I have successfully created a docker image for my demo application based
on the official karaf 4.2.8 docker hub image.
Here's what I did:
1. Copied etc/org.ops4j.pax.url.mvn.cfg from a karaf 4.2.8 instance,
and added my own maven repo
https://gist.github.com/steinarb/62a070f6481d3e84c334a78f1c771e80#file-org-ops4j-pax-url-mvn-cfg-L113
2. Copied org.apache.karaf.features.cfg and added the feature repo of
the sample application
https://gist.github.com/steinarb/fec331bcd37a0a7bdba850d33fcd0555#file-org-apache-karaf-features-cfg-L28
and added the sample application to the list of boot features
https://gist.github.com/steinarb/fec331bcd37a0a7bdba850d33fcd0555#file-org-apache-karaf-features-cfg-L52
3. Created a Dockerfile adding to the official docker hub karaf 4.2.8
image, copying in the two modified config files
https://gist.github.com/steinarb/444ad82a63dcf156101a064666ed590c#file-dockerfile-L1
4. Built the image from the Dockerfile
docker build -t steinarb/ukelonn-demo:v2 .
5. Started the image
docker run -p 8101:8101 -p 8181:8181 -d steinarb/ukelonn-demo:v2
6. The container took a while to start but eventually I could connect
to http://localhost:8181/ukelonn where I found the demo version of
my application running
The downsides of this approach, are:
1. Container startup is slow since features and their dependencies must
be downloaded
2. The maven repons may not be reachable at startup time and container
startup will fail
3. What the maven repos serve may potensially not be the same on each
startup
All of the above is, I guess, sort of contrary to docker philosophy,
which is to pack up everything the application needs to start, so that
the application can be started in a predictable and reproducable manner.
However, for this particular application, this is actually what I want:
I can create a docker image and deploy to docker hub, and on any docker
installation that can pull from dockerhub and can reach internet, I can
spin up the latest snapshot with the two commands "docker pull" and
"docker run".
