From a45c597a0c54cb9b94349d7e8003bd257ce17ade Mon Sep 17 00:00:00 2001 From: zhangyue Date: Fri, 29 May 2026 10:36:48 +0800 Subject: [PATCH 1/3] ci: stabilize iluvatar runner budget --- .github/ci_config.yml | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/.github/ci_config.yml b/.github/ci_config.yml index 18d633a9f..e1e9ef2f5 100644 --- a/.github/ci_config.yml +++ b/.github/ci_config.yml @@ -55,16 +55,16 @@ platforms: gpu: type: unittest resources: - ngpus: 0 + ngpus: 1 gpu_style: none memory: 32GB shm_size: 16g - timeout: 3600 + timeout: 7200 queue_timeout: 600 junit_path: test-results.xml stages: - name: test - run: pytest tests/ --devices iluvatar -n 8 -v --tb=short --junitxml=/workspace/results/test-results.xml + run: pytest tests/ --devices iluvatar -n 4 -v --tb=short --junitxml=/workspace/results/test-results.xml metax: runner_label: Metax From f4e516aba5eb8fe8d8a7c7996546a306afead091 Mon Sep 17 00:00:00 2001 From: zhangyue Date: Fri, 29 May 2026 10:39:12 +0800 Subject: [PATCH 2/3] fix(ci): pin reusable workflows to content-tag images --- .ci | 2 +- .github/workflows/ci_test.yml | 4 ++-- .github/workflows/ci_v2_shadow.yml | 4 ++-- 3 files changed, 5 insertions(+), 5 deletions(-) diff --git a/.ci b/.ci index 45d1046ec..5893c29cc 160000 --- a/.ci +++ b/.ci @@ -1 +1 @@ -Subproject commit 45d1046ec42ea73d1a75d5a96eb1a8204d47cbcd +Subproject commit 5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 diff --git a/.github/workflows/ci_test.yml b/.github/workflows/ci_test.yml index 6700fed06..fd7fb2eae 100644 --- a/.github/workflows/ci_test.yml +++ b/.github/workflows/ci_test.yml @@ -27,10 +27,10 @@ on: jobs: ci: - uses: InfiniTensor/ci/.github/workflows/infiniops-ci.yml@45d1046ec42ea73d1a75d5a96eb1a8204d47cbcd + uses: InfiniTensor/ci/.github/workflows/infiniops-ci.yml@5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 with: config_path: .github/ci_config.yml - ci_ref: 45d1046ec42ea73d1a75d5a96eb1a8204d47cbcd + ci_ref: 5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 max_parallel: 10 platform: ${{ github.event_name == 'workflow_dispatch' && (inputs.platform == 'all' && 'nvidia,iluvatar,metax,moore,cambricon,ascend' || inputs.platform) || 'nvidia,iluvatar,metax,moore,cambricon,ascend' }} secrets: inherit diff --git a/.github/workflows/ci_v2_shadow.yml b/.github/workflows/ci_v2_shadow.yml index 2f49a2b1a..2c3731aed 100644 --- a/.github/workflows/ci_v2_shadow.yml +++ b/.github/workflows/ci_v2_shadow.yml @@ -28,10 +28,10 @@ on: jobs: ci-v2-shadow: - uses: InfiniTensor/ci/.github/workflows/infiniops-ci-v2-shadow.yml@45d1046ec42ea73d1a75d5a96eb1a8204d47cbcd + uses: InfiniTensor/ci/.github/workflows/infiniops-ci-v2-shadow.yml@5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 with: config_path: .github/ci_config.yml - ci_ref: 45d1046ec42ea73d1a75d5a96eb1a8204d47cbcd + ci_ref: 5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 max_parallel: 10 platform: ${{ github.event_name == 'workflow_dispatch' && (inputs.platform == 'all' && 'nvidia,iluvatar,metax,moore,cambricon,ascend' || inputs.platform == 'active' && 'nvidia,iluvatar,metax,moore,cambricon,ascend' || inputs.platform) || 'nvidia,iluvatar,metax,moore,cambricon,ascend' }} secrets: inherit From 8a56c691760d10e1df8e86aafc2a7de6997ba0a0 Mon Sep 17 00:00:00 2001 From: zhangyue Date: Fri, 29 May 2026 11:18:21 +0800 Subject: [PATCH 3/3] fix(ci): support static iluvatar device lease --- .ci | 2 +- .github/ci_config.yml | 13 +++++++------ .github/workflows/ci_test.yml | 4 ++-- .github/workflows/ci_v2_shadow.yml | 4 ++-- 4 files changed, 12 insertions(+), 11 deletions(-) diff --git a/.ci b/.ci index 5893c29cc..b45d360c8 160000 --- a/.ci +++ b/.ci @@ -1 +1 @@ -Subproject commit 5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 +Subproject commit b45d360c8cc529747ee31c5451d7eac96ac9f309 diff --git a/.github/ci_config.yml b/.github/ci_config.yml index e1e9ef2f5..4ab616010 100644 --- a/.github/ci_config.yml +++ b/.github/ci_config.yml @@ -24,7 +24,7 @@ platforms: memory: 32GB shm_size: 16g # Prevent PyTorch default 64MB shared memory limit. timeout: 3600 - queue_timeout: 600 + queue_timeout: 3600 junit_path: test-results.xml # env: # Uncomment to inject extra env vars into the container. # MY_VAR: value @@ -55,12 +55,13 @@ platforms: gpu: type: unittest resources: + gpu_ids: "0" ngpus: 1 gpu_style: none memory: 32GB shm_size: 16g timeout: 7200 - queue_timeout: 600 + queue_timeout: 7200 junit_path: test-results.xml stages: - name: test @@ -89,7 +90,7 @@ platforms: memory: 32GB shm_size: 16g timeout: 3600 - queue_timeout: 600 + queue_timeout: 3600 junit_path: test-results.xml stages: - name: test @@ -116,7 +117,7 @@ platforms: memory: 32GB shm_size: 16g timeout: 3600 - queue_timeout: 600 + queue_timeout: 3600 junit_path: test-results.xml stages: - name: test @@ -142,7 +143,7 @@ platforms: memory: 32GB shm_size: 16g timeout: 3600 - queue_timeout: 600 + queue_timeout: 3600 junit_path: test-results.xml stages: - name: test @@ -214,7 +215,7 @@ platforms: memory: 32GB shm_size: 16g timeout: 3600 - queue_timeout: 600 + queue_timeout: 3600 junit_path: test-results.xml stages: - name: test diff --git a/.github/workflows/ci_test.yml b/.github/workflows/ci_test.yml index fd7fb2eae..8dc54476f 100644 --- a/.github/workflows/ci_test.yml +++ b/.github/workflows/ci_test.yml @@ -27,10 +27,10 @@ on: jobs: ci: - uses: InfiniTensor/ci/.github/workflows/infiniops-ci.yml@5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 + uses: InfiniTensor/ci/.github/workflows/infiniops-ci.yml@b45d360c8cc529747ee31c5451d7eac96ac9f309 with: config_path: .github/ci_config.yml - ci_ref: 5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 + ci_ref: b45d360c8cc529747ee31c5451d7eac96ac9f309 max_parallel: 10 platform: ${{ github.event_name == 'workflow_dispatch' && (inputs.platform == 'all' && 'nvidia,iluvatar,metax,moore,cambricon,ascend' || inputs.platform) || 'nvidia,iluvatar,metax,moore,cambricon,ascend' }} secrets: inherit diff --git a/.github/workflows/ci_v2_shadow.yml b/.github/workflows/ci_v2_shadow.yml index 2c3731aed..25f02a11b 100644 --- a/.github/workflows/ci_v2_shadow.yml +++ b/.github/workflows/ci_v2_shadow.yml @@ -28,10 +28,10 @@ on: jobs: ci-v2-shadow: - uses: InfiniTensor/ci/.github/workflows/infiniops-ci-v2-shadow.yml@5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 + uses: InfiniTensor/ci/.github/workflows/infiniops-ci-v2-shadow.yml@b45d360c8cc529747ee31c5451d7eac96ac9f309 with: config_path: .github/ci_config.yml - ci_ref: 5893c29cc9db55d41740aa58f2f7e5eeaa7fc9e9 + ci_ref: b45d360c8cc529747ee31c5451d7eac96ac9f309 max_parallel: 10 platform: ${{ github.event_name == 'workflow_dispatch' && (inputs.platform == 'all' && 'nvidia,iluvatar,metax,moore,cambricon,ascend' || inputs.platform == 'active' && 'nvidia,iluvatar,metax,moore,cambricon,ascend' || inputs.platform) || 'nvidia,iluvatar,metax,moore,cambricon,ascend' }} secrets: inherit