xc-llm-ascend/.github/workflows/schedule_nightly_test_a2.yaml

#
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# This file is a part of the vllm-ascend project.
#

# This workflow related to the resources atlas 800 A2
# We will not limit the concurrency of jobs on A2
name: Nightly-A2

on:
  schedule:
      # Run test at 24:00 Beijing time (UTC+8)
      - cron: "0 16 * * *"
  workflow_dispatch:
  pull_request:
    branches:
      - 'main'
    types: [labeled, synchronize]

permissions:
  contents: read
  pull-requests: read
  issues: read

# Bash shells do not use ~/.profile or ~/.bashrc so these shells need to be explicitly
# declared as "shell: bash -el {0}" on steps that need to be properly activated.
# It's used to activate ascend-toolkit environment variables.
defaults:
  run:
    shell: bash -el {0}

# only cancel in-progress runs of the same workflow
concurrency:
  group: ascend-nightly-${{ github.ref }}-a2
  cancel-in-progress: true

jobs:
  parse-trigger:
    name: Parse trigger and determine test scope
    runs-on: linux-aarch64-a2b3-0
    outputs:
      should_run: ${{ steps.parse.outputs.should_run }}
      test_filter: ${{ steps.parse.outputs.test_filter }}
      is_pr_event: ${{ steps.parse.outputs.is_pr_event }}
    steps:
      - name: Parse trigger
        id: parse
        uses: actions/github-script@v7
        with:
          script: |
            const eventName = context.eventName;

            function parseNightlyComment(body) {
              if (!body) return null;
              const match = body.trim().match(/^\/nightly(?:\s+(.+))?$/m);
              if (!match) return null;
              const args = (match[1] || '').trim();
              if (!args || args === 'all') return 'all';
              // Wrap with commas for exact-name matching: ",name1,name2,"
              return ',' + args.split(/\s+/).join(',') + ',';
            }

            // schedule / workflow_dispatch: run all tests with pre-built image
            if (eventName === 'schedule' || eventName === 'workflow_dispatch') {
              core.setOutput('should_run', 'true');
              core.setOutput('test_filter', 'all');
              core.setOutput('is_pr_event', 'false');
              return;
            }

            // pull_request (labeled / synchronize)
            if (eventName === 'pull_request') {
              const labels = context.payload.pull_request.labels.map(l => l.name);
              if (!labels.includes('nightly-test')) {
                core.setOutput('should_run', 'false');
                core.setOutput('test_filter', '');
                core.setOutput('is_pr_event', 'true');
                return;
              }
              // Search comments for latest /nightly command
              const prNumber = context.payload.pull_request.number;
              const comments = await github.paginate(github.rest.issues.listComments, {
                owner: context.repo.owner,
                repo: context.repo.repo,
                issue_number: prNumber,
                per_page: 100,
              });
              let testFilter = null;
              for (let i = comments.length - 1; i >= 0; i--) {
                const result = parseNightlyComment(comments[i].body);
                if (result !== null) { testFilter = result; break; }
              }
              // No /nightly comment found: do not run any tests
              if (testFilter === null) {
                core.info('nightly-test label present but no /nightly comment found; skipping.');
                core.setOutput('should_run', 'false');
                core.setOutput('test_filter', '');
                core.setOutput('is_pr_event', 'true');
                return;
              }
              core.setOutput('should_run', 'true');
              core.setOutput('test_filter', testFilter);
              core.setOutput('is_pr_event', 'true');
              return;
            }

            // Fallback
            core.setOutput('should_run', 'false');
            core.setOutput('test_filter', '');
            core.setOutput('is_pr_event', 'false');

  single-node-tests:
    name: single-node
    needs: [parse-trigger]
    if: always() && needs.parse-trigger.outputs.should_run == 'true'
    strategy:
      fail-fast: false
      matrix:
        test_config:
          - name: test_custom_op
            os: linux-aarch64-a2b3-1
            tests: tests/e2e/nightly/single_node/ops/singlecard_ops
          - name: test_custom_op_multi_card
            os: linux-aarch64-a2b3-4
            tests: tests/e2e/nightly/single_node/ops/multicard_ops_a2/
    uses: ./.github/workflows/_e2e_nightly_single_node.yaml
    with:
      runner: ${{ matrix.test_config.os }}
      tests: ${{ matrix.test_config.tests }}
      name: ${{ matrix.test_config.name }}
      image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-a2'
      is_pr_test: >-
        ${{
          needs.parse-trigger.outputs.is_pr_event == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}
      is_run: >-
        ${{
          needs.parse-trigger.outputs.should_run == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}

  single-node-yaml-tests:
    name: single-node
    needs: [parse-trigger]
    if: always() && needs.parse-trigger.outputs.should_run == 'true'
    strategy:
      fail-fast: false
      matrix:
        test_config:
          - name: qwen3-32b
            os: linux-aarch64-a2b3-4
            config_file_path: Qwen3-32B.yaml
          - name: qwen3-next-80b-a3b-instruct
            os: linux-aarch64-a2b3-4
            config_file_path: Qwen3-Next-80B-A3B-Instruct-A2.yaml
          - name: qwen3-32b-int8
            os: linux-aarch64-a2b3-4
            config_file_path: Qwen3-32B-Int8-A2.yaml
    uses: ./.github/workflows/_e2e_nightly_single_node.yaml
    with:
      runner: ${{ matrix.test_config.os }}
      image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-a2'
      config_file_path: ${{ matrix.test_config.config_file_path }}
      name: ${{ matrix.test_config.name }}
      is_pr_test: >-
        ${{
          needs.parse-trigger.outputs.is_pr_event == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}
      is_run: >-
        ${{
          needs.parse-trigger.outputs.should_run == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}


  multi-node-tests:
    name: multi-node
    needs: [parse-trigger, single-node-tests, single-node-yaml-tests]
    if: always() && needs.parse-trigger.outputs.should_run == 'true'
    strategy:
      fail-fast: false
      max-parallel: 1
      matrix:
        test_config:
          - name: multi-node-deepseek-dp
            config_file_path: DeepSeek-R1-W8A8-A2.yaml
            size: 2
          - name: multi-node-qwen3-235b-dp
            config_file_path: Qwen3-235B-A22B-A2.yaml
            size: 2
    uses: ./.github/workflows/_e2e_nightly_multi_node.yaml
    with:
      soc_version: a2
      runner: linux-amd64-cpu-8-hk
      image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:nightly-a2'
      replicas: 1
      size: ${{ matrix.test_config.size }}
      config_file_path: ${{ matrix.test_config.config_file_path }}
      vllm_ascend_ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.ref_name }}
      is_pr_test: >-
        ${{
          needs.parse-trigger.outputs.is_pr_event == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}
      is_run: >-
        ${{
          needs.parse-trigger.outputs.should_run == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}
    secrets:
      KUBECONFIG_B64: ${{ secrets.KUBECONFIG_HK_001_INTERNAL_B64 }}

  single-node-accuracy-tests:
    needs: [parse-trigger]
    if: always() && needs.parse-trigger.outputs.should_run == 'true'
    strategy:
      fail-fast: false
      matrix:
        test_config:
          - name: accuracy-group-1
            os: linux-aarch64-a2b3-1
            model_list:
              - Qwen3-8B
              - Qwen2-Audio-7B-Instruct
              - Qwen3-8B-W8A8
              - Qwen3-VL-8B-Instruct
              - Qwen2.5-Omni-7B
          - name: accuracy-group-2
            os: linux-aarch64-a2b3-1
            model_list:
              - ERNIE-4.5-21B-A3B-PT
              - InternVL3_5-8B-hf
              - Molmo-7B-D-0924
              - Llama-3.2-3B-Instruct
              - llava-onevision-qwen2-0.5b-ov-hf
          - name: accuracy-group-3
            os: linux-aarch64-a2b3-2
            model_list:
              - Qwen3-30B-A3B
              - Qwen3-VL-30B-A3B-Instruct
              - Qwen3-30B-A3B-W8A8
          - name: accuracy-group-4
            os: linux-aarch64-a2b3-4
            model_list:
              - Qwen3-Next-80B-A3B-Instruct
              - Qwen3-Omni-30B-A3B-Instruct
    uses: ./.github/workflows/_e2e_nightly_single_node_models.yaml
    with:
      vllm: v0.16.0
      runner: ${{ matrix.test_config.os }}
      model_list: ${{ toJson(matrix.test_config.model_list) }}
      image: 'swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.1-910b-ubuntu22.04-py3.11'
      is_run: >-
        ${{
          needs.parse-trigger.outputs.should_run == 'true' && (
            needs.parse-trigger.outputs.test_filter == 'all' ||
            contains(needs.parse-trigger.outputs.test_filter, format(',{0},', matrix.test_config.name))
          )
        }}
      upload: false

  doc-test:
    name: doc-test
    needs: [parse-trigger]
    if: always() && (github.event_name == 'schedule' || github.event_name == 'workflow_dispatch')
    strategy:
      # Each version should be tested
      fail-fast: false
      matrix:
        vllm_version: [releases-v0.13.0, releases-v0.13.0-openeuler, main, main-openeuler]
    runs-on: linux-aarch64-a2b3-1
    container:
      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/vllm-ascend:${{ matrix.vllm_version }}
    steps:
      - name: Check NPU/CANN and git info
        run: |
          echo "====> Print NPU/CANN info"
          npu-smi info
          cat /usr/local/Ascend/ascend-toolkit/latest/"$(uname -i)"-linux/ascend_toolkit_install.info

          echo "====> Print vllm-ascend git info"
          cd /vllm-workspace/vllm-ascend
          git --no-pager log -1 || true
          echo "====> Print vllm git info"
          cd /vllm-workspace/vllm
          git --no-pager log -1 || true

      - name: Checkout vllm-project/vllm-ascend repo
        uses: actions/checkout@v6

      - name: Run vllm-ascend/tests/e2e/run_doctests.sh
        run: |
          # PWD: /__w/vllm-ascend/vllm-ascend
          # Make sure e2e tests are latest
          echo "Replacing /vllm-workspace/vllm-ascend/tests/e2e ..."
          rm -rf /vllm-workspace/vllm-ascend/tests/e2e
          mkdir -p /vllm-workspace/vllm-ascend/tests
          # Overwrite e2e and examples
          cp -r tests/e2e /vllm-workspace/vllm-ascend/tests/
          cp -r examples /vllm-workspace/vllm-ascend/

          # Simulate container to enter directory
          cd /workspace

          # Run real test
          echo "Test:"
          /vllm-workspace/vllm-ascend/tests/e2e/run_doctests.sh