JupyterHub on a cluster with local spawning and batch spawning

[Brian Fulton-Howard]

As I understand it, I need to use some combination of wrapspawner, localspawner, sudospawner and batchspawner. I want to run jupyterhub as an unprivileged user, but the admins are happy to edit the sudoers file to make it work. What I'd like to do is give users in my lab a choice to start on the LSF cluster (batchspawner) or locally (localspawner). Wrapspawner would provide the interface to make a selection.

Is this possible, and how would I go about doing it? I looked in the documentation for the various spawners and didn't find any detailed information.

The cluster supports PAM, so I don't need any special authenticator.

[Michael Milligan]

Hi Brian -

It is possible, and it sounds like you already mostly have the solution! The example config at the end of the batchspawner README should be roughly what you want, except that you will use the SudoSpawner instead of LocalProcessSpawner. Specifically, you need to install all of the spawners you are using alongside Jupyterhub (e.g. pip install wrapspawner, batchspawner, and sudospawner into the same virtualenv as jupyterhub) -- the ProfilesSpawner configuration will tell Jupyterhub which spawner to use in response to the user's menu selection.

Sorry about the confusing state of the documentation about how to do this. Batchspawner and wrapspawner were originally distributed in the same repository, and when we split them apart it sounds like we may not have included enough context on one side or the other. We're happy to get suggestions about how to improve the docs if you'd like to give us some feedback when you get a chance.

Regards,

Michael

--
You received this message because you are subscribed to the Google Groups "Project Jupyter" group.
To unsubscribe from this group and stop receiving emails from it, send an email to jupyter+unsubscribe@googlegroups.com.
To post to this group, send email to jup...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/jupyter/a3f21888-0495-4fe0-b598-ca7e226bd61f%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Brian Fulton-Howard]

Thanks!

I have a potential addition to my config file. I had to subclass the LsfSpawner class because it appears incomplete and doesn't have all of the options required for our cluster. Does this look right?

c.JupyterHub.spawner_class = 'wrapspawner.ProfilesSpawner'
c.Spawner.http_timeout = 120
#------------------------------------------------------------------------------
# BatchSpawnerBase configuration
#   Providing default values that we may omit in the profiles
#------------------------------------------------------------------------------
c.BatchSpawnerBase.req_runtime = '12:00'
#------------------------------------------------------------------------------
# ProfilesSpawner configuration
#------------------------------------------------------------------------------
# List of profiles to offer for selection. Signature is:
#   List(Tuple( Unicode, Unicode, Type(Spawner), Dict ))
# corresponding to profile display name, unique key, Spawner class,
# dictionary of spawner config options.
#
# The first three values will be exposed in the input_template as {display},
# {key}, and {type}
#

## Override LsfSpawner to be compatible with Minerva

class MinervaSpawner(batchspawner.LsfSpawner):
    batch_script = Unicode('''#!/bin/sh
#BSUB -R "span[hosts=1]"        # Only spawn job on one server
#BSUB -P {account}
#BSUB -q {queue}
#BSUB -m {partition}
#BSUB -R rusage[mem={memory}]
#BSUB -n {nprocs}
#BSUB -J spawner-jupyterhub
#BSUB -o {homedir}/.jupyterhub.lsf.out
#BSUB -e {homedir}/.jupyterhub.lsf.err''')

    req_partition = Unicode('', \
        help="Partition name to submit job to resource manager"
        ).tag(config=True)

    req_account = Unicode('', \
        help="Account name to submit job to resource manager"
        ).tag(config=True)

c.ProfilesSpawner.profiles = [
   ( "Local server", 'local', 'jupyterhub.spawner.SudoSpawner', {'ip':'0.0.0.0'} ),
   ('Bode - 2 cores, 4 GB, 8 hours', 'bode2c4g12h', 'MinervaSpawner',
    dict(req_nprocs='2', req_queue='premium', req_runtime='8:00', req_memory='4000',
         req_partition='bode', req_account='acc_LOAD')),
   ('Manda - 8 cores, 128 GB, 4 hours', 'manda128gb', 'MinervaSpawner',
    dict(req_nprocs='8', req_queue='premium', req_runtime='4:00', req_memory='16000',
         req_partition='bode', req_account='acc_LOAD')),
   ('Bode - 2 cores, 4 GB, 24 hours', 'mesabi2c4gb24h', 'batchspawner.TorqueSpawner',
    dict(req_nprocs='2', req_queue='premium', req_runtime='24:00', req_memory='4gb',
         req_partition='bode', req_account='acc_LOAD')),
   ]

On Thursday, March 15, 2018 at 6:47:28 PM UTC-4, Michael Milligan wrote:

Hi Brian -

It is possible, and it sounds like you already mostly have the solution! The example config at the end of the batchspawner README should be roughly what you want, except that you will use the SudoSpawner instead of LocalProcessSpawner. Specifically, you need to install all of the spawners you are using alongside Jupyterhub (e.g. pip install wrapspawner, batchspawner, and sudospawner into the same virtualenv as jupyterhub) -- the ProfilesSpawner configuration will tell Jupyterhub which spawner to use in response to the user's menu selection.

Sorry about the confusing state of the documentation about how to do this. Batchspawner and wrapspawner were originally distributed in the same repository, and when we split them apart it sounds like we may not have included enough context on one side or the other. We're happy to get suggestions about how to improve the docs if you'd like to give us some feedback when you get a chance.

Regards,

Michael

On Thu, Mar 15, 2018 at 4:58 PM, Brian Fulton-Howard <fult...@gmail.com> wrote:

As I understand it, I need to use some combination of wrapspawner, localspawner, sudospawner and batchspawner. I want to run jupyterhub as an unprivileged user, but the admins are happy to edit the sudoers file to make it work. What I'd like to do is give users in my lab a choice to start on the LSF cluster (batchspawner) or locally (localspawner). Wrapspawner would provide the interface to make a selection.

Is this possible, and how would I go about doing it? I looked in the documentation for the various spawners and didn't find any detailed information.

The cluster supports PAM, so I don't need any special authenticator.

--
You received this message because you are subscribed to the Google Groups "Project Jupyter" group.

To unsubscribe from this group and stop receiving emails from it, send an email to jupyter+u...@googlegroups.com.

[Michael Milligan]

Yes, in general that's the correct idiom. You probably need to explicitly import your base class though (i.e. add "import batchspawner.LsfSpawner" before you declare your subclass).

That said, both of those req_* options look like good candidates for adding to the base class, especially since req_partition already exists in at least one other batchspawner subclass.

To unsubscribe from this group and stop receiving emails from it, send an email to jupyter+unsubscribe@googlegroups.com.

To post to this group, send email to jup...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/jupyter/1f26020c-41cf-4dc4-84d7-41a879c68bea%40googlegroups.com.

[Brian Fulton-Howard]

Should I do a pull request?

[Michael Milligan]

Sure, I just opened an issue so I don't forget about it (https://github.com/jupyterhub/batchspawner/issues/62) so I'm happy to get a PR implementing it!

To unsubscribe from this group and stop receiving emails from it, send an email to jupyter+unsubscribe@googlegroups.com.

To post to this group, send email to jup...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/jupyter/49e1e9c4-32a6-4843-8738-3cc3c468305e%40googlegroups.com.