Conda Solver Triage & Environment Stabilization

Pain point

Conda installs can hang at "Solving environment…" or fail with UnsatisfiableError, especially when

environments have accumulated mixed channels, overly broad version specs, or pip-installed packages.

Where it happens: computational research (local, VM, HPC) when bootstrapping or updating toolchains.

Repro/diagnostic steps

1. Capture: `conda info`, `conda config --show`, `conda list --explicit`, and the exact failing command.
2. Identify: channels + priority, Python version, and whether pip has been used inside the env.

Root causes (common)

• Channel mixing and inconsistent priority settings.
• Overly permissive or conflicting version constraints.
• Large, "kitchen sink" environments instead of task-specific envs.
• pip-installed packages complicating solver expectations.

Fix workflow

1. Create a *minimal* new env (pin Python + 1–2 core deps) and verify the solver is healthy.
2. If solves are slow, use mamba/libmamba as a drop-in solver and reduce the spec surface area.
3. Standardize channels (choose a primary ecosystem, avoid ad-hoc mixing).
4. Add dependencies incrementally (binary search conflicts).
5. Export a lock-style spec once stable; avoid piecemeal upgrades in production envs.

Expected result

• Environment solves complete consistently (minutes, not hours).
• Recreating the env from `environment.yml` yields the same working stack.

References

• https://docs.conda.io/docs/user-guide/tasks/manage-channels.html
• https://github.com/conda/conda/issues/8051
• https://bioconda.github.io/faqs.html
• https://stackoverflow.com/questions/74781771/how-we-can-resolve-solving-environment-failed-with-initial-frozen-solve-retry/76799565