Sporadic failures using mamba-version: "*"

[victorlin]

(originally from nextstrain/augur#1125)

Description

I'm aware that using mamba-version is an experimental feature, and some instability can be expected.

However, I've noticed that it's been failing more often than usual on at least two repos managed by my team @nextstrain.

The error is not very useful:

##[error]The process '/usr/share/miniconda/condabin/conda' failed with exit code 1

I've compared a failing run with a successful re-run of the same workflow. There are some notable differences in the output of conda config --show:

75d74
< allowlist_channels: []
110c109
< default_python: 3.10
---
> default_python: 3.9
123a123
> experimental_solver: classic
125d124
< fetch_threads: 5
137d135
< number_channel_notices: 5
165d162
< solver: classic
181a179
> whitelist_channels: []
197c195
< no change     /usr/share/miniconda/lib/python3.10/site-packages/xontrib/conda.xsh
---
> no change     /usr/share/miniconda/lib/python3.9/site-packages/xontrib/conda.xsh

I do not understand why there are differences in Python versions and other conda config between two runs of the same workflow.

Known Workarounds

1. Use Mambaforge

1. Remove mamba-version.
2. Add miniforge-variant: Mambaforge (and optionally miniforge-version).

Example: nextstrain/cli@4a76497

2. Use the new libmamba solver

Note that this workaround can introduce a separate issue when used on windows-2022: conda/conda#12161.

1. Remove mamba-version.

2. Run this after the setup-miniconda step:

   - run: |
       conda install -n base conda-libmamba-solver
       conda config --set solver libmamba

3. Replace mamba install with conda install.

Example: nextstrain/augur@344ed51

[joamatab]

observing the same today

https://github.com/gdsfactory/gdsfactory/actions/runs/3906659862/jobs/6675080754

[joshuailevy]

Also observing this. https://github.com/andersen-lab/Freyja/actions/runs/3907050506. Thanks @victorlin for posting the workaround.

[jongyoul]

I also found the same error on https://github.com/apache/zeppelin/actions/runs/3901061183/jobs/6665542729

[traversaro]

If you are not interested in using defaults, a possible strategy is to use mambaforge instead of miniconda, see for example:

• Fix Conda CI on Linux avoiding to use the defaults channel by installing Mambaforge robotology/robotology-superbuild#840
• Fix Conda CI on Linux avoiding to use the defaults channel robotology/robometry#141
• Fix conda-ci by using mambaforge directly robotology/yarp-devices-ros2#44
• Fix conda-forge GitHub Action CI by using mambaforge directly and by supporting graphviz >= 3 on Windows gazebosim/gazebo-classic#3287
• Fix conda-ci GitHub Action CI by using mambaforge directly robotology/idyntree-yarp-tools#42

[victorlin]

@traversaro thanks, I've updated the list of workarounds in the issue description and put that one first since it is the more popular one.

[victorlin]

If you are not interested in using defaults, a possible strategy is to use mambaforge instead of miniconda

Note that defaults is included in the generated .condarc regardless of miniforge-variant. To truly get away from it, you'll have to use nodefaults, --override-channels, or conda config --remove channels defaults. See #207 for more info.

[kthyng]

I have been having the following error with ubuntu-latest:

PackagesNotFoundError: The following packages are missing from the target environment:
    - defaults/linux-64::conda==22.11.1=py310h06a4308_4[md5=07e2e8a698cf1e7f580b9efcae72184e]

with this in my test.yaml

      - uses: conda-incubator/setup-miniconda@v2
        with:
          mamba-version: "*" # activate this to build with mamba.
          channels: conda-forge, defaults # These need to be specified to use mamba
          channel-priority: true
          environment-file: ci/environment-py${{ matrix.python-version }}.yml

          activate-environment: test_env_cmocean
          use-only-tar-bz2: true # IMPORTANT: This needs to be set for caching to work properly!

(example PR with errors earlier: matplotlib/cmocean#88)

and workaround 1 above worked for me!

[rosepearson]

This is affecting me too on https://github.com/rosepearson/GeoFabrics/actions/runs/3925803401/jobs/6710960393. I'll look at work around 1 thanks,

[padix-key]

I created a minimum working example for this issue using different OS and Python versions: https://github.com/padix-key/gha-mamba

In this setup this issue only affects Ubuntu builds: https://github.com/padix-key/gha-mamba/actions/runs/4044454081

[goanpeca]

Hi @padix-key thanks for the example!

[padix-key]

In an updated version of https://github.com/padix-key/gha-mamba (updated setup-miniconda and platform version), this issue did not appear anymore. Can this issue be closed?