dockerNode without remoting

Jenkinsfile.groovy

stage 'build' {

  withDocker(image: 'node_js:4.4',
             environment: [foo: 'bar'],
             volumes: [node_cache: '/node_modules'],
             commands: """
                npm install
                npm test
                ...
            """,
            stash: [jar: '/dist/app.tar.gz'])

get actually executed as :

# Create volume
id=sha(jobID+node_cache)
docker volume inspect $id
# if none exists
node_cache = ${docker volume create $id} 

# create container
container = $(docker create -v $node_cache:/node_modules node_js:4.4 /tmp/command.sh)

# create command script
echo $commands > docker cp - container:/tmp/command.sh

# run container
docker start -d container
# container is ran detached, so we don't need longRunningTask hack and status polling.

# wait for container to stop
# (a single process could listen to all docker 'stop|die' events then dispatch to running pipelines)
docker events --filter container=$container

# if jenkins has been restarted, pipeline resume would need to check container status
docker inspect container=$container

# pipe logs info web console for live view
docker logs -f $container

# stash
docker cp $container:/dist/app.tar.gz some_path_on_master


For actual usage in a pipeline, will also need some :

  def workspace = dockerVolume()
  withDocker(image: 'git'),
             volumes: [workspace: '/work'],
             workdir:'/work',
             commands: "git clone $scm")
             
could be made simpler : 

  (def workspace = ) dockerVolume($scm)
             
such a workspace volume could be implicitely bound to all subsequent containers in pipeline.