problem is that e.g. for development I want to install nodemon globally, but not for poduction.
In perfect solution I imagine running something like that
docker-compose -e ENV=dev build
docker-compose -e ENV=dev up
Keep in mind, that I still don't fully get docker, so if you caught some of mine misconceptions about docker, you can point them out.
You could take some clues from "Using Compose in production"
You’ll almost certainly want to make changes to your app configuration that are more appropriate to a live environment. These changes may include:
Removing any volume bindings for application code, so that code stays inside the container and can’t be changed from outside
Binding to different ports on the host
Setting environment variables differently (e.g., to decrease the verbosity of logging, or to enable email sending)
Specifying a restart policy (e.g., restart: always) to avoid downtime
Adding extra services (e.g., a log aggregator)
The advice is then not quite similar to the example you mention:
For this reason, you’ll probably want to define an additional Compose file, say production.yml, which specifies production-appropriate configuration. This configuration file only needs to include the changes you’d like to make from the original Compose file.
docker-compose -f docker-compose.yml -f production.yml up -d
This overriding mechanism is better than trying to mix dev and prod logic in one compose file, with environment variable to try and select one.
Note: If you name your second dockerfile docker-compose.override.yml, a simple docker-compose up would read the overrides automatically.
But in your case, a name based on the environment is clearer.
Docker Compose will read docker-compose.yml and docker-compose.override.yml by default. Understanding-Multiple-Compose-Files
You can set a default docker-compose.yml and different overwrite compose file. For example, docker-compose.prod.yml docker-compose.test.yml. Keep them in the same place.
Then create a symbolic link named docker-compose.override.yml for each env.
Track docker-compose.{env}.yml files and add docker-compose.override.yml to .gitignore.
In prod env: ln -s ./docker-compose.prod.yml ./docker-compose.override.yml
In test env: ln -s ./docker-compose.test.yml ./docker-compose.override.yml
The project structure will then look like this:
project\
- docker-compose.yml # tracked
- docker-compose.prod.yml # tracked
- docker-compose.test.yml # tracked
- docker-compose.override.yml # ignored & linked to override composefile for current env
- src/
- ...
Then you have done. In each environment, you can use the compose-file with the same command docker-compose up
If you are not sure, use docker-compose config to check if it's been override properly.