Differences

This shows you the differences between two versions of the page.

--- institute_lorentz:institutelorentz_maris_slurm_jupyterhub [2017/03/07 09:51] – [Access] lenocil
+++ institute_lorentz:institutelorentz_maris_slurm_jupyterhub [2020/01/15 13:34] (current) – removed lenocil
@@ Line 1: / Line 1: @@
-====== JupyterHub with Slurm on Maris ======
-:!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!:
-**As of February 2017 the maris' jupyterhub python environment is located at /marisdata/MARISHUB**. If you want
-commands such as ipcontroller or ipyengine to work within the maris jupyterhub setup you must include it in your paths. See the examples below.
-For instance in bash:
-<code>
-export PATH=/marisdata/MARISHUB/bin:$PATH
-</code>
-:!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!::!:
-JupyterHub is a multi-user server for Jupyter (AKA IPython) notebooks. Though JupyterHub you can create and share documents that contain live code, equations, visualizations and text notes (comments).
-Maris cluster provides a JupyterHub service which uses dedicated ''slurm'' partitions accessible only by
-authorized users. Users should submit access requests via email or preferably in person to ''support''.
-JupyterHub spawns single-user notebooks by requesting consumable resources to ''slurm'' using precoded parameters in order to prevent erroneous usages. Resources are allocated upon each successful login via the JupyterHub web interface. No notebooks will be spawned in the case of insufficient consumable resources (Memory and CPU).
-===== Access =====
-Maris' JupyterHub facilities can be reached (N.B.: **not** from home) via a web browser at
-https://jupyterhub.lorentz.leidenuniv.nl/
-{{ :institute_lorentz:jupyterhub.png?800 |}}
-:!: Please note that maris' jupyterhub has **admin_access enabled**. Admin users  have the ability to take actions on users' behalf, such as stopping and restarting their servers.
-===== Launching Notebooks =====
-Currently all users that have access to maris will also be able to use JupyterHub provided they have been allowed to use slurm. Upon a successful login, a dropdown menu lets users choose what type of notebook profile to spawn among a few predefined ones:
-{{ :institute_lorentz:spawn.png?1200 |}}
-If the requested  resources as defined by the profile chosen are available a single-user notebook instance will be launched enabling files management, terminals, different kerneks, etc... via a simple web interface.
-{{ :institute_lorentz:j1.png?1200 |}}
-:!: ''slurm'' will write output and error files relative to the spawning of each notebook instance using the filename scheme ''${HOME}/jupyterhub_%u_%j.[log|err]'', where `HOME' refers to your home directory and %u and %j to your username and slurm job number respectively. If launching a notebook fails, please read these output files before contacting ''support''.
-===== Jupyter Enabled Extensions =====
-==== JupyterHub `Files' Tab ====
-The //Files// tab which is the default tab on which a successful login will be redirected allows users to create, modify and remove files from they home directories. Furthermore, it allows to initiate a terminal application and a notebook. Currently, only notebooks which use the Python3 kernel can be instantiated.
-==== JupyterHub `Running' Tab ====
-This tab will show any running Jupyter processes, such as notebooks and terminals.
-==== JupyterHub `IPython Clusters' Tab ====
-JupyterHub users can distribute the load of their calculations to different maris' nodes using ipyparallel. It is highly suggested that users read  http://ipyparallel.readthedocs.io/en/latest/intro.html before proceeding.
-In this section we will only describe how to configure an IPython parallel computing cluster given the current ''slurm'' setup on Maris. For further info please read https://ipython.org/ipython-doc/3/parallel/.
-:!: If you are non familiar with slurm, please consult [[:slurm_tutorial|Slurm: A quick start tutorial]], [[institute_lorentz:institutelorentz_maris_slurm|slurm on maris]] and http://slurm.schedmd.com/ before proceeding.
-:!: IPython parallel clusters configured without ''slurm'' will be terminated abruptly and automatically by the system. This might bring your JupyterHub notebook instance to a freeze state which can be cleared at any time through ''slurm''
-<code>
-scancel <job ID>
-</code>
-where job ID is the job number relative to the single notebook application.
-=== Examples: IPython parallel cluster profile ===
-All IPython profiles are stored in ${HOME}/.ipython unless specified differently. You can do so by overriding the environment variable IPYTHONDIR.
-First of all, create a new parallel profile. Open a terminal and type
-  ipython profile create dummy --parallel
-This will create a new directory in your .ipython directory called `profile_dummy'
-  ls .ipython
-  profile_dummy
-This directory (profile) will contain several files that can be customized to your needs. More important, you must customize its files in order to let ''slurm'' manage and use your parallel profiles via JupyterHub's interface.
-Firstly, modify the file `ipcontroller_config.py' and make sure that the following lines are present
-   c.HubFactory.client_ip = '*'
-   c.HubFactory.engine_ip = '*'
-Secondly, edit the file `ipcluster_config.py' to force ipyparallel use slurm, for instance
-<code>
-c.IPClusterEngines.engine_launcher_class = 'SlurmEngineSetLauncher'
-c.IPClusterEngines.n = 2
-c.IPClusterStart.controller_launcher_class = 'SlurmControllerLauncher'
-c.SlurmLauncher.queue = u'weak-computation'
-c.SlurmEngineSetLauncher.batch_template ='''#!/usr/bin/bash
-#SBATCH -A your-account-name
-#SBATCH -p {queue}
-#SBATCH -n {n}
-#SBATCH -D /marisdata/your-username/somewhere
-#SBATCH -o slurm_%u_%N_%j.out # stdout
-#SBATCH -e slurm_%u_%N_%j.err # stderr
-export PATH=/marisdata/MARISHUB/bin:$PATH
-srun ipengine --profile-dir={profile_dir}
-'''
-c.SlurmControllerLauncher.batch_template ='''#!/usr/bin/bash
-#SBATCH -A your-account-name
-#SBATCH -p {queue}
-#SBATCH -c 1
-#SBATCH -N 1
-#SBATCH -D /marisdata/your-username/somewhere
-#SBATCH -o slurm_%u_%N_%j.out # stdout
-#SBATCH -e slurm_%u_%N_%j.err # stderr
-export PATH=/marisdata/MARISHUB/bin:$PATH
-ipcontroller --profile-dir={profile_dir}
-'''
-</code>
-:!: Read the slurm manuals and choose the $SBATCH options appropriately for your needs.
-In this example we note:
-  * We must modify the variable PATH to include **/marisdata/MARISHUB/bin**, that is export PATH=/marisdata/MARISHUB/bin:$PATH.
-  * The number of engines can be set using either the web interface via the `Tab Ipython Clusters' or by setting `c.IPClusterEngines.n = #'.
-  * An appropriate slurm account must be charged for things to work.
-  * The slurm partition to which to submit your calculations can be set using `c.SlurmLauncher.queue'.
-== Start the Profile ==
-Navigate to the `IPython Cluster' tab, select a profile, the number of engines and push `start'. Happy programming.
-{{ :institute_lorentz:j2.png?nolink |}}
-==== JupyterHub `Conda' Tab ====
-This tab allows users to list and manage conda environments using jupyter's web interface. In maris' setup, conda environments are displayed if they are in the following paths:
-<code>
-export CONDA_ENVS_PATH=/marisdata/$USER/.conda/envs:/home/$USER/.cond/envs${CONDA_ENVS_PATH:+:${CONDA_ENVS_PATH}}
-</code>
-{{ :institute_lorentz:j3.png |}}
-:!: The jupyterhub environment uses  nb_conda v2.0.0. This version is known to have several bugs/limitations. Among which we cite the impossibility of creating conda environments through the web interface. This limitation can be overcome taking some __risks__ (**ONLY** if you know what you are doing) by editing your jupyter configuration file usually located at ''~/.jupyter/jupyter_notebook_config.py'' adding the line
-<code>
-c.NotebookApp.disable_check_xsrf = True
-</code>
-===== Custom Jupyter Kernels =====
-To install a custom python kernel you can use ipython.
-<code>
-ipython kernel install --name=my-python-kernel --user
-</code>
-Note that in the command above the python version of the kernel will be the same as the python version that ipython uses. Also, the option `--user' makes sure that the kernel is installed in your local kernel directory which can vary from system to system. In most GNU/Linux distributions these are located at
-<code>
-${HOME}/.ipython/kernels
-${HOME}/.local/share/jupyter/kernels/
-</code>
-In any cases, to see where your local installation of ipython would look for installed  kernels and/or to display the specs of the existing ones, type
-<code>
-python kernelspec list
-[TerminalIPythonApp] WARNING | Subcommand `ipython kernelspec` is deprecated and will be removed in future versions.
-[TerminalIPythonApp] WARNING | You likely want to use `jupyter kernelspec` in the future
-Available kernels:
-  py2                     /home/xxx/.local/share/jupyter/kernels/py2
-  py2 test1               /home/xxx/.local/share/jupyter/kernels/py2 test1
-  python3.6-2017-02-01    /home/xxx/.local/share/jupyter/kernels/python3.6-2017-02-01
-  python3                 /marisdata/MARISHUB/share/jupyter/kernels/python3
-</code>
-Jupyter will display all conda kernels and all installed kernels you might have in the `new notebook' tab:
-{{ :institute_lorentz:j4.png |}}

Computer Documentation Wiki

User Tools

Site Tools

Differences

Page Tools