Thoughts on automated jupyter lab management script

(Thanks @sebastian-wahl for triggering this. I like the idea of having all these tasks in a single script!)

A few thoughts on the automation in scripts/run_chromium_through_ssh_tunnel.sh.

• Pass functions by SSH rather than using heredocs.

• Re-use discovering jupyter lab.

• Clean up chrome user dirs after shutdown. Use mktemp -d rather than explicit path.

• Specify envs rather than hard-code packages.

Pass functions by SSH rather than using heredocs.

According to https://stackoverflow.com/a/22107893 this

function get_hostname () { echo ${HOSTNAME}; }
ssh <remote_host> "$(typeset -f get_hostname); get_hostname;"

is equivalent to

ssh <remote_host> <<EOF
    echo \${HOSTNAME}
EOF

Re-use discovering jupyter lab

What about

function discover_jl_port () {
	page=$(${HOME}/miniconda3/bin/python \
            ${HOME}/miniconda3/bin/jupyter notebook list" \
            | grep http | awk '{print$1}')
	port=$(echo $page | awk -F: '{print$3}' | awk -F/ '{print$1}')
}

if [[ "$task" == "start" ]] || [[ "$task" == "connect" ]]; then

    port=$(ssh $hname "$(typeset -f discover_jl_port); discover_jl_port")
    if [[ -z "$port" ]] ; then
        …

elif [[ "$task" == "stop" ]]; then

    port=$(ssh $hname "$(typeset -f discover_jl_port); discover_jl_port")
    if [[ ! -z "$port" ]] ; then
        …

Clean up temporary chrome user dir

(Also an issue for the other script.)

tmp_user_data=$(mktemp -d)
chromium-browser --new-window \
    --user-data-dir="${tmp_user_data}" \
    --proxy-server="socks5://localhost:${socks_5_port}" $page
rm -rf ${tmp_user_data}

Specify envs rather than hard-code packages

Currently, ae3db2fb790b6bb7f8f47604d7bc4f49d3f8c609/scripts/remote_jupyter_manager.sh#L116ff. hard-codes one python env. This should be generalized to allow for easier specification of envs.

With this py3_std.yml,

name: py3_std
channels:
  - defaults
  - conda-forge
dependencies:
  - numpy
  - matplotlib
  - scipy
  - ipykernel
  - git
  - python=3
  - cdo
  - netcdf4
  - xarray
  - libiconv
  - cartopy

the following installs a py3_std env that is identical to the hard-coded one:

function _conda_create () {
    source ${HOME}/miniconda3_20180525/bin/activate base;
    conda env create -f /dev/stdin;
}

cat py3_std.yml | ssh $hname "$(typeset -f _conda_create); _conda_create;"