[Cylc 8 - cylc-uiserver] Stuck at building dependencies when try to install cylc-uiserver via pip / conda

Cylc Cylc Support
1

Hi Cylc Team,

I’m trying to download all Python dependencies of cylc-uiserv from a machine with Internet connection for offline installation in a production environment without Internet connection.

First, tried ‘pip3 download’ with a bare Python 3.8 Env., like this

[jliu@asp2a-login-nea01 ~]$ pip3 list
Package    Version
---------- -------
pip        22.0.4
setuptools 56.0.0

pip3 download -vvv --no-binary=":all:" --progress-bar=on  --dest=$HOME/SRC/cylc-uiserver cylc-uiserver

After about 4 hours, stuck at building dependencies of jupyterlab-pygments.

Then, went for anaconda3/2021.11 with Python 3.9, like this

source  /home/project/17001770/app/anaconda3/2021.11/bin/activate

conda config --set unsatisfiable_hints True

export CONDA_PKGS_DIRS=/home/project/17001770/conda_pkgs/cylc-uiserver

conda install -vvv  --download-only -c conda-forge cylc-uiserver

It took more than 10 hours in examining conflict for 394 packages and downlaoded nothing in the spcified target dirctory. Wondering if missed something from the command “conda install --download-only” .

Wondering life woudl be easier if it could provide a full lists of Python dependencis of Cylc 8 ( inclusives cylc-uiserver, cylc-uiserver’[hub]') based on a bare Python 3.x Env. as an alternative workaround for off-line installation, such as output from pip3 freeze.

PS,

  1. pip donwload worked fine for cylc-flow, cylc-rose, metomi-rose.
  2. tried to upload the output, but it’s not allowed.

Thanks for your time

2

We recommend using mamba rather than conda (which is getting increasingly slow), see https://cylc.discourse.group/t/trouble-installing-cylc-8-1-0/584/2.

For installation on systems without internet access we use conda pack.

conda pack --dest-prefix=/path/to/cylc -o packed-env.tgz

The --dest-prefix option specifies the path to where cylc will be installed on the destination platform and avoids any need to “unpack” after untarring. https://conda.github.io/conda-pack/cli.html

We use conda pack for all our installations. On our build host we create 2 environments (using mamba), one containing the full installation and one containing the minimal environment needed on remote job platforms (cylc-flow-base + metomi-rose-base). We then pack all the required environments for the many platforms where we install Cylc.

3

For the record, the latest build of the Cylc documentation (which is currently only available via the “nightly” docs link) includes the following note:

Note

We recommend using the fast Mamba environment solver to install Cylc. Mamba can be used as a drop-in replacement for the conda command, or as a conda command plugin. The classic conda environment solver may be too slow for a complex package like Cylc.