-
Notifications
You must be signed in to change notification settings - Fork 604
Description
Summary
CRITICAL: All CI test jobs on the main branch (26.02) are failing during conda environment setup due to contradictory CUDA version constraints in the rapids-xgboost==26.2.* dependency chain. The package has conflicting requirements that make it unsolvable for any CUDA version:
- CUDA 12.x: Direct constraint requires CUDA ≥13
- CUDA 13.x: Transitive dependencies require CUDA <13
This is a critical packaging/build error that blocks all testing on the main branch.
Failing test/component: conda environment creation (ALL Python tests on ALL CUDA versions)
Failure observed in:
- CUDA 12.2.2: https://github.com/rapidsai/cuml/actions/runs/19538382314/job/55938025612
- CUDA 13.0.2: https://github.com/rapidsai/cuml/actions/runs/19538382314/job/55938023313
- All other CUDA versions affected
Environment
- OS: All (rockylinux8, ubuntu, etc.)
- Python version: All (3.10, 3.11, 3.12, 3.13)
- CUDA version: ALL (both 12.x and 13.x confirmed broken)
- GPU: All
- Dependencies: All dependency configurations (oldest-deps, latest-deps)
- Branch: main (26.02)
Test Details
- Test file: N/A (failure occurs during environment setup, before tests run)
- Test name: All tests on ALL CUDA versions
Scenario 1: CUDA 12.x Failure
Error message:
error libmamba Could not solve for environment specs
The following packages are incompatible
├─ cuda-version =12.2 * is requested and can be installed;
└─ rapids-xgboost =26.2,>=0.0.0a0 * is not installable because it requires
└─ cuda-version >=13,<14.0a0 *, which conflicts with any installable versions previously reported.
Full error context (job 55938025612, log lines 1204-1211):
2025-11-20T13:27:32.8283476Z error libmamba Could not solve for environment specs
2025-11-20T13:27:32.8285222Z The following packages are incompatible
2025-11-20T13:27:32.8287242Z ├─ cuda-version =12.2 * is requested and can be installed;
2025-11-20T13:27:32.8288803Z └─ rapids-xgboost =26.2,>=0.0.0a0 * is not installable because it requires
2025-11-20T13:27:32.8290535Z └─ cuda-version >=13,<14.0a0 *, which conflicts with any installable versions previously reported.
2025-11-20T13:27:32.8825107Z critical libmamba Could not solve for environment specs
2025-11-20T13:27:32.9079141Z [rapids-conda-retry] conda returned exit code: 1
Scenario 2: CUDA 13.x Failure
Error message:
error libmamba Could not solve for environment specs
The following packages are incompatible
├─ cuda-version =13.0 * is requested and can be installed;
└─ rapids-xgboost =26.2,>=0.0.0a0 * is not installable because it requires
└─ libxgboost =3.0.5 rapidsai_h* but there are no viable options
└─ libxgboost 3.0.5 would require
└─ librmm >=25.12.0a8,<25.13.0a0 * but there are no viable options
└─ librmm [versions] would require
└─ cuda-version >=12,<13.0a0 *, which conflicts with any installable versions previously reported.
Full error context (job 55938023313, log lines 1220-1268):
2025-11-20T13:27:19.6040868Z error libmamba Could not solve for environment specs
2025-11-20T13:27:19.6050065Z The following packages are incompatible
2025-11-20T13:27:19.6055643Z ├─ cuda-version =13.0 * is requested and can be installed;
[... dependency chain showing libxgboost → librmm → cuda-version <13 conflict ...]
2025-11-20T13:27:19.6118187Z └─ cuda-version >=12,<13.0a0 *, which conflicts with any installable versions previously reported.
2025-11-20T13:27:19.7099577Z critical libmamba Could not solve for environment specs
Root Cause Analysis
The rapids-xgboost package version 26.2 has contradictory CUDA version constraints in its dependency chain that make it unsolvable for any CUDA version:
The Contradiction
-
Top-level constraint (rapids-xgboost → cuda-version):
rapids-xgboost =26.2directly requirescuda-version >=13,<14.0a0- This blocks all CUDA 12.x installations
-
Transitive constraint (rapids-xgboost → libxgboost → librmm → cuda-version):
rapids-xgboost =26.2requireslibxgboost =3.0.5libxgboost =3.0.5requireslibrmm >=25.12.0a8,<25.13.0a0- Compatible
librmmversions requirecuda-version >=12,<13.0a0 - This blocks all CUDA 13.x installations
Dependency Chain Visualization
rapids-xgboost 26.2
├─ DIRECT: cuda-version >=13,<14.0a0 ❌ Blocks CUDA 12.x
└─ libxgboost 3.0.5
└─ librmm 25.12.x
└─ cuda-version >=12,<13.0a0 ❌ Blocks CUDA 13.x
Result: No CUDA version (12.x or 13.x) can satisfy both constraints simultaneously.