SSH does not work

[olevski]

A user reported this.

Trying to use ssh or git clone with ssh fails:

 workspace ❯ source ▶ $ ▶ git clone git@github.com:SwissDataScienceCenter/renku-data-services.git
Cloning into 'renku-data-services'...
OpenSSL version mismatch. Built against 30000020, you have 30500040
fatal: Could not read from remote repository.

Example image: harbor.renkulab.io/renku-build/renku-build:renku-01k71dt4heexs2wfmt0a6b2np3

To reproduce:

docker run -ti --rm --entrypoint launcher harbor.renkulab.io/renku-build/renku-build:renku-01k71dt4heexs2wfmt0a6b2np3 bash

Then run

git clone git@github.com:SwissDataScienceCenter/renku-data-services.git

If the image is run without the entrypoint set to launcher then ssh works:

docker run -ti --rm --entrypoint bash harbor.renkulab.io/renku-build/renku-build:renku-01k71dt4heexs2wfmt0a6b2np3

[olevski]

Information about the image and versions:

Run Images:
  ghcr.io/swissdatasciencecenter/renku-frontend-buildpacks/base-image:0.0.8

Rebasable: false

Buildpacks:
  ID                                        VERSION        HOMEPAGE
  renku/conda-nodefaults                    0.0.8          -
  paketo-buildpacks/ca-certificates         3.9.1          https://github.com/paketo-buildpacks/ca-certificates
  paketo-buildpacks/miniconda               0.10.4         -
  paketo-buildpacks/conda-env-update        0.7.14         https://github.com/paketo-buildpacks/conda-env-update
  paketo-buildpacks/python-start            0.14.19        https://github.com/paketo-buildpacks/python-start
  renku/python-dependency-manager           0.0.8          -
  vscodium                                  0.3.0          -

[leafty]

I suspect different installed package versions between the build image and the run image; or between the layer and the run image.

[SalimKayal]

It will probably get fixed when I publish my changes for multi-architecture.

[kirschnj]

The problem here is that the conda environment shadows the system libraries, and we are invoking conda's ssh. It works after running unset LD_LIBRARY_PATH. Credits to ChatGPT for explaining it to me.

[leafty]

See: #166. It looks like conda tends to break the system by adding things to $LD_LIBRARY_PATH. I wonder if we can check if the issue exists when we have it (e.g. with the jupyter frontend) and when we don't.