bsub jupyter lab

start jupyter lab (python and R) server on LSF farm5 node, i.e. notebooks with access to /lustre, via bsub.
This page has several sections:

• How to start jupyter lab
• How to use your own R libraries
• How to install your own R libraries in your project or home dir
• How to use your own python libraries
• How to set up a personal jupyter lab conda environment
• How to set up a personal jupyter lab conda environment with your own R kernel
• How to copy data to/from personal computer to Sanger /lustre without VPN
• How to activate hgi conda on farm5

instructions

ssh on farm5, clone repo and run start script with arguments:
! This requires conda to be activated, see instructions below if not done already !

git clone https://github.com/wtsi-hgi/bsub_jupyter_lab.git && cd bsub_jupyter_lab
./bsub_jupyter_lab.sh -g hgi -c 4 -m 50000 -q normal

• -g is your LSF group (not hgi)
• -c is the number of CPUs requested
• -m is the memory requested (50000 means 50G)
• -q is the LSF queue

there are also 2 optional arguments

• -d /lustre/path_to_notebook_dir (to set notebook's browser directory)
• -e /lustre/path_to_conda_env (to use your own jupyter lab conda env, see instructions below to create)

The address:port and token of the server will be given in file jupyter_lab.log (created in current directory). That is, wait until you see a line like the following in jupyter_lab.log , and paste address in your browser:

    Or copy and paste one of these URLs:
        http://node-10-4-1:53074/?token=ea9bba78eb0840154b45acfc90dc9395e66c8d6fbcb2d4be

If you are working remotely, you need to be connected via VPN or use SSH tunneling through the web proxy to access the node from your web browser.
With the latter, you only need to forward the port of the host that is running Jupyter Lab on to a port on your local machine: To do so,

1. In a terminal, run the Tunnelling SSH command that you will find in log file jupyter_lab.log (for the example above, that would be ssh -L 53074:node-10-4-1.internal.sanger.ac.uk:53074 [email protected],
   which forwards port 53074 on node-10-4-1 to port 53074 on your localhost, jumping through the SSH gateway.)
2. Point your browser to the address 127.0.0.1 that is also shown in jupyter_lab.log (for example, http://127.0.0.1:53074/?token=ea9bba78eb0840154b45acfc90dc9395e66c8d6fbcb2d4be).

R libraries

2 R versions are currently available in the notebook: R 3.6.1 and R 4.0.0 .
You can add your own installed libraries (must be compatible with either R 3.6.1 or R 4.0.0) in an R notebook:

.libPaths(/lustre/path_to_installed_libraries)
library(your_library_name)

How to install your own libraries in your project or home dir

# set local/personal libs directory, e.g.:
R_version = paste0(as.character(R.version['major']),
       '.', as.character(R.version['minor']))
local_libs = paste0('~/personal_rlibs_R_',R_version) # can be changed to any new dir in your home dir or project area

# create directory if doesn't exist already:
dir.create(local_libs, showWarnings = FALSE)
# tell R to use this libs directory
.libPaths(local_libs)
# Install BiocManager if not done already:
if (!requireNamespace("BiocManager", quietly = TRUE))
    install.packages("BiocManager", lib=local_libs, repos='http://cran.us.r-project.org')
BiocManager::install(lib=local_libs)

# now R is ready to install packages in local_libs
# example install of a package in local_libs directory via Bioconductor:
BiocManager::install('purrr', lib=local_libs)

# example install of a package in local_libs directory without Bioconductor:
install.packages("anocva", lib=local_libs, repos='http://cran.us.r-project.org')

# test that libraries are there and loaded correctly
library(purrr)
library(anocva)
# check that local_libs directory indeed has `purrr` and `anocva` dirs:
list.files(local_libs)

# in future sessions, don't forget to run .libPaths(local_libs) so that R can load these libraries.

python libraries

In Jupyter, you can open a terminal and try install packages with a --target directory, e.g.
pip install pandas --target /lustre/path_to_new_pip_libraries
and then in a python notebook:

    import sys
    sys.path.append('/lustre/path_to_new_pip_libraries')
    import pandas

If that gives conflict errors with the packages already installed, you can create and use your own jupyter-lab/python conda environment, see instructions below.

How to copy data to or from personal computer to Sanger lustre without VPN

This, and all SSH related issues, are documented at https://ssg-confluence.internal.sanger.ac.uk/display/FARM/All+things+SSH .
In short, you can use rsync.
First, in a terminal, run:

ssh -L2222:farm5-login.internal.sanger.ac.uk:22 [email protected] 

where your_username is your Sanger username. It should ask for your Sanger ssh password and ask where you want to go (enter farm5-login). Once connected, keep that terminal open, and in a new one run the rsync command to copy your files:

rsync -rvz -e 'ssh -p 2222' --progress local_file_to_copy your_username@localhost:/lustre/path_where_you_want_the_files_to_go/

jupyterlab conda env

You can also use your own conda environment, so that you can install any conda/pip/R packages directly in the environment. Create a new minimal environment with (or use minimal_conda_env.txt) :

conda create --prefix /lustre/path_to_your_new_env/jupyterlab_env -c conda-forge jupyterlab

and use optional argument -eof script ./bsub_jupyter_lab.sh to reference it.

jupyterlab conda env R kernel

You can also use your own conda environment + your own R kernel (that will use R from the new conda env):

 
# full setup of your own Jupiter lab + R kernel environment:
 
# choose a path you have access to (/lustre/somewhere or preferably /software/somewhere if you can):
export INSTALL_DIR=/lustre/scratch118/humgen/hgi/users/mercury/test
 
# first, clone jupyter repo in INSTALL_DIR (this is important later on to setup the new R kernel config):
cd $INSTALL_DIR
git clone https://github.com/wtsi-hgi/bsub_jupyter_lab.git
 
# now, install your own jupyterlab conda environment, also in INSTALL_DIR:
conda create --prefix $INSTALL_DIR/jupyterlab_env -c conda-forge jupyterlab
 
# check that environment can be loaded ok:
conda activate $INSTALL_DIR/jupyterlab_env
 
#  now, you need to add your own R jupyter kernel in that new conda env:
# first, add R to the conda env:
conda install -c conda-forge r-base # this should instal R version 4 (let me know if you wanted a different version here).
# next add R kernel and tell jupyter lab configuration about it:
# check which R will be used by default
which R # in my case it will say “/lustre/scratch118/humgen/hgi/users/mercury/test/jupyterlab_env/bin/R” which is good because I want it to be in from my INSTALL_DIR
# go into R
R
# check that R default library path is in INSTALL_DIR
.libPaths() # in my case it will say [1] "/lustre/scratch118/humgen/hgi/users/mercury/test/jupyterlab_env/lib/R/library"
# now install the IRkernel package in that default location
install.packages('IRkernel')
IRkernel::installspec(displayname = 'Rperso')  # to register the kernel in the current R installation
# That command will install the R kernel in your home dir
# in my case, it said [InstallKernelSpec] Installed kernelspec ir in /nfs/users/nfs_m/mercury/.local/share/jupyter/kernels/ir
 
# now exit R, and copy that kernel config from your home directory to the bsub_jupyter_lab directory you just cloned:
cp -r ~/.local/share/jupyter/kernels/ir $INSTALL_DIR/bsub_jupyter_lab/kernels/
 
# check that you can use your new conda env via jupyter notebook, and that it sees your personal R kernel:
cd bsub_jupyter_lab
./bsub_jupyter_lab.sh -g hgi -c 4 -m 50000 -q normal -e $INSTALL_DIR/jupyterlab_env
# notice I used the -e option, and please replace -g with our own farm5 LSF group (not ‘hgi’).
# open your browser to the jupyter lab URL, and check that you see R kernel 'Rperso'
 
# You should see the “Rperso” kernel available in the list of kernels, and R command .libPaths() should point to the R library path in your new conda env.
 
# now you can install any other library in your environment!
# also, if you need rstudio shortcuts, you will need: jupyter labextension install @techrah/text-shortcuts

activate hgi conda on farm5

ssh farm5-login
gn5@farm5-head2:~$ /software/hgi/installs/anaconda3/bin/conda init bash
<now you must log out and into farm5 again>
<check that conda activate works:>
gn5@farm5-head2:~$ conda activate hgi_base