Fix test timeout issue in pr-test (#17681)

This commit is contained in:
Kangyan-Zhou
2026-01-24 11:46:08 -08:00
committed by GitHub
parent 2c2c4e446b
commit 8656a146a6
+31 -31
View File
@@ -376,7 +376,7 @@ jobs:
# Skip for scheduled runs (they run stages independently) and when target_stage is set # 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' 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 runs-on: x64-kernel-build-node
timeout-minutes: 60 timeout-minutes: 240
strategy: strategy:
matrix: matrix:
include: include:
@@ -425,7 +425,7 @@ jobs:
# Skip for scheduled runs (they run stages independently) and when target_stage is set # 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' 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 runs-on: arm-kernel-build-node
timeout-minutes: 60 timeout-minutes: 240
strategy: strategy:
matrix: matrix:
include: include:
@@ -479,7 +479,7 @@ jobs:
!inputs.target_stage && !inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true' needs.check-changes.outputs.sgl_kernel == 'true'
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
steps: steps:
@@ -519,7 +519,7 @@ jobs:
!inputs.target_stage && !inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true' needs.check-changes.outputs.sgl_kernel == 'true'
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
steps: steps:
@@ -559,7 +559,7 @@ jobs:
!inputs.target_stage && !inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true' needs.check-changes.outputs.sgl_kernel == 'true'
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
CI: true CI: true
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
@@ -612,7 +612,7 @@ jobs:
!inputs.target_stage && !inputs.target_stage &&
needs.check-changes.outputs.sgl_kernel == 'true' needs.check-changes.outputs.sgl_kernel == 'true'
runs-on: ${{ needs.check-changes.outputs.b200_runner }} runs-on: ${{ needs.check-changes.outputs.b200_runner }}
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }} RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
steps: steps:
@@ -685,7 +685,7 @@ jobs:
!inputs.target_stage && !inputs.target_stage &&
needs.check-changes.outputs.jit_kernel == 'true' needs.check-changes.outputs.jit_kernel == 'true'
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
steps: steps:
@@ -719,7 +719,7 @@ jobs:
) )
) )
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
steps: steps:
@@ -766,7 +766,7 @@ jobs:
) )
) )
runs-on: ubuntu-latest runs-on: ubuntu-latest
timeout-minutes: 60 timeout-minutes: 240
steps: steps:
- name: Free disk space - name: Free disk space
run: | run: |
@@ -812,7 +812,7 @@ jobs:
) )
) )
runs-on: 1-gpu-5090 runs-on: 1-gpu-5090
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-5090 RUNNER_LABELS: 1-gpu-5090
IS_BLACKWELL: "1" IS_BLACKWELL: "1"
@@ -866,7 +866,7 @@ jobs:
) )
) )
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
strategy: strategy:
@@ -916,7 +916,7 @@ jobs:
) )
) )
runs-on: 2-gpu-runner runs-on: 2-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 2-gpu-runner RUNNER_LABELS: 2-gpu-runner
strategy: strategy:
@@ -965,7 +965,7 @@ jobs:
) )
) )
runs-on: 1-gpu-5090 runs-on: 1-gpu-5090
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-5090 RUNNER_LABELS: 1-gpu-5090
IS_BLACKWELL: "1" IS_BLACKWELL: "1"
@@ -1013,7 +1013,7 @@ jobs:
) )
) )
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-runner RUNNER_LABELS: 1-gpu-runner
strategy: strategy:
@@ -1062,7 +1062,7 @@ jobs:
) )
) )
runs-on: 2-gpu-runner runs-on: 2-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 2-gpu-runner RUNNER_LABELS: 2-gpu-runner
steps: steps:
@@ -1107,7 +1107,7 @@ jobs:
) )
) )
runs-on: 1-gpu-5090 runs-on: 1-gpu-5090
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 1-gpu-5090 RUNNER_LABELS: 1-gpu-5090
IS_BLACKWELL: "1" IS_BLACKWELL: "1"
@@ -1158,7 +1158,7 @@ jobs:
) )
) )
runs-on: 2-gpu-runner runs-on: 2-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 2-gpu-runner RUNNER_LABELS: 2-gpu-runner
steps: steps:
@@ -1206,7 +1206,7 @@ jobs:
) )
) )
runs-on: ${{ needs.check-changes.outputs.b200_runner }} runs-on: ${{ needs.check-changes.outputs.b200_runner }}
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }} RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
strategy: strategy:
@@ -1259,7 +1259,7 @@ jobs:
) )
) )
runs-on: 4-gpu-h100 runs-on: 4-gpu-h100
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 4-gpu-h100 RUNNER_LABELS: 4-gpu-h100
steps: steps:
@@ -1304,7 +1304,7 @@ jobs:
) )
) )
runs-on: ${{ needs.check-changes.outputs.b200_runner }} runs-on: ${{ needs.check-changes.outputs.b200_runner }}
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }} RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
steps: steps:
@@ -1345,7 +1345,7 @@ jobs:
) )
) )
runs-on: 1-gpu-runner runs-on: 1-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
strategy: strategy:
fail-fast: false fail-fast: false
matrix: matrix:
@@ -1369,7 +1369,7 @@ jobs:
run: | run: |
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh diffusion CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh diffusion
- name: Run diffusion server tests - name: Run diffusion server tests
timeout-minutes: 60 timeout-minutes: 240
run: | run: |
cd python cd python
CONTINUE_ON_ERROR_FLAG="" CONTINUE_ON_ERROR_FLAG=""
@@ -1396,7 +1396,7 @@ jobs:
) )
) )
runs-on: 2-gpu-runner runs-on: 2-gpu-runner
timeout-minutes: 60 timeout-minutes: 240
strategy: strategy:
fail-fast: false fail-fast: false
matrix: matrix:
@@ -1421,7 +1421,7 @@ jobs:
CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh diffusion CUSTOM_BUILD_SGL_KERNEL=${{needs.check-changes.outputs.sgl_kernel}} bash scripts/ci/ci_install_dependency.sh diffusion
- name: Run diffusion server tests - name: Run diffusion server tests
timeout-minutes: 60 timeout-minutes: 240
run: | run: |
cd python cd python
CONTINUE_ON_ERROR_FLAG="" CONTINUE_ON_ERROR_FLAG=""
@@ -1447,7 +1447,7 @@ jobs:
) )
) )
runs-on: 4-gpu-h100 runs-on: 4-gpu-h100
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 4-gpu-h100 RUNNER_LABELS: 4-gpu-h100
strategy: strategy:
@@ -1500,7 +1500,7 @@ jobs:
) )
) )
runs-on: 8-gpu-h200 runs-on: 8-gpu-h200
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 8-gpu-h200 RUNNER_LABELS: 8-gpu-h200
strategy: strategy:
@@ -1559,7 +1559,7 @@ jobs:
) )
) )
runs-on: 8-gpu-h20 runs-on: 8-gpu-h20
timeout-minutes: 60 timeout-minutes: 240
env: env:
SGLANG_CI_RDMA_ALL_DEVICES: "mlx5_1,mlx5_2,mlx5_3,mlx5_4" SGLANG_CI_RDMA_ALL_DEVICES: "mlx5_1,mlx5_2,mlx5_3,mlx5_4"
RUNNER_LABELS: 8-gpu-h20 RUNNER_LABELS: 8-gpu-h20
@@ -1613,7 +1613,7 @@ jobs:
) )
) )
runs-on: 4-gpu-h100 runs-on: 4-gpu-h100
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 4-gpu-h100 RUNNER_LABELS: 4-gpu-h100
steps: steps:
@@ -1662,7 +1662,7 @@ jobs:
) )
) )
runs-on: 8-gpu-h200 runs-on: 8-gpu-h200
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 8-gpu-h200 RUNNER_LABELS: 8-gpu-h200
steps: steps:
@@ -1711,7 +1711,7 @@ jobs:
) )
) )
runs-on: ${{ needs.check-changes.outputs.b200_runner }} runs-on: ${{ needs.check-changes.outputs.b200_runner }}
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }} RUNNER_LABELS: ${{ needs.check-changes.outputs.b200_runner }}
strategy: strategy:
@@ -1765,7 +1765,7 @@ jobs:
) )
) )
runs-on: 4-gpu-gb200 runs-on: 4-gpu-gb200
timeout-minutes: 60 timeout-minutes: 240
env: env:
RUNNER_LABELS: 4-gpu-gb200 RUNNER_LABELS: 4-gpu-gb200
strategy: strategy: