Let external container know about subnets created by multiple docker-compose projects - docker-compose

Question

I have a multi-client setup, each client having similar application stack. To simplify it a bit, let's say a typical tomcat web app working on a db.
Clients must be completely isolated from each other, each with their on application stack, including a standalone tomcat).
VirtualHost configurations on httpd map domains to the corresponding tomcat instance via AJP.
For example:
client1.example.com -> 172.18.0.2:8009
client2.example.com -> 172.19.0.2:8009
I found docker-compose very useful within each client. One of the things I like is that it even takes care of creating private subnets for each client and containers within that client don't even need to know other containers' IP address, but can use aliases that are automatically set up by the links configuration. Pretty neat.
httpd
clients
client1
docker-compose.yml
tomcat
db
client2
docker-compose.yml
tomcat
db
Now the problem. httpd (which I would also want to dockerize) needs to reach tomcat for all clients. However, it doesn't know their IP.
How can I realize this mapping in VirtualHost? I.e. how to solve the question marks below?
<VirtualHost *:443>
ServerName client1.example.com
..
ProxyPass / ajp://?????:8009/
..
</VirtualHost>
One thing I have tried is to have some global env file, with all subnets and IPs explicit, like:
CLIENT1_TOMCAT=172.18.0.2
CLIENT2_TOMCAT=172.19.0.2
and use this in both each client's yml and httpd yml.
That's pretty ugly though, because custom IP configuration within a yml is either for all or for no service. I was hoping I could set the one for tomcat and let the others automatic, but they end up overlapping. All the sudden, specifying IP configurations for all services and all clients, makes the whole thing a lot less elegant.
I also know it is possible to inspect a container's IP like this:
docker inspect -f '{{range .NetworkSettings.Networks}}{{.IPAddress}}{{end}}' client1_tomcat
However, this requires the container to exist already. Also, if I use this method then I have to modify httpd's configuration every time the container is recreated, as it may change IPs.
I guess that somehow I wold like httpd's yml to "include" the clients' ones, so that it can know about their subnets. Still, keeping clients' subnets separated.
Any idea how I should approach this?
UPDATE
working setup based on #tcnj's answer:
Content of httpd/docker-compose.yml:
version: '2'
services:
httpd:
image: httpd
ports:
- "80:80"
..
networks:
- client1_default
- client2_default
external_links:
- client1_tomcat_1
- client2_tomcat_1
networks:
client1_default:
external: true
client2_default:
external: true

Answer 1

The easiest way to solve your problem is to put httpd on all of the client networks, and then refer to the tomcat containers by their name.
Eg:
You have client1 with network client1_default, and with the container client1_tomcat.
You also have client2 with network client2_default, and with the container client2_tomcat.
If you're starting httpd from the command line then you can add --network client1_default --network client2_default as extra options to docker run before the image name. If you're using docker compose for httpd as well then you can use the following:
httpd:
...
networks:
- client1_default
- client2_default
...
In your httpd config you can then have:
<VirtualHost *:443>
ServerName client1.example.com
...
ProxyPass / ajp://client1_tomcat:8009/
...
</VirtualHost>
<VirtualHost *:443>
ServerName client2.example.com
...
ProxyPass / ajp://client2_tomcat:8009/
...
</VirtualHost>