Added AWS EC2 dynamic runner support

.github/workflows/sycl_linux_build_and_test.yml

@@ -64,6 +64,11 @@ on:
         type: number
         required: false
         default: 4
+      uniq:
+        description: Unique string to name dynamic runers in AWS
+        type: string
+        required: false
+        default: ${{ github.run_id }}-${{ github.run_attempt }}
 
 jobs:
   build:
@@ -191,8 +196,40 @@ jobs:
           const script = require('./generate_test_matrix.js');
           script({core, process});
 
+aws-start-matrix:
+    name: Start AWS Matrix
+    needs: [ build, resolve_matrix ]
+    if: ${{ inputs.lts_config != '' }}
+    strategy:
+      fail-fast: false
+      max-parallel: ${{ inputs.max_parallel }}
+      matrix:
+        include: ${{ fromJSON(needs.resolve_matrix.outputs.lts) }}
+    runs-on: ubuntu-latest
+    environment: aws
+    steps:
+      - uses: actions/checkout@v3
+        if: ${{ matrix.aws-type }}
+        with:
+          path: llvm
+      - run: npm install ./llvm/devops/actions/aws-ec2
+        if: ${{ matrix.aws-type }}
+      - name: Start AWS EC2 runner
+        if: ${{ matrix.aws-type }}
+        uses: ./llvm/devops/actions/aws-ec2
+        with:
+          label: ${{ matrix.runs-on }}
+          GH_PERSONAL_ACCESS_TOKEN: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
+          AWS_ACCESS_KEY: ${{ secrets.AWS_ACCESS_KEY }}
+          AWS_SECRET_KEY: ${{ secrets.AWS_SECRET_KEY }}
+          aws-ami:  ${{ matrix.aws-ami }}
+          aws-spot: ${{ matrix.aws-spot }}
+          aws-type: ${{ matrix.aws-type }}
+          aws-disk: ${{ matrix.aws-disk }}
+          one-job:  ${{ matrix.one-job }}
+
   llvm_test_suite:
-    needs: [build, resolve_matrix]
+    needs: [build, resolve_matrix, aws-start-matrix]
     if: ${{ inputs.lts_config != '' }}
     strategy:
       fail-fast: false
@@ -240,3 +277,30 @@ jobs:
         results_name_suffix: ${{ matrix.config }}_${{ inputs.build_artifact_suffix }}
         cmake_args: '${{ matrix.cmake_args }} ${{ inputs.lts_cmake_extra_args }}'
 
+  aws-stop-matrix:
+    name: Stop AWS Matrix
+    needs: [ aws-start-matrix, resolve_matrix, llvm_test_suite ]
+    if: ${{ always() && inputs.lts_config != '' }}
+    strategy:
+      fail-fast: false
+      max-parallel: ${{ inputs.max_parallel }}
+      matrix:
+        include: ${{ fromJSON(needs.resolve_matrix.outputs.lts) }}
+    runs-on: ubuntu-latest
+    environment: aws
+    steps:
+      - uses: actions/checkout@v3
+        if: ${{ matrix.aws-type }}
+        with:
+          path: llvm
+      - run: npm install ./llvm/devops/actions/aws-ec2
+        if: ${{ matrix.aws-type }}
+      - name: Stop AWS EC2 runner
+        if: ${{ matrix.aws-type }}
+        uses: ./llvm/devops/actions/aws-ec2
+        with:
+          label: ${{ matrix.runs-on }}
+          mode: stop
+          GH_PERSONAL_ACCESS_TOKEN: ${{ secrets.GH_PERSONAL_ACCESS_TOKEN }}
+          AWS_ACCESS_KEY: ${{ secrets.AWS_ACCESS_KEY }}
+          AWS_SECRET_KEY: ${{ secrets.AWS_SECRET_KEY }}

.github/workflows/sycl_nightly.yml

@@ -13,6 +13,7 @@ jobs:
   ubuntu2004_build_test:
     if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
+    secrets: inherit
     with:
       build_cache_root: "/__w/"
       build_artifact_suffix: default
@@ -22,6 +23,7 @@ jobs:
   ubuntu2004_opaque_pointers_build_test:
     if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
+    secrets: inherit
     with:
       build_cache_root: "/__w/"
       build_cache_suffix: opaque_pointers

.github/workflows/sycl_post_commit.yml

@@ -14,12 +14,14 @@ jobs:
   linux_default:
     name: Linux Default
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
+    secrets: inherit
     with:
       build_cache_root: "/__w/llvm"
       build_artifact_suffix: default
   linux_no_assert:
     name: Linux (no assert)
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
+    secrets: inherit
     with:
       build_cache_root: "/__w/llvm"
       build_cache_suffix: gcc_no_assertions

.github/workflows/sycl_precommit.yml

@@ -36,6 +36,7 @@ jobs:
     needs: lint
     if: always() && (success() || contains(github.event.pull_request.labels.*.name, 'ignore-lint'))
     uses: ./.github/workflows/sycl_linux_build_and_test.yml
+    secrets: inherit
     with:
       build_cache_root: "/__w/"
       build_cache_size: "8G"

devops/actions/aws-ec2/action.yml

@@ -0,0 +1,80 @@
+name: aws-ec2
+description: Starts AWS EC2 instance with github actions runner agent in it to process one job
+
+inputs:
+  label:
+    description: "Name of the unique label assigned to the runner used as 'runs-on' property for the following jobs"
+    required: true
+  GH_PERSONAL_ACCESS_TOKEN:
+    description: "Github personal access token with repo permission"
+    required: true
+  AWS_ACCESS_KEY:
+    description: "AWS access id"
+    required: true
+  AWS_SECRET_KEY:
+    description: "WS access secret key"
+    required: true
+  aws-region:
+    description: "AWS EC2 region"
+    required: false
+    default: "us-east-2" # Ohio
+  aws-ami:
+    description: "AWS AMI id. Makes sense only for start mode"
+    required: false
+    default: "ami-0966bccbb521ccb24" # Ubuntu 22.04 (ami-02f3416038bdb17fb with /dev/sda1 disk) with docker installed and gh_runner (1001) like this:
+    # sudo -s
+    # apt-get update
+    # curl -fsSL https://get.docker.com -o /tmp/get-docker.sh
+    # sh /tmp/get-docker.sh # or "yum install -y docker" for Amazon Linux or RHEL/CentOS
+    # groupadd -g 1001 gh_runner; useradd gh_runner -u 1001 -g 1001 -m -s /bin/bash; usermod -aG docker gh_runner; usermod -aG video gh_runner
+    # sync; shutdown -h now
+
+    # "ami-02ec0f344128253f9" # Amazon Linux 2 AMI with NVIDIA TESLA GPU Driver (ami-06bf0a3f89fe08f0a with /dev/xvda disk) with docker installed and gh_runner (1001) like this:
+    # sudo -s
+    # yum update -y
+    # amazon-linux-extras install docker
+    # sudo systemctl --now enable docker
+    # distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.repo | sudo tee /etc/yum.repos.d/nvidia-container-toolkit.repo
+    # yum-config-manager --disable amzn2-graphics; yum clean expire-cache; yum install -y nvidia-docker2; systemctl restart docker
+    # groupadd -g 1001 gh_runner; useradd gh_runner -u 1001 -g 1001 -m -s /bin/bash; usermod -aG docker gh_runner; usermod -aG video gh_runner
+    # sync; shutdown -h now
+
+    # "ami-0ccda708841dde988" # Amazon Linux 2 AMI with AMD Radeon Pro Driver (ami-0bb1072e787242eb6 with /dev/xvda disk) with docker installed and gh_runner (1001) like this:
+    # sudo -s
+    # sh ./get-docker.sh # or "yum install -y docker" for Amazon Linux or RHEL/CentOS
+    # amazon-linux-extras install docker
+    # sudo systemctl --now enable docker
+    # groupadd -g 1001 gh_runner; useradd gh_runner -u 1001 -g 1001 -m -s /bin/bash; usermod -aG docker gh_runner; usermod -aG video gh_runner
+    # sync; shutdown -h now
+
+  aws-type:
+    description: "AWS EC2 instance type. Makes sense only for start mode"
+    required: false
+    default: "[ \"t2.micro\" ]"
+    # "[ \"g4dn.2xlarge\" ]": 1 NVIDIA T4 GPU, 8 CPU, 32 GB RAM 
+    # "[ \"g4ad.4xlarge\" ]": 1 AMD Radeon Pro V520 GPU, 16 CPU, 64 GB RAM (overloaded pool so do not use spot here now)
+  aws-spot:
+    description: "Enable usage of spot instances to save money (less reliable). Makes sense only for start mode"
+    requred: false
+    default: true
+  aws-disk:
+    description: "AWS EC2 instance AMI specific disk device path and size in GB (8 by default). Makes sense only for start mode"
+    required: false
+    default: "/dev/sda1:16" # Github actions container engine will fail with lack of disk space for 8GB
+  aws-timebomb:
+    description: "AWS EC2 instance maximum live time. Makes sense only for start mode"
+    required: false
+    default: 1h
+
+  mode:
+    description: "Mode of operation: start or stop"
+    required: false
+    default: start
+  one-job:
+    description: "Will terminate AWS EC2 instance after one job (not waiting for stop job) saving money"
+    required: false
+    default: true
+
+runs:
+  using: node12
+  main: ./aws-ec2.js