[CI] Enable matrix test for UT #1677
Conversation
RUIJIEZHONG66166
force-pushed
the
ruijie/enable_ut_matrix
branch
from
d206dfa to
5491aa6
Compare
| rm -rf $(dirname ${CONDA_EXE})/../envs/xpu_op_${ZE_AFFINITY_MASK} | ||
| conda create -n xpu_op_${ZE_AFFINITY_MASK} python=${{ inputs.python }} cmake ninja -y | ||
| source activate xpu_op_${ZE_AFFINITY_MASK} | ||
| conda remove --all -y -n $CONDA_ENV_NAME || \ |
There was a problem hiding this comment.
Command on a previous line, conda clean -ay is a big problem if you have a few runners running on the same system. This effectively means that you can damage environment you create/work in one runner from another and get all sorts of fancy random failures.
There was a problem hiding this comment.
Got it. Thank you~~
.github/workflows/_linux_ut.yml
Outdated
| - name: Create unique workspace | ||
| run: | | ||
| # Create unique conda env for each UT test | ||
| echo "CONDA_ENV_NAME=xpu_op_${ZE_AFFINITY_MASK}_${{ matrix.test.name }}" >> $GITHUB_ENV |
There was a problem hiding this comment.
That's not truly unique actually. Meaning that if there are 2 of these workloads running in parallel, for example for different PRs, these will start to fail. To address this you can reuse approach from #1282. It will work only with Linux though. For Windows you will need something else:
| echo "CONDA_ENV_NAME=xpu_op_${ZE_AFFINITY_MASK}_${{ matrix.test.name }}" >> $GITHUB_ENV | |
| random=$(head /dev/urandom | tr -dc A-Za-z0-9_ | head -c ${1:-5} | xargs) | |
| echo "CONDA_ENV_NAME=xpu_op_${ZE_AFFINITY_MASK}_${{ matrix.test.name }}_${random}" >> $GITHUB_ENV |
There was a problem hiding this comment.
Got it. Thank you so much for your help~~
Enable matrix test method to accelerate the UT test.