I have two separate docker-compose.yml files in two different folders:
~/front/docker-compose.yml
~/api/docker-compose.yml
How can I make sure that a container in front can send requests to a container in api?
I know that --default-gateway option can be set using docker run for an individual container, so that a specific IP address can be assigned to this container, but it seems that this option is not available when using docker-compose.
Currently I end up doing a docker inspect my_api_container_id and look at the gateway in the output. It works but the problem is that this IP is randomly attributed, so I can't rely on it.
Another form of this question might thus be:
Can I attribute a fixed IP address to a particular container using docker-compose?
But in the end what I'm looking after is:
How can two different docker-compose projects communicate with each other?
You just need to make sure that the containers you want to talk to each other are on the same network. Networks are a first-class docker construct, and not specific to compose.
# front/docker-compose.yml
version: '2'
services:
front:
...
networks:
- some-net
networks:
some-net:
driver: bridge
...
# api/docker-compose.yml
version: '2'
services:
api:
...
networks:
- front_some-net
networks:
front_some-net:
external: true
Note: Your app’s network is given a name based on the “project name”, which is based on the name of the directory it lives in, in this case a prefix front_ was added
They can then talk to each other using the service name. From front you can do ping api and vice versa.
UPDATE: As of compose file version 3.5:
This now works:
version: "3.5"
services:
proxy:
image: hello-world
ports:
- "80:80"
networks:
- proxynet
networks:
proxynet:
name: custom_network
docker-compose up -d will join a network called 'custom_network'. If it doesn't exist, it will be created!
root#ubuntu-s-1vcpu-1gb-tor1-01:~# docker-compose up -d
Creating network "custom_network" with the default driver
Creating root_proxy_1 ... done
Now, you can do this:
version: "2"
services:
web:
image: hello-world
networks:
- my-proxy-net
networks:
my-proxy-net:
external:
name: custom_network
This will create a container that will be on the external network.
I can't find any reference in the docs yet but it works!
Just a small adittion to #johnharris85's great answer,
when you are running a docker compose file, a "default" network is created
so you can just add it to the other compose file as an external network:
# front/docker-compose.yml
version: '2'
services:
front_service:
...
...
# api/docker-compose.yml
version: '2'
services:
api_service:
...
networks:
- front_default
networks:
front_default:
external: true
For me this approach was more suited because I did not own the first docker-compose file and wanted to communicate with it.
All containers from api can join the front default network with following config:
# api/docker-compose.yml
...
networks:
default:
external:
name: front_default
See docker compose guide: using a pre existing network (see at the bottom)
The previous posts information is correct, but it does not have details on how to link containers, which should be connected as "external_links".
Hope this example make more clear to you:
Suppose you have app1/docker-compose.yml, with two services (svc11 and svc12), and app2/docker-compose.yml with two more services (svc21 and svc22) and suppose you need to connect in a crossed fashion:
svc11 needs to connect to svc22's container
svc21 needs to connect to svc11's container.
So the configuration should be like this:
this is app1/docker-compose.yml:
version: '2'
services:
svc11:
container_name: container11
[..]
networks:
- default # this network
- app2_default # external network
external_links:
- container22:container22
[..]
svc12:
container_name: container12
[..]
networks:
default: # this network (app1)
driver: bridge
app2_default: # external network (app2)
external: true
this is app2/docker-compose.yml:
version: '2'
services:
svc21:
container_name: container21
[..]
networks:
- default # this network (app2)
- app1_default # external network (app1)
external_links:
- container11:container11
[..]
svc22:
container_name: container22
[..]
networks:
default: # this network (app2)
driver: bridge
app1_default: # external network (app1)
external: true
Everybody has explained really well, so I'll add the necessary code with just one simple explanation.
Use a network created outside of docker-compose (an "external" network) with docker-compose version 3.5+.
Further explanation can be found here.
First docker-compose.yml file should define network with name giveItANamePlease as follows.
networks:
my-network:
name: giveItANamePlease
driver: bridge
The services of first docker-compose.yml file can use network as follows:
networks:
- my-network
In second docker-compose file, we need to proxy the network by using the network name which we have used in first docker-compose file, which in this case is giveItANamePlease:
networks:
my-proxy-net:
external:
name: giveItANamePlease
And now you can use my-proxy-net in services of a second docker-compose.yml file as follows.
networks:
- my-proxy-net
Since Compose 1.18 (spec 3.5), you can just override the default network using your own custom name for all Compose YAML files you need. It is as simple as appending the following to them:
networks:
default:
name: my-app
The above assumes you have version set to 3.5 (or above if they don't deprecate it in 4+).
Other answers have pointed the same; this is a simplified summary.
UPDATE: As of docker-compose file version 3.5:
I came across a similar problem and I solved it by adding a small change in one of my docker-compose.yml project.
For instance, we have two API's scoring and ner. Scoring API needs to send a request to the ner API for processing the input request. In order to do that they both are supposed to share the same network.
Note: Every container has its own network which is automatically created at the time of running the app inside docker. For example ner API network will be created like ner_default and scoring API network will be named as scoring default. This solution will work for version: '3'.
As in the above scenario, my scoring API wants to communicate with ner API then I will add the following lines. This means Whenever I create the container for ner API then it automatically added to the scoring_default network.
networks:
default:
external:
name: scoring_default
ner/docker-compose.yml
version: '3'
services:
ner:
container_name: "ner_api"
build: .
...
networks:
default:
external:
name: scoring_default
scoring/docker-compose.yml
version: '3'
services:
api:
build: .
...
We can see this how the above containers are now a part of the same network called scoring_default using the command:
docker inspect scoring_default
{
"Name": "scoring_default",
....
"Containers": {
"14a6...28bf": {
"Name": "ner_api",
"EndpointID": "83b7...d6291",
"MacAddress": "0....",
"IPv4Address": "0.0....",
"IPv6Address": ""
},
"7b32...90d1": {
"Name": "scoring_api",
"EndpointID": "311...280d",
"MacAddress": "0.....3",
"IPv4Address": "1...0",
"IPv6Address": ""
},
...
}
You can add a .env file in all your projects containing COMPOSE_PROJECT_NAME=somename.
COMPOSE_PROJECT_NAME overrides the prefix used to name resources, as such all your projects will use somename_default as their network, making it possible for services to communicate with each other as they were in the same project.
NB: You'll get warnings for "orphaned" containers created from other projects.
So many answers!
First of all, avoid hyphens in entities names such as services and networks. They cause issues with name resolution.
Example: my-api won't work. myapi or api will work.
What worked for me is:
# api/docker-compose.yml
version: '3'
services:
api:
container_name: api
...
ports:
- 8081:8080
networks:
- mynetwork
networks:
mynetwork:
name: mynetwork
and
# front/docker-compose.yml
version: '3'
services:
front:
container_name: front
...
ports:
- 81:80
networks:
- mynetwork
networks:
mynetwork:
name: mynetwork
NOTE: I added ports to show how services can access each other, and how they are accessible from the host.
IMPORTANT: If you don't specify a network name, docker-compose will craft one for you. It uses the name of the folder the docker_compose.yml file is in. In this case: api_mynetwork and front_mynetwork. That will prevent communication between containers since they will by on different network, with very similar names.
Note that the network is defined exactly the same in both file, so you can start either service first and it will work. No need to specify which one is external, docker-compose will take care of managing that for you.
From the host
You can access either container using the published ports defined in docker-compose.yml.
You can access the Front container: curl http://localhost:81
You can access the API container: curl http://localhost:8081
From the API container
You can access the Front container using the original port, not the one you published in docker-compose.yml.
Example: curl http://front:80
From the Front container
You can access the API container using the original port, not the one you published in docker-compose.yml.
Example: curl http://api:8080
For using another docker-compose network you just do these(to share networks between docker-compose):
Run the first docker-compose project by up -d
Find the network name of the first docker-compose by: docker network ls(It contains the name of the root directory project)
Then use that name by this structure at below in the second docker-compose file.
second docker-compose.yml
version: '3'
services:
service-on-second-compose: # Define any names that you want.
.
.
.
networks:
- <put it here(the network name that comes from "docker network ls")>
networks:
- <put it here(the network name that comes from "docker network ls")>:
external: true
I would ensure all containers are docker-compose'd to the same network by composing them together at the same time, using:
docker compose --file ~/front/docker-compose.yml --file ~/api/docker-compose.yml up -d
If you are
trying to communicate between two containers from different docker-compose projects and don't want to use the same network (because let's say they would have PostgreSQL or Redis container on the same port and you would prefer to not changing these ports and not use it at the same network)
developing locally and want to imitate communication between two docker compose projects
running two docker-compose projects on localhost
developing especially Django apps or Django Rest Framework (drf) API and running app inside container on some exposed port
getting Connection refused while trying to communicate between two containers
And you want to
container api_a communicate to api_b (or vice versa) without the same "docker network"
(example below)
you can use "host" of the second container as IP of your computer and port that is mapped from inside Docker container. You can obtain IP of your computer with this script (from: Finding local IP addresses using Python's stdlib):
import socket
def get_ip():
s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
try:
# doesn't even have to be reachable
s.connect(('10.255.255.255', 1))
IP = s.getsockname()[0]
except:
IP = '127.0.0.1'
finally:
s.close()
return IP
Example:
project_api_a/docker-compose.yml:
networks:
app-tier:
driver: bridge
services:
api:
container_name: api_a
image: api_a:latest
depends_on:
- postgresql
networks:
- app-tier
inside api_a container you are running Django app:
manage.py runserver 0.0.0.0:8000
and second docker-compose.yml from other project:
project_api_b/docker-compose-yml :
networks:
app-tier:
driver: bridge
services:
api:
container_name: api_b
image: api_b:latest
depends_on:
- postgresql
networks:
- app-tier
inside api_b container you are running Django app:
manage.py runserver 0.0.0.0:8001
And trying to connect from container api_a to api_b then URL of api_b container will be:
http://<get_ip_from_script_above>:8001/
It can be especially valuable if you are using even more than two(three or more) docker-compose projects and it's hard to provide common network for all of it - it's good workaround and solution
To connect two docker-compose you need a network and putting both docker-composes in that network,
you could create netwrok with docker network create name-of-network,
or you could simply put network declaration in networks option of docker-compose file and when you run docker-compose (docker-compose up) the network would be created automatically.
put the below lines in both docker-compose files
networks:
net-for-alpine:
name: test-db-net
Note: net-for-alpine is internal name of the network and it will be used inside of the docker-compose files and could be different,
test-db-net is external name of the network and must be same in two docker-compose files.
Assume we have docker-compose.db.yml and docker-compose.alpine.yml
docker-compose.apline.yml would be:
version: '3.8'
services:
alpine:
image: alpine:3.14
container_name: alpine
networks:
- net-for-alpine
# these two command keeps apline container running
stdin_open: true # docker run -i
tty: true # docker run -t
networks:
net-for-alpine:
name: test-db-net
docker-compose.db.yml would be:
version: '3.8'
services:
db:
image: postgres:13.4-alpine
container_name: psql
networks:
- net-for-db
networks:
net-for-db:
name: test-db-net
To test the network, go inside alpine container
docker exec -it alpine sh
then with following commands you could check the network
# if it returns 0 or see nothing as a result, network is established
nc -z psql (container name)
or
ping pgsql
I'm running multiple identical docker-compose.yml files in different directories, using .env files to make a slight difference. And use Nginx Proxy Manage to communicate with other services. here is my file:
make sure you have created a public network
docker network create nginx-proxy-man
/domain1.com/docker-compose.yml, /domain2.com/docker-compose.yml, ...
version: "3.9"
services:
webserver:
build:
context: ./bin/${PHPVERSION}
container_name: "${COMPOSE_PROJECT_NAME}-${PHPVERSION}"
...
networks:
- default # network outside
- internal # network internal
database:
build:
context: "./bin/${DATABASE}"
container_name: "${COMPOSE_PROJECT_NAME}-${DATABASE}"
...
networks:
- internal # network internal
networks:
default:
external: true
name: nginx-proxy-man
internal:
internal: true
.env file just change COMPOSE_PROJECT_NAME
COMPOSE_PROJECT_NAME=domain1_com
.
.
.
PHPVERSION=php56
DATABASE=mysql57
webserver.container_name: domain1_com-php56 - will join the default network (name: nginx-proxy-man), previously created for Nginx Proxy Manager to be accessible from the outside.
Note: container_name is unique in the same network.
database.container_name: domain1_com-mysql57 - easier to distinguish
In the same docker-compose.yml, the services will connect to each other via the service name because of the same network domain1_com_internal. And to be more secure, set this network with the option internal: true
Note, if you don't explicitly specify networks for each service, but just use a common external network for both docker-compose.yml, then it's likely that domain1_com will use domain2_com's database.
Another option is just running up the first module with the 'docker-compose' check the ip related with the module, and connect the second module with the previous net like external, and pointing the internal ip
example
app1 - new-network created in the service lines, mark as external: true at the bottom
app2 - indicate the "new-network" created by app1 when goes up, mark as external: true at the bottom, and set in the config to connect, the ip that app1 have in this net.
With this, you should be able to talk with each other
*this way is just for local-test focus, in order to don't do an over complex configuration
** I know is very 'patch way' but works for me and I think is so simple some other can take advantage of this
Answer for Docker Compose '3' and up
By default Docker Compose uses a bridge network to provision inter-container communication. Read this article for more info about inter-container networking.
What matters for you, is that by default Docker Compose creates a hostname that equals the service name in the docker-compose.yml file. Consider the following docker-compose.yml:
version: '3.9'
services:
server:
image: node:16.9.0
container_name: server
tty: true
stdin_open: true
depends_on:
- mongo
command: bash
mongo:
image: mongo
environment:
MONGO_INITDB_DATABASE: my-database
When you run docker-compose up, Docker will create a default network and assigns the service name as hostname for both mongo and server.
You can now access the backend container via:
docker exec -it server bash
And now you can ping the mongo container using Dockers internal network (default on port 27017 in this case):
curl -v http://mongo:27017/my-database
That's it. The same applies for your setup.
I have had a similar example where I was working with separate docker-compose files working on a docker swarm with an overlay network to do that all I had to do is change the networks parameters as so:
first docker-compose.yaml
version: '3.9'
.
.
.
networks:
net:
driver: overlay
attachable: true
docker-compose -p app up
since I have specified the app name as app using -p the initial network will be app_net.
Now in order to run another docker-compose with multiple services that will use the same network you will need to set these as so:
second docker-compose.yaml
version: '3.9'
.
.
.
networks:
net-ref:
external: true
name: app_net
docker stack deploy -c docker-compose.yml mystack
No matter what name you give to the stack the network will not be affected and will always refer to the existing external network called app_net.
PS: It's important to make sure to check your docker-compose version.
version: '2'
services:
bot:
build: .
volumes:
- '.:/home/node'
- /home/node/node_modules
networks:
- my-rede
mem_limit: 100m
memswap_limit: 100m
cpu_quota: 25000
container_name: 236948199393329152_585042339404185600_bot
command: node index.js
environment:
NODE_ENV: production
networks:
my-rede:
external:
name: name_rede_externa
Follow up of JohnHarris answer, just adding some more details which may be useful to someone: Lets take two docker-compose file and connect them through networks:
1st foldername/docker-compose.yml:
version: '2'
services:
some-contr:
container_name: []
build: .
...
networks:
- somenet
ports:
- "8080:8080"
expose:
# Opens port 8080 on the container
- "8080"
environment:
PORT: 8080
tty: true
networks:
boomnet:
driver: bridge
2nd docker-compose.yml:
version: '2'
services:
pushapiserver:
container_name: [container_name]
build: .
command: "tail -f /dev/null"
volumes:
- ./:/[work_dir]
working_dir: /[work dir]
image: [name of image]
ports:
- "8060:8066"
environment:
PORT: 8066
tty: true
networks:
- foldername_somenet
networks:
foldername_somenet:
external: true
Now you can make api calls to one another services(b/w diff containers)like:
http://pushapiserver:8066/send_push call from some code in files for 1st docker-compose.yml
Two common mistakes (atleast i made them few times):
take note of [foldername] in which your docker-compose.yml file is present. Please see above in 2nd docker-compose.yml i have added foldername in network bc docker create network by [foldername]_[networkname]
Port: this one is very common. Please note i have used 8066 when trying to make connection i.e. http://pushapiserver:8066/... 8066 is port of docker container(2nd docker-compose.yml) so when trying to talk with different docker compose.
docker will use docker container port[8066] and not host machine mapped port
[8060]
Related
I'm trying to containerize my application. I use mongodb and 2 more micro services.
As you can see in the docker compose file below, I have some problems.
My requirements:
main_image needs to connect to MongoDB.
gui_image needs to connect to MongoDB.
gui_image needs to show its GUI on port 8080 (Can use another port as well)
gui_image has to read and write to a file inside my computer.
MongoDB has to access a volume inside my computer.
main_image needs access to the internet.
Here is my questions:
1- Does exposing ports in docker file and docker-compose the same thing?
2- How do I mount a volume to mongodb as best practice?
3- How to accomplish the requirements above with the diagram below in docker-compose?
Here is my docker-compose file:
version: "3"
services:
mongo:
image: mongo:latest
ports:
- 27017:27017
main_image:
build:
context: .
dockerfile: .\my_project\dockerfile
depends_on:
- mongo
gui_image:
build:
context: .
dockerfile: .\my_gui\dockerfile
ports:
- 8080:8080
- 27017:27017
depends_on:
- mongo
Here is my dockerfile under my_gui directory:
FROM continuumio/miniconda3
WORKDIR /app
COPY . .
RUN pip install dash
EXPOSE 8080
EXPOSE 27017
ENTRYPOINT [ "python","gui_script.py"]
And lastly, here is my dockerfile under my_project directory:
FROM continuumio/miniconda3
WORKDIR /app
COPY . .
EXPOSE 27017
ENTRYPOINT [ "python","main_script.py"]
1.The EXPOSE instruction in Dockerfile informs Docker that the container listens on the specified network ports at runtime(like when using docker run -p command).
However using ports in compose is a dynamic way of specifying these ports. So images like nginx or apache which are always supposed to run on port 80 inside the container will use EXPOSE in Dockerfile itself.
While an image which has dynamic port which may be controlled using an environment variable will then use expose in docker run or compose file.
some_webapi:
environment:
- ASPNETCORE_URLS=http://*:80
build:
context: .
dockerfile: ./Dockerfile
2.As documented on the docker hub page for mongo image (https://hub.docker.com/_/mongo/) you can use
volumes:
- '/path/to/your/pc/folder:/path/inside/docker'
3.And for the last question you might wanna use Networking in Compose.
By default Compose sets up a single network for your app. Each container for a service joins the default network and is both reachable by other containers on that network, and discoverable by them at a hostname identical to the container name.
Services can join networks like this
gui_image:
build:
context: .
dockerfile: .\my_gui\dockerfile
ports:
- 8080:8080
- 27017:27017
depends_on:
- mongo
networks:
- gui
And also you have to define all the networks used by services in global scope of compose file
version: '3'
services:
networks:
gui:
After that containers will be able to see each other even by their container_name which you can define in services
mongo:
image: mongo:latest
ports:
- 27017:27017
container_name: gui_mogno
then you will be able to connect to mongo with a connection string like this mongodb://gui_mogno:27017/
You can get more information about networking here
I have overlapping with default docker subnets, so I want to setup for some services custom network. docker-compose version:3
example of docker-compose.yml
version: '3'
services:
service1:
build:
dockerfile: Dockerfile
service2:
build:
dockerfile: Dockerfile1
service3:
build:
dockerfile: Dockerfile2
networks:
- net1
networks:
net1:
ipam:
driver: default
config:
- subnet: "172.101.0.0/24"
So when I execute the following command I got 2 networks :
docker-compose up --build -d service3
project_default and project_net1
But I need default network for other services so I can't change default settings.
What should I do to not create default network if it's not used for the service?
If you don't have a networks: specification for a given service, Docker Compose behaves as though you specified networks: [default]. In your example, service1 and service2 are on the default network and service3 is on the net1 network.
If you really need Docker Compose to not create the project_default network then you need to assign every container to some other network. From experimenting with a minimal docker-compose.yml file, explicitly adding networks: [net1] to the two services that don't already have it will cause the default network to not be created.
If your real issue is just around an IP address conflict, you're allowed to manually configure the default network and this might be easier.
version: '3'
services:
service1:
build: .
# with no networks:, so it gets the default
networks:
default:
ipam:
driver: default
config:
- subnet: "172.101.0.0/24"
My docker services require a network to work properly. I intend to share my compose file, but I don't want to manually create the network with the right parameters each time.
I would like the network to be created during the build process, in the Dockerfile or in the docker-compose file.
I don't think you can create a network in your Dockerfile, but you can do so in your docker-compose.yml file.
Have you looked at networking in Docker Compose?
You can specify your own networks with the networks key in your docker-compose.yml file. This lets you create custom networks and specify custom network drivers and options.
Each service in your docker-compose.yml file can specify what networks to connect to, with the service level networks key.
Here is an example from the Compose docs
version: "3"
services:
proxy:
build: ./proxy
networks:
- frontend
app:
build: ./app
networks:
- frontend
- backend
db:
image: postgres
networks:
- backend
networks:
frontend:
# Use a custom driver
driver: custom-driver-1
backend:
# Use a custom driver which takes special options
driver: custom-driver-2
driver_opts:
foo: "1"
bar: "2"
I am trying to have my own network name for my docker-compose files (server.yml and test.yml), as test.yml gets only started from time to time, but needs access to some services in the server.yml. I can make it work with docker-compose -p nameofproject up, but not with COMPOSE_PROJECT_NAME.
server.yml
version: '2'
networks:
mynetwork:
driver: bridge
services:
app1:
networks:
- mynetwork
environment:
POSTGRES_PASSWORD: somepassword
COMPOSE_PROJECT_NAME: serverstack
app2:
networks:
- mynetwork
environment:
COMPOSE_PROJECT_NAME: serverstack
depends_on:
- app1
My expectation is that when the container is starting I should see
Creating serverstackmynetwork_app_1
Creating serverstackmynetwork_app_2
the network should be named (docker network ls)
serverstack_mynetwork
just like when I do the following, which actually works
docker-compose -p serverstack up
And then I can connect just by using docker-compose up with the second file (which works just fine when using the -p option on the server.yml)
testing.yml
version: '2'
networks:
testapp_network:
external:
name: serverstack_mynetwork
services:
testapp:
networks:
- testapp_network
But using it without -p serverstack on the server.yml I see directories as names
Creating directoryofapp1_app1_1
Creating directoryofapp2_app2_1
so COMPOSE_PROJECT_NAME is being omitted and I also cannot connect to the server service though serverstack_mynetwork
I did add the COMPOSE_PROJECT_NAME: serverstack after building the image, but I would expect it should work anyhow. What am I missing?
I solved this by creating the ".env" file containing
COMPOSE_PROJECT_NAME=myprojectname
I have such docker-compose.yml (not a full list here):
version: '2'
services:
nginx:
build: ./nginx/
ports:
- 8080:80
links:
- php
volumes_from:
- app
networks:
app_subnet:
ipv4_address: 172.16.1.3
php:
build: ./php/
expose:
- 9000
volumes_from:
- app
networks:
app_subnet:
ipv4_address: 172.16.1.4
networks:
app_subnet:
driver: bridge
ipam:
config:
- subnet: 172.16.1.0/24
gateway: 172.16.1.1
After docker-compose up I got such an error:
User specified IP address is supported only when connecting to
networks with user configured subnets
So I'm creating subnet with docker network create --gateway 172.16.1.1 --subnet 172.16.1.0/24 app_subnet
But this doesn't solve the problem because docker-compose creates the subnet with name dev_app_subnet on the fly. And my subnet is not used - I'm getting the same error.
The main purpose of doing this is to assign static IP for nginx service - open my project web url from etc/hosts file.
[SOLVED] Found the solution. When pointing to the network, we should use flag "external" telling compose that network is already created and should be taken outside (otherwise it will be created on the fly with project prefix):
networks:
app_subnet:
external: true
So, after that docker-compose will attach containers to existing app_subnet
Before that the subnet must be created:
docker network create --gateway 172.16.1.1 --subnet 172.16.1.0/24 app_subnet
In my case is i first run docker-compose up failed, but the network already created, can see using docker network ls.
In this case, just docker-compose down, fix the yml, rerun docker-compose up is fine
It is probable that from previous run of the script the network interface was already created but without the subnet parameter.
For fixing it run
docker network ls -a
and remove the network that is blocking the creation of the service
docker network rm <network interface id>
Adding to the Qiushi's answer, you can use the docker-compose 3.9 version to specify the external network as below.
version: "3.9"
networks:
network1:
external: true
name: etl_subnet
Ref: https://docs.docker.com/compose/compose-file/compose-file-v3/#network-configuration-reference
To be specific if you're using docker stack deploy to deploy a swarm cluster. And you need to specify the scope of the subnet as 'swarm':
docker network create --gateway 172.16.1.1 --subnet 172.16.1.0/24
--scope swarm app_subnet
in the docker-compose.yml, you need to specify the external network as:
networks: default:
external:
name: etl_subnet
Use it as default.