We want to switch to a new Jenkins (ver. 2.176.1), where the slaves are started in a docker cloud on demand (using the docker plugin).
How to start an agent in the cloud, with a specific port mapping, which does not collide with other containers in the same cloud, but can be further used in the pipeline script?
Old setup
Our current jenkins does not use docker in any way, the nodes are always running. The usual build process for a web project uses maven. At some point, the application is started using the maven cargo plugin. Selenium tests are executed, using a selenium grid. The external ports for the running web project are configured on each jenkins slave.
Goal
Run this setup with on demand docker containers as slaves, still using the external tools.
Problem
The basic build of a test project works and the problem is with the selenium part.
Using one port mapping works for one container, of course there will be collisions if we run more at the same time.
First we tried to use a port range in the global docker agent template, from the docker plugin. This allows to start multiple containers, but we found no parameter with the actually used port in the pipeline scripts, so no way to set it for the tests.
Further tries included agent{ docker{ image 'my_image' args '-p...'} } or the "sidecar" approach from here https://jenkins.io/doc/book/pipeline/docker/ and setting the ports when the container is started, using the EXECUTOR_NUMBER parameter to make the ports unique. In both cases, the jenkins tries to start another container in the agent container. This is too late, as the mapped ports of the agent container can't be changed after the container was created. And n
Using something like docker inspect from withing a running slave failed, as we don't know the current container id either. Update see below
So how can we start a slave, map a know set of docker internal ports to a set of ports on the host, without colliding with other docker agents and still know which ports are used in the build scripts, aka. jenkins files?
Update/Workaround
First, it is possible to get the ID of the container, using the environment variable DOCKER_CONTAINER_ID. Another approach is the hostname of the current node, as this is also the container id and can be resolved in the scripts.
The resulting line looks like this:
HTTP_PORT = (sh(script: 'docker -H tcp://${BIDGE_IP}:4243 inspect --format=\'{{(index (index .NetworkSettings.Ports \"8080/tcp\") 0).HostPort}}\' `hostname` ', returnStdout: true)).trim()
The variable ${BRIDGE_IP} is defined in the jenkins global variables and is the docker network ip of the host, where the docker engine is running on.
