Conda 23.1 libmamba default solver transition breaking cross-environment reproducibility

Answer

Force every conda invocation to use the libmamba solver by setting solver: libmamba in a global or project‑specific .condarc and generating lock files with that solver explicitly.

Why this works

Conda 23.1 switched the default resolver to libmamba, which has stricter conflict detection and different version heuristics than the classic solver. If any stage (local workstation, CI runner, or container image) falls back to the classic solver, the same environment specification can resolve to a different set of packages, breaking reproducibility. By making libmamba the explicit, universal solver, the resolution algorithm is identical everywhere, and lock files produced under libmamba are portable because they are not tied to the deprecated classic solver.

Steps to enforce a single resolver

1. Set libmamba as the default solver globally (or per‑project):
   

   conda config --set solver libmamba --system   # global

   
   or, for a repository‑local setting:
   

   conda config --set solver libmamba --env   # writes to .condarc in the current environment

2. Commit the resulting .condarc (containing solver: libmamba) to version control so every developer, CI job, and Docker build reads the same file.
3. When generating lock files, explicitly request the libmamba solver:
   
   • conda-lock lock -f environment.yml --solver libmamba
   • pixi lock (pixi defaults to libmamba)
   • conda env export --from-history -n myenv > environment.yml (after ensuring the active environment uses libmamba)
4. In CI pipelines and Dockerfiles, load the same .condarc before any conda command:
   

   # Example Dockerfile snippet\nFROM continuumio/miniconda3\nCOPY .condarc /root/.condarc\nRUN conda config --show solver  # verification step\n# … proceed with environment creation\n

5. Optionally tighten channel handling to reduce variability:
   

   conda config --set channel_priority strict --system

Verification

• Run conda info --json and confirm \"solver\": \"libmamba\" appears on each machine or container.
• Create an environment from the same lock file on two systems (e.g., local Linux and a CI Docker image) and compare conda list --explicit outputs; identical package hashes indicate solver parity.
• Check that lock files were produced with the libmamba flag (or note that pixi’s default uses libmamba) to ensure they are solver‑agnostic.