Prerequisites | Installation | Configuration | Contributing | Test | License |

The JupyterHub spawner which utilizes Universal Containerizer Runtime (UCR) in Marathon so that you can use GPU.

The original idea of the Marathon integration is taken from https://github.com/vigsterkr/marathonspawner which uses Docker containerizer in Marathon.

JupyterHub 0.7 or above is required, which also means Python 3.3 or above.

Install ucrspawner to the system:

pip install ucrspawner

git clone https://github.com/dtaniwaki/ucrspawner
cd ucrspawner
python setup.py install

Tell JupyterHub to use UCRSpawner by adding the following line to your jupyterhub_config.py:

c.JupyterHub.spawner_class = 'ucrspawner.UCRSpawner'

Then, specify the endpoint of Marathon.

c.UCRSpawner.marathon_host = "MARATHON_HOST"

Users can use any Docker image of Jupyter docker stack with UCRSpawner.

Set the default app image if you want to change.

c.UCRSpawner.app_image = 'jupyterhub/singleuser'

Users can choose their own app image in the spawn option form as well.

Specify a user in Mesos slaves whose UID is the same as a user in your Docker image.

c.UCRSpawner.mesos_user = 'jovyan_in_mesos'

In the default user of docker stacks jovyan's UID is 1000, so you must have a user whose UID is 1000 in Mesos slaves as well. Otherwise, you need to change the UID of jovyan in Docker image instead.

You can also change the app ID prefix of Marathon.

c.UCRSpawner.app_prefix = 'jupyter'

While UCRSpawner has minimum computing resource limits, you can configure default and maximum resource limits.

c.UCRSpawner.cpu = 1
c.UCRSpawner.max_cpu = 4
c.UCRSpawner.mem = 256 # in MB
c.UCRSpawner.max_mem = 1024 # in MB
c.UCRSpawner.disk = 1000 # in MB
c.UCRSpawner.max_disk = 5000 # in MB
c.UCRSpawner.gpu = 0
c.UCRSpawner.max_gpu = 2

You can get available resources of meso slaves in your handlers, which inherit BaseHandler through this spawner.

self.get_current_user().spawner.get_mesos_slaves()

Specify Mesos master host in jupyterhub_config.py if the endpoint is different from the one which Marathon reports.

c.UCRSpawner.mesos_master_host = 'MESOS_MASTER_HOST'

python setup.py test

Or use tox to test this package in multiple python versions.

tox

You can launch JupyterHub and the dependencies with UCRSpawner by docker-compose.

docker-compose --project-name=ucrspawner up

Wait for a while until all the services up, then, you will get access to JupyterHub at http://localhost:8000, Marathon at http://localhost:8080 and Mesos Master at http://localhost:5050.

1. Fork it
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create new Pull Request

Copyright (c) 2018 Daisuke Taniwaki. See LICENSE for details.