Fix test timeout issue in pr-test (#17681)
This commit is contained in:
62
.github/workflows/pr-test.yml
vendored
62
.github/workflows/pr-test.yml
vendored
@@ -376,7 +376,7 @@ jobs:
|
||||
# Skip for scheduled runs (they run stages independently) and when target_stage is set
|
||||
if: github.event_name != 'schedule' && inputs.test_parallel_dispatch != true && !inputs.target_stage && needs.check-changes.outputs.sgl_kernel == 'true'
|
||||
runs-on: x64-kernel-build-node
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
strategy:
|
||||
matrix:
|
||||
include:
|
||||
@@ -425,7 +425,7 @@ jobs:
|
||||
# Skip for scheduled runs (they run stages independently) and when target_stage is set
|
||||
if: github.event_name != 'schedule' && inputs.test_parallel_dispatch != true && !inputs.target_stage && needs.check-changes.outputs.sgl_kernel == 'true'
|
||||
runs-on: arm-kernel-build-node
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
strategy:
|
||||
matrix:
|
||||
include:
|
||||
@@ -479,7 +479,7 @@ jobs:
|
||||
!inputs.target_stage &&
|
||||
needs.check-changes.outputs.sgl_kernel == 'true'
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
steps:
|
||||
@@ -519,7 +519,7 @@ jobs:
|
||||
!inputs.target_stage &&
|
||||
needs.check-changes.outputs.sgl_kernel == 'true'
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
steps:
|
||||
@@ -559,7 +559,7 @@ jobs:
|
||||
!inputs.target_stage &&
|
||||
needs.check-changes.outputs.sgl_kernel == 'true'
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
CI: true
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
@@ -612,7 +612,7 @@ jobs:
|
||||
!inputs.target_stage &&
|
||||
needs.check-changes.outputs.sgl_kernel == 'true'
|
||||
runs-on: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
steps:
|
||||
@@ -685,7 +685,7 @@ jobs:
|
||||
!inputs.target_stage &&
|
||||
needs.check-changes.outputs.jit_kernel == 'true'
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
steps:
|
||||
@@ -719,7 +719,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
steps:
|
||||
@@ -766,7 +766,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: ubuntu-latest
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
steps:
|
||||
- name: Free disk space
|
||||
run: |
|
||||
@@ -812,7 +812,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-5090
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-5090
|
||||
IS_BLACKWELL: "1"
|
||||
@@ -866,7 +866,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
strategy:
|
||||
@@ -916,7 +916,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 2-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 2-gpu-runner
|
||||
strategy:
|
||||
@@ -965,7 +965,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-5090
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-5090
|
||||
IS_BLACKWELL: "1"
|
||||
@@ -1013,7 +1013,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-runner
|
||||
strategy:
|
||||
@@ -1062,7 +1062,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 2-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 2-gpu-runner
|
||||
steps:
|
||||
@@ -1107,7 +1107,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-5090
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 1-gpu-5090
|
||||
IS_BLACKWELL: "1"
|
||||
@@ -1158,7 +1158,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 2-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 2-gpu-runner
|
||||
steps:
|
||||
@@ -1206,7 +1206,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
strategy:
|
||||
@@ -1259,7 +1259,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 4-gpu-h100
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 4-gpu-h100
|
||||
steps:
|
||||
@@ -1304,7 +1304,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
steps:
|
||||
@@ -1345,7 +1345,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 1-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
strategy:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
@@ -1369,7 +1369,7 @@ jobs:
|
||||
run: |
|
||||
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh diffusion
|
||||
- name: Run diffusion server tests
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
run: |
|
||||
cd python
|
||||
CONTINUE_ON_ERROR_FLAG=""
|
||||
@@ -1396,7 +1396,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 2-gpu-runner
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
strategy:
|
||||
fail-fast: false
|
||||
matrix:
|
||||
@@ -1421,7 +1421,7 @@ jobs:
|
||||
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh diffusion
|
||||
|
||||
- name: Run diffusion server tests
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
run: |
|
||||
cd python
|
||||
CONTINUE_ON_ERROR_FLAG=""
|
||||
@@ -1447,7 +1447,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 4-gpu-h100
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 4-gpu-h100
|
||||
strategy:
|
||||
@@ -1500,7 +1500,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 8-gpu-h200
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 8-gpu-h200
|
||||
strategy:
|
||||
@@ -1559,7 +1559,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 8-gpu-h20
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
SGLANG_CI_RDMA_ALL_DEVICES: "mlx5_1,mlx5_2,mlx5_3,mlx5_4"
|
||||
RUNNER_LABELS: 8-gpu-h20
|
||||
@@ -1613,7 +1613,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 4-gpu-h100
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 4-gpu-h100
|
||||
steps:
|
||||
@@ -1662,7 +1662,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 8-gpu-h200
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 8-gpu-h200
|
||||
steps:
|
||||
@@ -1711,7 +1711,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
|
||||
strategy:
|
||||
@@ -1765,7 +1765,7 @@ jobs:
|
||||
)
|
||||
)
|
||||
runs-on: 4-gpu-gb200
|
||||
timeout-minutes: 60
|
||||
timeout-minutes: 240
|
||||
env:
|
||||
RUNNER_LABELS: 4-gpu-gb200
|
||||
strategy:
|
||||
|
||||
Reference in New Issue
Block a user