Create an Azure VM #35
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: Create an Azure VM | |
| defaults: | |
| run: | |
| shell: bash | |
| on: | |
| workflow_dispatch: | |
| inputs: | |
| action: | |
| description: Action | |
| required: true | |
| type: choice | |
| options: | |
| - start | |
| - terminate all | |
| size: | |
| description: Instance | |
| required: true | |
| type: choice | |
| # All numbers are for spot instances in East US as of 2023-06-03. | |
| # Standard_HB120rs_v3: | |
| # - https://learn.microsoft.com/en-us/azure/virtual-machines/hbv3-series | |
| # - 120 AMD EPYC™ 7V73X (Milan-X) CPU cores | |
| # - 448 GB of RAM | |
| # - 2 * 960 GB SSD | |
| # - $0.37/hour | |
| # Standard_NC8as_T4_v3: | |
| # - https://learn.microsoft.com/en-us/azure/virtual-machines/nct4-v3-series | |
| # - 1x NVIDIA Tesla T4 GPU with 16 GB of RAM | |
| # - 8 AMD EPYC 7V12 (Rome) CPU cores | |
| # - 56 GB of RAM | |
| # - 1 * 360 GB SSD | |
| # - $0.14/hour | |
| # Standard_NV36ads_A10_v5: | |
| # - https://learn.microsoft.com/en-us/azure/virtual-machines/nva10v5-series | |
| # - 1x NVIDIA A10 GPU with 24 GB of RAM | |
| # - 36 AMD EPYC 74F3V(Milan) CPU cores | |
| # - 440 GB of RAM | |
| # - 1 * 1440 GB SSD | |
| # - $1.28/hour | |
| options: | |
| - Standard_HB120rs_v3 | |
| # - Standard_NC8as_T4_v3 | |
| # - Standard_NV36ads_A10_v5 | |
| jobs: | |
| github-check-for-runner: | |
| name: Check for runner (GitHub) | |
| runs-on: ubuntu-latest | |
| if: ${{ inputs.action == 'start' }} | |
| outputs: | |
| # NOTE: Returns the ID of the matching runner, or -1 if no matching runner is found. | |
| id: ${{ steps.id.outputs.value }} | |
| steps: | |
| - name: Query GitHub for runners | |
| id: runners | |
| env: | |
| # We require admin read access to the repository to query the runners. | |
| # TODO: Expires on Thu, Jun 13 2024. | |
| GH_TOKEN: ${{ secrets.NIX_CUDA_TEST_GH_TOKEN_ADMIN_READ }} | |
| run: | | |
| gh api \ | |
| -H "Accept: application/vnd.github+json" \ | |
| -H "X-GitHub-Api-Version: 2022-11-28" \ | |
| /repos/ConnorBaker/nix-cuda-test/actions/runners \ | |
| | jq -crS '.runners' \ | |
| | xargs -0 printf "json=%s" \ | |
| | tee -a "$GITHUB_OUTPUT" | |
| - name: Filter the runners by size and status | |
| id: matching-runners | |
| run: | | |
| jq -crS \ | |
| --arg size "${{ inputs.size }}" \ | |
| 'map(select(.name == $size and .status == "online"))' \ | |
| <<< '${{ steps.runners.outputs.json }}' \ | |
| | xargs -0 printf "json=%s" \ | |
| | tee -a "$GITHUB_OUTPUT" | |
| - name: Get the ID of the matching runners | |
| id: id | |
| run: | | |
| declare -ri length=$(jq -crS 'length' <<< '${{ steps.matching-runners.outputs.json }}') | |
| declare -i id=-1 | |
| if (( length == 0 )); then | |
| echo "No matching runner found." | |
| elif (( length == 1 )); then | |
| echo "Matching runner found." | |
| id=$(jq -crS '.[0].id' <<< '${{ steps.matching-runners.outputs.json }}') | |
| else | |
| echo "Multiple matching runners found!" | |
| exit 1 | |
| fi | |
| echo "value=$id" | tee -a "$GITHUB_OUTPUT" | |
| azure-check-for-instance: | |
| name: Check for instance (Azure) | |
| runs-on: ubuntu-latest | |
| permissions: | |
| contents: read | |
| id-token: write | |
| outputs: | |
| ip: ${{ steps.ip.outputs.value }} | |
| steps: | |
| - name: Log in to Azure | |
| uses: Azure/login@v2 | |
| with: | |
| client-id: ${{ secrets.AZURE_CLIENT_ID }} | |
| subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }} | |
| tenant-id: ${{ secrets.AZURE_TENANT_ID }} | |
| - name: Check for instance | |
| id: instance | |
| uses: azure/CLI@v1 | |
| with: | |
| azcliversion: latest | |
| inlineScript: | | |
| if | |
| az vm show \ | |
| --show-details \ | |
| --resource-group NixCudaTest \ | |
| --name "${{ inputs.size }}" \ | |
| | jq -crS . \ | |
| | xargs -0 printf "json=%s" \ | |
| | tee -a "$GITHUB_OUTPUT" | |
| then | |
| echo "Instance ${{ inputs.size }} exists" | |
| else | |
| echo "Instance ${{ inputs.size }} does not exist" | |
| fi | |
| # IP is "" if the instance does not exist. | |
| - name: Get instance IP | |
| id: ip | |
| run: | | |
| jq -crS \ | |
| '.publicIps' \ | |
| <<< '${{ steps.instance.outputs.json }}' \ | |
| | xargs -0 printf "value=%s" \ | |
| | tee -a "$GITHUB_OUTPUT" | |
| create-running-instance: | |
| name: Create instance | |
| runs-on: ubuntu-latest | |
| needs: | |
| - azure-check-for-instance | |
| - github-check-for-runner | |
| if: ${{ inputs.action == 'start' && needs.github-check-for-runner.outputs.id == -1 || needs.azure-check-for-instance.outputs.ip == 'null' }} | |
| permissions: | |
| contents: read | |
| id-token: write | |
| steps: | |
| - name: Test pre-conditions | |
| run: | | |
| declare -ri id=${{ needs.github-check-for-runner.outputs.id }} | |
| declare -r ip=${{ needs.azure-check-for-instance.outputs.ip }} | |
| declare GH_RUNNER_STATUS | |
| if (( id == -1 )) ; then | |
| GH_RUNNER_STATUS="does not exist" | |
| else | |
| GH_RUNNER_STATUS="exists" | |
| fi | |
| declare AZ_INSTANCE_STATUS | |
| if [[ -z "$ip" ]]; then | |
| AZ_INSTANCE_STATUS="does not exist" | |
| else | |
| AZ_INSTANCE_STATUS="exists" | |
| fi | |
| declare PRECONDITIONS_STATUS | |
| if [[ "$id" == "-1" && -z "$ip" ]]; then | |
| PRECONDITIONS_STATUS="met" | |
| else | |
| PRECONDITIONS_STATUS="not met" | |
| fi | |
| echo "Preconditions $PRECONDITIONS_STATUS: GitHub runner $GH_RUNNER_STATUS and Azure instance $AZ_INSTANCE_STATUS" | |
| if [[ "$PRECONDITIONS_STATUS" != "met" ]]; then | |
| exit 1 | |
| fi | |
| - uses: actions/checkout@v4 | |
| - name: Set up SSH keys | |
| run: | | |
| mkdir -p "$HOME/.ssh" | |
| echo "${{ secrets.AZURE_SSH_PRIVATE_KEY }}" > "$HOME/.ssh/id_rsa" | |
| chmod 600 "$HOME/.ssh/id_rsa" | |
| echo "${{ secrets.AZURE_SSH_PUBLIC_KEY }}" > "$HOME/.ssh/id_rsa.pub" | |
| chmod 644 "$HOME/.ssh/id_rsa.pub" | |
| - name: Log in to Azure | |
| uses: Azure/login@v2 | |
| with: | |
| client-id: ${{ secrets.AZURE_CLIENT_ID }} | |
| subscription-id: ${{ secrets.AZURE_SUBSCRIPTION_ID }} | |
| tenant-id: ${{ secrets.AZURE_TENANT_ID }} | |
| - name: Provision the instance | |
| id: provision | |
| uses: azure/CLI@v1 | |
| with: | |
| azcliversion: latest | |
| # TODO: This likely runs inside a docker image, which is why we need to provision keys twice. | |
| # However, it is odd that accessing GITHUB_WORKSPACE works. | |
| inlineScript: | | |
| declare -rA images=( | |
| [Standard_HB120rs_v3]="Canonical:0001-com-ubuntu-minimal-mantic:minimal-23_10-gen2:latest" | |
| ) | |
| if [[ -z "${images[${{ inputs.size }}]}" ]] ; then | |
| echo "No image available for size: ${{ inputs.size }}" | |
| exit 1 | |
| fi | |
| mkdir -p "$HOME/.ssh" | |
| echo "${{ secrets.AZURE_SSH_PRIVATE_KEY }}" > "$HOME/.ssh/id_rsa" | |
| chmod 600 "$HOME/.ssh/id_rsa" | |
| echo "${{ secrets.AZURE_SSH_PUBLIC_KEY }}" > "$HOME/.ssh/id_rsa.pub" | |
| chmod 644 "$HOME/.ssh/id_rsa.pub" | |
| az vm create \ | |
| --admin-username runner \ | |
| --enable-hibernation false \ | |
| --enable-hotpatching false \ | |
| --eviction-policy Delete \ | |
| --image "Canonical:0001-com-ubuntu-minimal-mantic:minimal-23_10-gen2:latest" \ | |
| --location eastus \ | |
| --max-price 1.0 \ | |
| --name "${{ inputs.size }}" \ | |
| --nic-delete-option Delete \ | |
| --priority Spot \ | |
| --resource-group NixCudaTest \ | |
| --security-type Standard \ | |
| --size "${{ inputs.size }}" \ | |
| --ssh-key-values "$HOME/.ssh/id_rsa.pub" \ | |
| --user-data "$GITHUB_WORKSPACE/ci/user-data-${{ inputs.size }}.sh" \ | |
| | jq -crS . \ | |
| | xargs -0 printf "json=%s" \ | |
| | tee -a "$GITHUB_OUTPUT" | |
| - name: Get the instance IP | |
| id: ip | |
| run: | | |
| declare -r ip=$(jq -crS '.publicIpAddress' <<< '${{ steps.provision.outputs.json }}') | |
| if [[ -z "$ip" ]]; then | |
| echo "Instance creation failed: no IP address" | |
| exit 1 | |
| fi | |
| echo "value=$ip" | tee -a "$GITHUB_OUTPUT" | |
| - name: Add instance to SSH config | |
| run: | | |
| cat >> "$HOME/.ssh/config" <<EOF | |
| Host ${{ inputs.size }} | |
| HostName ${{ steps.ip.outputs.value }} | |
| User runner | |
| IdentityFile ~/.ssh/id_rsa | |
| StrictHostKeyChecking no | |
| UserKnownHostsFile /dev/null | |
| EOF | |
| chmod 644 "$HOME/.ssh/config" | |
| - name: Ensure instance is reachable via ssh | |
| run: timeout 60s bash -c "until ssh ${{ inputs.size }} true; do sleep 15; done" | |
| - name: Generate registration token | |
| env: | |
| # We require admin read access to the repository to query the runners. | |
| # TODO: Expires on Thu, Jun 13 2024. | |
| GH_TOKEN: ${{ secrets.NIX_CUDA_TEST_GH_TOKEN_ADMIN_READ }} | |
| run: | | |
| gh api \ | |
| --method POST \ | |
| -H "Accept: application/vnd.github+json" \ | |
| -H "X-GitHub-Api-Version: 2022-11-28" \ | |
| /repos/ConnorBaker/nix-cuda-test/actions/runners/registration-token \ | |
| | jq -crS '.token' \ | |
| > "$GITHUB_WORKSPACE/.runner-token" | |
| - name: Copy the token to the runner | |
| run: scp "$GITHUB_WORKSPACE/.runner-token" "${{ inputs.size }}:~/.runner-token" | |
| - name: Set up self-hosted GitHub runner | |
| # NOTE: Must execute ./config.sh from the actions-runner directory as it uses relative paths. | |
| run: | | |
| ssh "${{ inputs.size }}" bash -c \ | |
| 'cd "$HOME/actions-runner" \ | |
| && ./config.sh \ | |
| --unattended \ | |
| --url "https://github.com/ConnorBaker/nix-cuda-test" \ | |
| --token "$(cat ~/.runner-token)" \ | |
| --name "${{ inputs.size }}" \ | |
| --labels "${{ inputs.size }}"' | |
| sleep 15 | |
| build-nix-cuda-test: | |
| name: Build nix-cuda-test | |
| runs-on: | |
| # - Standard_HB120rs_v3 | |
| - self-hosted | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - name: Set up Nix | |
| uses: ./.github/actions/setup-nix | |
| - name: Set up Attic | |
| uses: ./.github/actions/setup-attic | |
| with: | |
| cache-key: ${{ secrets.NIX_CUDA_TEST_ATTIC_CACHE_KEY }} | |
| - name: Build nix-cuda-test | |
| run: nix build --print-build-logs --no-link .#nix-cuda-test |