ci: stabilize optional backend workflows (#5945)

## Summary

- allow Paddle-only C++ matrix entries to fail without cancelling or
blocking the other backend jobs
- temporarily disable Paddle installation and tests in the CUDA workflow
- replace the manual uv installer with astral-sh/setup-uv@v9, pin uv
0.12.0, and limit macOS dependency downloads to avoid runner
socket-buffer exhaustion

## Failure analysis

The referenced macOS ARM job successfully built and repaired the wheel.
It failed while installing wheel test dependencies because downloading
lmdb raised No buffer space available (os error 55). The macOS wheel
remains blocking so release runs cannot silently publish without that
artifact; limiting uv download concurrency targets the observed runner
resource failure.

## Validation

- prek run --files .github/workflows/test_cc.yml
.github/workflows/test_cuda.yml .github/workflows/build_wheel.yml
- ruff check .
- ruff format .
- actionlint on the changed workflows, excluding existing custom-runner
and undefined optional matrix-field diagnostics

Coding agent: Codex
Codex version: codex-cli 0.144.6
Model: gpt-5.6-sol
Reasoning effort: xhigh

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **Chores**
  * Improved wheel build reliability across supported operating systems.
* Limited macOS build download concurrency for more consistent
performance.

* **Tests**
* Improved test workflow resilience when Paddle-enabled checks encounter
failures.
  * Disabled Paddle-specific setup in CUDA test runs.
* Allowed CUDA and backend checks to continue independently when Paddle
is unavailable.
  * Continued requiring TensorFlow and PyTorch for applicable test runs.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Co-authored-by: njzjz-bot <njzjz-bot@users.noreply.github.com>

Author: njzjz

.github/workflows/build_wheel.yml

@@ -56,7 +56,9 @@ jobs:
           # https://github.com/pypa/setuptools_scm/issues/480
           fetch-depth: 0
       - name: Install uv
-        run: curl --proto '=https' --tlsv1.2 -LsSf https://github.com/astral-sh/uv/releases/download/0.2.24/uv-installer.sh | sh
+        uses: astral-sh/setup-uv@c771a70e6277c0a99b617c7a806ffedaca235ff9 # v9.0.0
+        with:
+          version: "0.12.0"
         if: runner.os != 'Linux'
       - name: Build wheels
         uses: pypa/cibuildwheel@v4.1
@@ -68,6 +70,9 @@ jobs:
           CUDA_VERSION: ${{ matrix.cuda_version }}
           DP_PKG_NAME: ${{ matrix.dp_pkg_name }}
           CIBW_BUILD_FRONTEND: "build[uv]"
+          # uv defaults to 50 concurrent downloads, which can exhaust the
+          # macOS runner's socket buffers while installing wheel test extras.
+          UV_CONCURRENT_DOWNLOADS: ${{ startsWith(matrix.platform_id, 'macosx_') && '8' || '50' }}
       - uses: actions/upload-artifact@v7
         with:
           name: cibw-cp${{ matrix.python }}-${{ matrix.platform_id }}-cu${{ matrix.cuda_version }}-${{ strategy.job-index }}

.github/workflows/test_cc.yml

@@ -15,7 +15,12 @@ jobs:
   testcc:
     name: Test C++
     runs-on: ${{ matrix.check_memleak == false && github.repository_owner == 'deepmodeling' && 'cpu' || 'ubuntu-22.04' }}
+    # Paddle's external interface download is occasionally unavailable. Keep
+    # exercising these configurations without making that outage blocking.
+    continue-on-error: ${{ matrix.enable_paddle }}
     strategy:
+      # An allowed Paddle failure must not cancel the other backend jobs.
+      fail-fast: false
       matrix:
         # Only regular jobs run the LAMMPS tests. Use self-hosted CPU runners
         # for those jobs in deepmodeling, with a hosted fallback for forks.

.github/workflows/test_cuda.yml

@@ -13,6 +13,11 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.ref || github.run_id }}
   cancel-in-progress: true
 name: Test CUDA
+env:
+  # Temporarily disable Paddle in CUDA CI while its external downloads are
+  # unreliable. Keep this switch explicit so the coverage can be restored.
+  DP_CI_ALLOW_MISSING_PADDLE: "1"
+  DP_ENABLE_PADDLE: "0"
 jobs:
   test_python:
     name: Test Python on CUDA
@@ -53,7 +58,6 @@ jobs:
       - run: |
           export PYTORCH_ROOT=$(python -c 'import torch;print(torch.__path__[0])')
           export TENSORFLOW_ROOT=$(python -c 'import importlib.util,pathlib;print(pathlib.Path(importlib.util.find_spec("tensorflow").origin).parent)')
-          pip install --find-links "https://www.paddlepaddle.org.cn/packages/nightly/cu126/paddlepaddle-gpu/" --index-url https://pypi.org/simple --trusted-host www.paddlepaddle.org.cn --trusted-host paddlepaddle.org.cn "paddlepaddle-gpu==3.4.0.dev20260310"
           source/install/uv_with_retry.sh pip install --system -v -e .[gpu,test,lmp,cu12,torch,jax] mpi4py --reinstall-package deepmd-kit
           # See https://github.com/jax-ml/jax/issues/29042
           source/install/uv_with_retry.sh pip install --system -U 'nvidia-cublas-cu12>=12.9.0.13'
@@ -62,7 +66,7 @@ jobs:
           DP_ENABLE_NATIVE_OPTIMIZATION: 1
           DP_ENABLE_PYTORCH: 1
       - run: dp --version
-      - run: python -m pytest source/tests
+      - run: python -m pytest source/tests --ignore=source/tests/pd
         env:
           NUM_WORKERS: 0
           CUDA_VISIBLE_DEVICES: 0
@@ -106,7 +110,6 @@ jobs:
       - run: |
           export PYTORCH_ROOT=$(python -c 'import torch;print(torch.__path__[0])')
           export TENSORFLOW_ROOT=$(python -c 'import importlib.util,pathlib;print(pathlib.Path(importlib.util.find_spec("tensorflow").origin).parent)')
-          pip install --find-links "https://www.paddlepaddle.org.cn/packages/nightly/cu126/paddlepaddle-gpu/" --index-url https://pypi.org/simple --trusted-host www.paddlepaddle.org.cn --trusted-host paddlepaddle.org.cn "paddlepaddle-gpu==3.4.0.dev20260310"
           source/install/uv_with_retry.sh pip install --system -v -e .[gpu,test,lmp,cu12,torch,jax] mpi4py --reinstall-package deepmd-kit
           # See https://github.com/jax-ml/jax/issues/29042
           source/install/uv_with_retry.sh pip install --system -U 'nvidia-cublas-cu12>=12.9.0.13'
@@ -127,12 +130,10 @@ jobs:
           CMAKE_GENERATOR: Ninja
           DP_VARIANT: cuda
           DP_USE_MPICH2: 1
+          ENABLE_PADDLE: FALSE
       - run: |
           export LD_LIBRARY_PATH=$CUDA_PATH/lib64:/usr/lib/x86_64-linux-gnu/:$GITHUB_WORKSPACE/dp_test/lib:$LD_LIBRARY_PATH
           export PATH=$GITHUB_WORKSPACE/dp_test/bin:$PATH
-          cp $GITHUB_WORKSPACE/source/build_tests/paddle_inference_install_dir/paddle/lib/* $GITHUB_WORKSPACE/dp_test/lib/
-          cp $GITHUB_WORKSPACE/source/build_tests/paddle_inference_install_dir/third_party/install/onednn/lib/* $GITHUB_WORKSPACE/dp_test/lib/
-          cp $GITHUB_WORKSPACE/source/build_tests/paddle_inference_install_dir/third_party/install/mklml/lib/* $GITHUB_WORKSPACE/dp_test/lib/
           python -m pytest -s source/lmp/tests || (cat log.lammps && exit 1)
           python -m pytest source/ipi/tests
         env:
@@ -141,6 +142,7 @@ jobs:
           TF_INTER_OP_PARALLELISM_THREADS: 1
           LAMMPS_PLUGIN_PATH: ${{ github.workspace }}/dp_test/lib/deepmd_lmp
           CUDA_VISIBLE_DEVICES: 0
+          ENABLE_PADDLE: 0
   pass:
     name: Pass testing on CUDA
     needs: [test_python, test_cc]

source/tests/consistent/common.py

@@ -62,11 +62,16 @@
 INSTALLED_JAX = Backend.get_backend("jax")().is_available()
 INSTALLED_PD = Backend.get_backend("paddle")().is_available()
 INSTALLED_ARRAY_API_STRICT = find_spec("array_api_strict") is not None
+# CUDA CI may explicitly omit Paddle while its external artifacts are
+# unavailable, without disabling the remaining cross-backend checks.
+CI_ALLOW_MISSING_PADDLE = os.environ.get("DP_CI_ALLOW_MISSING_PADDLE") == "1"
 
 if (
     os.environ.get("CI")
     and not RUN_TF2_BACKEND_TESTS
-    and not (INSTALLED_TF and INSTALLED_PT and INSTALLED_PD)
+    and not (
+        INSTALLED_TF and INSTALLED_PT and (INSTALLED_PD or CI_ALLOW_MISSING_PADDLE)
+    )
 ):
     raise ImportError("TensorFlow, PyTorch or Paddle should be tested in the CI")