---
name: openBalena tests

on:
  workflow_call:

# https://docs.github.com/en/actions/security-guides/automatic-token-authentication
# https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#permissions
permissions:
  contents: read
  id-token: "write"  # AWS GitHub OIDC required: write
  packages: read

# https://docs.github.com/en/actions/using-jobs/using-concurrency
concurrency:
  group: ${{ github.workflow }}-${{ github.event.number || github.ref }}
  # cancel jobs in progress for updated PRs, but not merge or tag events
  cancel-in-progress: ${{ github.event.action == 'synchronize' }}

env:
  # Stack ID
  # arn:aws:cloudformation:us-east-1:491725000532:stack/balena-tests-s3-certs/814dea60-404d-11ed-b06f-0a7d458f8ba5
  AWS_S3_CERTS_BUCKET: balena-tests-certs
  # (kvm) nested virtualisation not supported on AWS/EC2 instance types|classes other than X.metal
  AWS_EC2_INSTANCE_TYPES: "r6i.2xlarge r6a.2xlarge r5.2xlarge r5n.2xlarge r5b.2xlarge r5a.2xlarge m5.2xlarge m5n.2xlarge m5a.2xlarge m6i.2xlarge c6a.2xlarge c6i.2xlarge c5n.2xlarge c5.2xlarge c5a.2xlarge"
  AWS_EC2_LAUNCH_TEMPLATE: lt-02e10a4f66261319d
  AWS_IAM_USERNAME: balena-tests-iam-User-1GXO3XP12N6LL
  AWS_LOGS_RETENTION: "30"
  AWS_VPC_SECURITY_GROUP_IDS: sg-057937f4d89d9d51c
  AWS_VPC_SUBNET_IDS: "subnet-02d18a08ea4058574 subnet-0a026eae1df907a09"
  # otherwise it tries to send data to an endpoint provided by a private project
  # https://github.com/balena-io/analytics-backend
  # .. which is not part of openBalena
  BALENARC_NO_ANALYTICS: "1"  # https://github.com/balena-io/balena-cli/blob/master/lib/events.ts#L62-L70
  DEBUG: "0"  # https://github.com/balena-io/balena-cli/issues/2447
  RETRY: "3"

jobs:
  test:
    runs-on: ["self-hosted", "X64", "distro:jammy"]  # balenaOS (balena-public-pki) tests require socat v1.7.4
    timeout-minutes: 60
    strategy:
      fail-fast: false
      matrix:
        target:
          - compose-private-pki
          - balena-public-pki
        include:
          # tests Docker (compose) flow using self-signed PKI
          - target: compose-private-pki
            launch_template_version: ${{ vars.AWS_EC2_LT_VERSION || '6' }}
            # https://docs.renovatebot.com/modules/datasource/aws-machine-image/
            # amiFilter=[{"Name":"owner-id","Values":["099720109477"]},{"Name":"name","Values":["ubuntu/images/hvm-ssd-gp3/ubuntu-noble-24.04-amd64-server-*"]},{"region":"us-east-1"}]
            # currentImageName=unknown
            ami: ami-04b70fa74e45c3917
            subdomain:  ${{ vars.DNS_SUBDOMAIN || 'auto' }}
            dns_tld: ${{ vars.DNS_TLD || 'balena-devices.com' }}

          # .. balenaCloud flow with Let's Encrypt (ACME) PKI
          - target: balena-public-pki
            launch_template_version: ${{ vars.AWS_EC2_LT_VERSION || '6' }}
            # https://docs.renovatebot.com/modules/datasource/aws-machine-image/
            # amiFilter=[{"Name":"owner-id","Values":["491725000532"]},{"Name":"name","Values":["balenaOS-installer-secureboot-*-generic-amd64"]},{"region":"us-east-1"}]
            # currentImageName=unknown
            ami: ami-03a3995797dee84fa
            # https://dash.cloudflare.com/001b3ed2352612aaa068aca1b0022736/balena-devices.com/dns
            subdomain:  ${{ vars.DNS_SUBDOMAIN || 'auto' }}
            dns_tld: ${{ vars.DNS_TLD || 'balena-devices.com' }}
            environment: ${{ vars.BALENARC_BALENA_URL || 'balena-cloud.com' }}
            fleet: ${{ vars.BALENA_FLEET || 'balena/open-balena' }}

    environment:
      name: ${{ matrix.target }}

    steps:
    - uses: actions/checkout@9a9194f87191a7e9055e3e9b95b8cfb13023bb08

    # https://github.com/unfor19/install-aws-cli-action
    - name: Setup awscli
      uses: unfor19/install-aws-cli-action@e8b481e524a99f37fbd39fdc1dcb3341ab091367 # v1

    - uses: aws-actions/configure-aws-credentials@55f725fcb586ca16a5ed5b0d75d464defbfa831b
      with:
        aws-region: ${{ vars.AWS_REGION || 'us-east-1' }}
        role-session-name: github-${{ github.job }}-${{ github.run_id }}-${{ github.run_attempt }}
        # balena-io/environments-bases: aws/balenacloud/ephemeral-tests/balena-tests-iam.yml
        role-to-assume: ${{ vars.AWS_IAM_ROLE }}

    # https://docs.aws.amazon.com/systems-manager/latest/userguide/session-manager-working-with-install-plugin.html#install-plugin-debian
    - name: install session-manager-plugin
      if: matrix.target == 'compose-private-pki'
      run: |
        # shellcheck disable=SC2153
        runner_arch="$(echo "${RUNNER_ARCH}" | tr '[:upper:]' '[:lower:]' | sed 's/x64/64bit/g')"

        session-manager-plugin || (curl -sSfo session-manager-plugin.deb "https://s3.amazonaws.com/session-manager-downloads/plugin/latest/ubuntu_${runner_arch}/session-manager-plugin.deb" \
          && sudo dpkg -i session-manager-plugin.deb \
          && rm -f session-manager-plugin.deb)

    # https://github.com/balena-io-examples/setup-balena-action
    - name: Setup balena CLI
      uses: balena-io-examples/setup-balena-action@main
      with:
        # renovate: datasource=github-releases depName=balena-io/balena-cli
        cli-version: v18.2.17

    # https://github.com/pdcastro/ssh-uuid#why
    # https://github.com/pdcastro/ssh-uuid#linux-debian-ubuntu-others
    - name: install ssh(scp)-uuid
      if: matrix.target == 'balena-public-pki'
      shell: bash
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        mkdir -p "${RUNNER_TEMP}/ssh-uuid"

        wget -q -O "${RUNNER_TEMP}/ssh-uuid/ssh-uuid" https://raw.githubusercontent.com/pdcastro/ssh-uuid/master/ssh-uuid.sh \
          && chmod +x "${RUNNER_TEMP}/ssh-uuid/ssh-uuid" \
          && ln -s "${RUNNER_TEMP}/ssh-uuid/ssh-uuid" "${RUNNER_TEMP}/ssh-uuid/scp-uuid"

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        balena version

        "${RUNNER_TEMP}/ssh-uuid/scp-uuid" --help

        grep -q "${RUNNER_TEMP}/ssh-uuid" "${GITHUB_PATH}" \
          || echo "${RUNNER_TEMP}/ssh-uuid" >> "${GITHUB_PATH}"

    - name: install cloud-init
      if: matrix.target == 'compose-private-pki'
      shell: bash
      run: sudo apt update && sudo apt install -y cloud-init

    - name: generate SSH private key
      id: generate-key-pair
      run: |
        set -ue

        verbose='+x'
        if [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]]; then
            verbose='-x'
        fi
        set ${verbose}

        key_name="${{ matrix.target }}-${GITHUB_RUN_ID}-${GITHUB_RUN_NUMBER}-${GITHUB_RUN_ATTEMPT}"
        echo "key_name=${key_name}" >>"${GITHUB_OUTPUT}"

        set +x
        private_key_material="$(aws ec2 create-key-pair \
          --key-name "${key_name}" | jq -r .KeyMaterial)"

        public_key="$(aws ec2 describe-key-pairs --include-public-key \
          --key-name "${key_name}" | jq -re .KeyPairs[].PublicKey)"

        # https://stackoverflow.com/a/70384422/1559300
        # https://docs.github.com/en/actions/using-workflows/workflow-commands-for-github-actions#masking-a-value-in-log
        while read -r line; do
            echo "::add-mask::${line}"
        done <<< "${private_key_material}"

        ssh_private_key="$(cat << EOF
        ${private_key_material}
        EOF
        )"
        echo "ssh_private_key<<EOF" >>"${GITHUB_OUTPUT}"
        set ${verbose}

        { echo "${ssh_private_key}"; echo "EOF"; } >>"${GITHUB_OUTPUT}"

        echo "ssh_public_key=${public_key}" >> "${GITHUB_OUTPUT}"

      env:
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}

    # https://github.com/webfactory/ssh-agent
    - uses: webfactory/ssh-agent@dc588b651fe13675774614f8e6a936a468676387 # v0.9.0
      with:
        ssh-private-key: ${{ steps.generate-key-pair.outputs.ssh_private_key }}

    - name: (pre)register balenaOS test device
      id: register-test-device
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        balena_device_uuid="$(openssl rand -hex 16)"

        # https://www.balena.io/docs/learn/more/masterclasses/advanced-cli/#52-preregistering-a-device
        with_backoff balena device register '${{ matrix.fleet }}' --uuid "${balena_device_uuid}"

        device_id="$(balena device "${balena_device_uuid}" | grep ^ID: | cut -c20-)"

        # the actual version deployed depends on the AWS EC2/AMI, defined in AWS_EC2_LAUNCH_TEMPLATE
        os_version="$(balena os versions ${{ vars.DEVICE_TYPE || 'generic-amd64' }} | head -n 1)"

        # shellcheck disable=SC2046
        balena config generate \
          --version "${os_version}" \
          --device "${balena_device_uuid}" \
          --network ethernet \
          --appUpdatePollInterval 10 \
          $([[ '${{ vars.DEVELOPMENT_MODE || 'false' }}' =~ true ]] && echo '--dev') \
          --output config.json

        with_backoff balena tag set balena ephemeral-test-device --device "${balena_device_uuid}"

        github_vars=(GITHUB_ACTOR GITHUB_BASE_REF GITHUB_HEAD_REF GITHUB_JOB \
          GITHUB_REF GITHUB_REF_NAME GITHUB_REF_TYPE GITHUB_REPOSITORY \
          GITHUB_REPOSITORY_OWNER GITHUB_RUN_ATTEMPT GITHUB_RUN_ID GITHUB_RUN_NUMBER \
          GITHUB_SHA GITHUB_WORKFLOW RUNNER_ARCH RUNNER_NAME RUNNER_OS)

        for github_var in "${github_vars[@]}"; do
            balena tag set "${github_var}" "${!github_var}" --device "${balena_device_uuid}"
        done

        echo "balena_device_uuid=${balena_device_uuid}" >> "${GITHUB_OUTPUT}"
        echo "balena_device_id=${device_id}" >> "${GITHUB_OUTPUT}"

    # https://github.com/balena-io/balena-cli/issues/1543
    - name: pin balenaOS test device to draft release
      if: matrix.target == 'balena-public-pki'
      run: |
        set -uae

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        pr_id='${{ github.event.pull_request.id }}'
        head_sha='${{ github.event.pull_request.head.sha || github.event.head_commit.id }}'
        release_id="$(with_backoff balena releases '${{ matrix.fleet }}' --json \
          | jq -r --arg pr_id "${pr_id}" --arg head_sha "${head_sha}" '.[]
          | select(.release_tag[].tag_key=="balena-ci-commit-sha")
          | select(.release_tag[].value==$head_sha)
          | select(.release_tag[].tag_key=="balena-ci-id")
          | select(.release_tag[].value==$pr_id).commit')"

        with_backoff balena device pin \
          ${{ steps.register-test-device.outputs.balena_device_uuid }} \
          "${release_id}"

        with_backoff balena device ${{ steps.register-test-device.outputs.balena_device_uuid }}

    - name: configure balenaOS test device environment
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        with_backoff balena env add VERBOSE "${{ vars.VERBOSE || 'false' }}" \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add BALENARC_NO_ANALYTICS '1' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add DNS_TLD '${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add DB_HOST db \
          --service api \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add REDIS_HOST redis:6379 \
          --service api \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'


        # to allow devices running locally to communicate to the local API, we can route
        # to the local Docker network aliases instead of public DNS, since (a) DNS_TLD is
        # guestfwd(ed) in QEMU to a special internal IP 10.0.2.100; (b) is proxied to
        # haproxy network alias on device; and (c) made public with a wildcard DNS record
        # (e.g.)
        #
        # $ dig +short api.auto.balena-devices.com
        # 10.0.2.100
        #

        with_backoff balena env add API_HOST 'api.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        # not used but required for config.json to be valid
        with_backoff balena env add DELTA_HOST 'delta.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add REGISTRY2_HOST 'registry2.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add VPN_HOST 'cloudlink.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add HOST 'api.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --service api \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add TOKEN_AUTH_CERT_ISSUER 'api.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --service api \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add REGISTRY2_TOKEN_AUTH_ISSUER 'api.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --service registry \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add REGISTRY2_TOKEN_AUTH_REALM 'https://api.${{ matrix.subdomain }}.${{ matrix.dns_tld }}/auth/v1/token' \
          --service registry \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add REGISTRY2_S3_REGION_ENDPOINT 's3.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add WEBRESOURCES_S3_HOST 's3.${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        # https://github.com/balena-io/cert-manager/blob/master/entry.sh#L255-L278
        # cert-manager will restore the last wildcard certificate from AWS/S3 to avoid
        # being rate limited by LetsEncrypt/ACME
        with_backoff balena env add AWS_S3_BUCKET '${{ env.AWS_S3_CERTS_BUCKET }}' \
          --service cert-manager \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        # FIXME: still required?
        with_backoff balena env add COMMON_REGION '${{ env.AWS_REGION }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add SUPERUSER_EMAIL 'admin@${{ matrix.subdomain }}.${{ matrix.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add ORG_UNIT openBalena \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'


        # unstable/unsupported functionality
        with_backoff balena env add HIDE_UNVERSIONED_ENDPOINT 'false' \
          --service api \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add RELEASE_ASSETS_TEST 'true' \
          --service sut \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

    - name: configure balenaOS test device secrets
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        # cert-manager requires it to get whoami information for the user
        with_backoff balena env add API_TOKEN '${{ secrets.BALENA_API_KEY }}' \
          --service cert-manager \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        # cert-manager requires is to request wildcard SSL certificate from LetsEncrypt
        with_backoff balena env add CLOUDFLARE_API_TOKEN '${{ secrets.CLOUDFLARE_API_TOKEN }}' \
          --service cert-manager \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        # AWS credentials to backup/restore PKI assets
        with_backoff balena env add AWS_ACCESS_KEY_ID '${{ env.AWS_ACCESS_KEY_ID }}' \
          --service cert-manager \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add AWS_SECRET_ACCESS_KEY '${{ env.AWS_SECRET_ACCESS_KEY }}' \
          --service cert-manager \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

    - name: provision balenaOS ephemeral SUT
      id: balena-sut
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        # shellcheck disable=SC2046,SC2043
        for subnet_id in ${{ env.AWS_VPC_SUBNET_IDS }}; do
            # spot, on-demand
            for market_type in ${{ vars.MARKET_TYPES || 'spot' }}; do
                for instance_type in ${AWS_EC2_INSTANCE_TYPES}; do
                    # https://docs.aws.amazon.com/cli/latest/reference/ec2/run-instances.html
                    response="$(aws ec2 run-instances \
                      $([[ -n '${{ matrix.ami }}' ]] && echo '--image-id  ${{ matrix.ami }}') \
                      --launch-template 'LaunchTemplateId=${{ env.AWS_EC2_LAUNCH_TEMPLATE }},Version=${{ matrix.launch_template_version }}' \
                      --instance-type "${instance_type}" \
                      $([[ $market_type =~ spot ]] && echo '--instance-market-options MarketType=spot') \
                      --security-group-ids '${{ env.AWS_VPC_SECURITY_GROUP_IDS }}' \
                      --subnet-id "${subnet_id}" \
                      --associate-public-ip-address \
                      --user-data file://config.json \
                      --tag-specifications "ResourceType=instance,Tags=[{Key=Name,Value=open-balena-tests},{Key=matrix.target,Value=${{ matrix.target }}},{Key=MarketType,Value=${market_type}},{Key=Owner,Value=${{ env.AWS_IAM_USERNAME }}},{Key=GITHUB_SHA,Value=${GITHUB_SHA}-tests},{Key=GITHUB_RUN_ID,Value=${GITHUB_RUN_ID}-tests},{Key=GITHUB_RUN_NUMBER,Value=${GITHUB_RUN_NUMBER}-tests},{Key=GITHUB_RUN_ATTEMPT,Value=${GITHUB_RUN_ATTEMPT}-tests}]" || true)"

                    [[ -n $response ]] && break
                done
                [[ -n $response ]] && break
            done
            [[ -n $response ]] && break
        done

        [[ -z $response ]] && exit 1

        instance_id="$(echo "${response}" | jq -r '.Instances[].InstanceId')"

        aws ec2 wait instance-running --instance-ids "${instance_id}"
        with_backoff aws ec2 wait instance-status-ok --instance-ids "${instance_id}"

        echo "instance_id=${instance_id}" >> "${GITHUB_OUTPUT}"

      env:
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}

    - name: provision balenaCloud SSH key
      id: provision-ssh-key
      # wait for cloud-config
      # https://github.com/balena-os/cloud-config
      timeout-minutes: 5
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        function cleanup() {
            aws ec2 describe-instances --instance-ids ${{ steps.balena-sut.outputs.instance_id }} \
              | jq -r .Reservations[].Instances[].StateReason
        }
        trap 'cleanup' EXIT

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        if ! [[ -e "${HOME}/.ssh/id_rsa" ]]; then
            echo '${{ steps.generate-key-pair.outputs.ssh_private_key }}' > "${HOME}/.ssh/id_rsa"
            echo '${{ steps.generate-key-pair.outputs.ssh_public_key }}' > "${HOME}/.ssh/id_rsa.pub"
        fi

        echo "::notice::check $(balena keys | wc -l) keys"

        match=''
        for key in $(balena keys | grep -v ID | awk '{print $1}'); do
            fp=$(balena key "${key}" | tail -n 1 | ssh-keygen -E md5 -lf /dev/stdin | awk '{print $2}')
            if [[ $fp =~ $(ssh-keygen -E md5 -lf "${HOME}/.ssh/id_rsa" | awk '{print $2}') ]]; then
                match="${key}"
                break
            fi
        done

        if [[ -z $match ]]; then
            balena key add "${GITHUB_SHA}" "${HOME}/.ssh/id_rsa.pub"
        else
            balena keys
        fi

        while ! [[ "$(ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          cat /mnt/boot/config.json | jq -r .uuid)" =~ ${{ steps.register-test-device.outputs.balena_device_uuid }} ]]; do

            echo "::warning::Still working..."
            sleep "$(( (RANDOM % 5) + 5 ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

        echo "key_id=${GITHUB_SHA}" >> "${GITHUB_OUTPUT}"

    - name: wait for balenaCloud application
      timeout-minutes: 10
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        function cleanup() {
            aws ec2 describe-instances --instance-ids ${{ steps.balena-sut.outputs.instance_id }} \
              | jq -r .Reservations[].Instances[].StateReason
        }
        trap 'cleanup' EXIT

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        balena whoami && ssh-add -l

        while [[ "$(curl -X POST --silent --retry ${{ env.RETRY }} --fail \
          'https://api.${{ matrix.environment }}/supervisor/v1/device' \
          --header 'authorization: Bearer ${{ secrets.BALENA_API_KEY }}' \
          --header 'Content-Type:application/json' \
          --data '{"uuid": "${{ steps.register-test-device.outputs.balena_device_uuid }}", "method": "GET"}' \
          --compressed | jq -r '.update_pending')" =~ ^true$ ]]; do

            sleep "$(( ( RANDOM % ${{ env.RETRY }} ) + ${{ env.RETRY }} ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

        # wait for services to start running
        while with_backoff ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          'balena ps -q | xargs balena inspect | jq -r .[].State.Status' \
          | grep -E 'created|restarting|removing|paused|exited|dead'; do

            echo "::warning::Still working..."
            sleep "$(( (RANDOM % 30) + 30 ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

        # wait for Docker healthchecks
        while with_backoff ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          'balena ps -q | xargs balena inspect \
          | jq -r ".[] | select(.State.Health.Status!=null).Name + \":\" + .State.Health.Status"' \
          | grep -E ':starting|:unhealthy'; do

            echo "::warning::Still working..."
            sleep "$(( (RANDOM % 30) + 30 ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

    # (TBC) https://www.balena.io/docs/reference/supervisor/docker-compose/
    # due to lack of long form depends_on support in compositions, restart to ensure all
    # components are running with the latest configuration; preferred over restart via
    # Supervisor API restart due to potential HTTP [timeouts](https://github.com/balena-os/balena-supervisor/issues/1157)
    - name: restart balenaEngine composition
      timeout-minutes: 10
      if: matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        function cleanup() {
            aws ec2 describe-instances --instance-ids ${{ steps.balena-sut.outputs.instance_id }} \
              | jq -r .Reservations[].Instances[].StateReason
        }
        trap 'cleanup' EXIT

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        balena whoami && ssh-add -l

        with_backoff ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          "balena ps -aq | xargs balena inspect \
          | jq -re '.[]
          | select(.Name | contains(\"_supervisor\") | not).Id' \
          | xargs balena restart"

        # wait for Docker healthchecks
        while with_backoff ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          'balena ps -q | xargs balena inspect \
          | jq -r ".[]
          | select(.State.Health.Status!=null).Name + \":\" + .State.Health.Status"' \
          | grep -E ':starting|:unhealthy'; do

            echo "::warning::Still working..."
            sleep "$(( (RANDOM % 30) + 30 ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

    - name: SUT&DUT (balena)
      if: matrix.target == 'balena-public-pki'
      timeout-minutes: 20
      # https://giters.com/gfx/example-github-actions-with-tty
      # https://github.com/actions/runner/issues/241#issuecomment-924327172
      shell: 'script -q -e -c "bash {0}"'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        function cleanup() {
            aws ec2 describe-instances --instance-ids ${{ steps.balena-sut.outputs.instance_id }} \
              | jq -r .Reservations[].Instances[].StateReason
        }
        trap 'cleanup' EXIT

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        balena whoami && ssh-add -l

        (with_backoff ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          "balena ps -aq | xargs balena inspect \
          | jq -re '.[] | select(.Name | contains(\"sut_\")).Id' \
          | xargs balena logs -f") &

        # tests service is working while its status == running
        status=''
        while [[ "$status" =~ Running ]]; do
            status="$(curl --silent --retry ${{ env.RETRY }} --fail \
              'https://api.${{ matrix.environment }}/supervisor/v2/applications/state' \
              --header 'authorization: Bearer ${{ secrets.BALENA_API_KEY }}' \
              --header 'Content-Type:application/json' \
              --data '{"uuid": "${{ steps.register-test-device.outputs.balena_device_uuid }}", "method": "GET"}' \
              --compressed | jq -r '.[].services.sut.status')"

            echo "::warning::Still working..."
            sleep "$(( ( RANDOM % ${{ env.RETRY }} ) + ${{ env.RETRY }} ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

        # .. once the service exits with status == exited, it is assumed to be finished
        status=''
        while ! [[ "$status" =~ exited ]]; do
            echo "::warning::Still working..."
            status="$(curl --silent --retry ${{ env.RETRY }} --fail \
              'https://api.${{ matrix.environment }}/supervisor/v2/applications/state' \
              --header 'authorization: Bearer ${{ secrets.BALENA_API_KEY }}' \
              --header 'Content-Type:application/json' \
              --data '{"uuid": "${{ steps.register-test-device.outputs.balena_device_uuid }}", "method": "GET"}' \
              --compressed | jq -r '.[].services.sut.status')"

            sleep "$(( ( RANDOM % ${{ env.RETRY }} ) + ${{ env.RETRY }} ))s"

            aws ec2 wait instance-running --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.balena-sut.outputs.instance_id }} || exit 1
        done

        # .. check its exit code
        expected_exit_code=0
        actual_exit_code="$(with_backoff ssh-uuid -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null \
          ${{ steps.register-test-device.outputs.balena_device_uuid }}.balena \
          "balena ps -aq | xargs balena inspect \
          | jq -re '.[] | select(.Name | contains(\"sut_\")).State.ExitCode'")"

        [[ $expected_exit_code -eq $actual_exit_code ]] || false

      env:
        ATTEMPTS: 2

    - name: provision Ubuntu ephemeral SUT
      id: ubuntu-sut
      timeout-minutes: 20
      if: matrix.target == 'compose-private-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        function cleanup() {
           rm -f user-data.yml
        }
        trap 'cleanup' EXIT

        aws sts get-caller-identity

        # https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/user-data.html
        # https://cloudinit.readthedocs.io/en/latest/reference/modules.html#update-etc-hosts
        cat <<EOF >user-data.yml
        #cloud-config
        output : { all : '| tee -a /var/log/cloud-init-output.log' }
        # https://cloudinit.readthedocs.io/en/latest/reference/modules.html#update-etc-hosts
        manage_etc_hosts: localhost

        packages:
          - git
          - jq
          - wget

        write_files:
          - path: /root/.env
            permissions: "0766"
            content: |
              DNS_TLD=${{ matrix.subdomain }}.${{ matrix.dns_tld }}
              PRODUCTION_MODE=false
              VERBOSE=${{ vars.VERBOSE }}

          - path: /root/functions
            permissions: "0777"
            content: |
              # https://coderwall.com/p/--eiqg/exponential-backoff-in-bash
              function with_backoff() {
                  local max_attempts=\${ATTEMPTS-5}
                  local timeout=\${TIMEOUT-1}
                  local attempt=0
                  local exitCode=0

                  set +e
                  while [[ \$attempt < \$max_attempts ]]
                  do
                      "\$@"
                      exitCode=\$?

                      if [[ \$exitCode == 0 ]]
                      then
                        break
                      fi

                      echo "Failure! Retrying in \$timeout.." 1>&2
                      sleep "\$timeout"
                      attempt=\$(( attempt + 1 ))
                      timeout=\$(( timeout * 2 ))
                  done

                  if [[ \$exitCode != 0 ]]
                  then
                      echo "You've failed me for the last time! (\$*)" 1>&2
                  fi
                  set -e
                  return \$exitCode
              }

          # docs/getting-started.md
          - path: /root/getting-started.sh
            permissions: "0777"
            content: |
              #!/usr/bin/env bash

              set -ax

              [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

              source /root/functions

              apt-get update
              which openssl || apt-get install -y make openssl
              which git || apt-get install -y make git
              which jq || apt-get install -y make jq
              which make || apt-get install make

              which yq || with_backoff wget -q https://github.com/mikefarah/yq/releases/latest/download/yq_linux_amd64 -O /usr/bin/yq
              chmod +x /usr/bin/yq
              yq --version

              which docker || curl -fsSL https://get.docker.com | sh -
              usermod -aG docker ubuntu
              systemctl enable docker && systemctl start docker
              chown ubuntu:docker /var/run/docker.sock

              id -u balena || useradd -s /bin/bash -m -G docker,sudo balena
              echo 'balena ALL=(ALL) NOPASSWD: ALL' > /etc/sudoers.d/balena

              while ! docker ps; do sleep \$(((RANDOM%3)+1)); done
              with_backoff docker login \
                --username='${{ secrets.DOCKERHUB_USER }}' \
                --password='${{ secrets.DOCKERHUB_TOKEN }}'

              with_backoff docker login ghcr.io \
                --username=token \
                --password=${{ secrets.GITHUB_TOKEN }}

              if [ ! -f /sys/fs/cgroup/cgroup.controllers ]; then
                  echo "cgroups v2 is disabled"
              else
                  echo "cgroups v2 is enabled"
                  source /etc/default/grub
                  sed -i '/GRUB_CMDLINE_LINUX/d' /etc/default/grub
                  echo GRUB_CMDLINE_LINUX=\$(printf '\"%s systemd.unified_cgroup_hierarchy=0\"\n' "\${GRUB_CMDLINE_LINUX}") > /etc/default/grub
                  update-grub
                  reboot
              fi

              tmphosts="\$(mktemp)"
              cat </etc/hosts | sed "s/\$(hostname)\$/\$(hostname) api.${{ matrix.subdomain }}.${{ matrix.dns_tld }}/g" >"\${tmphosts}" \
                && cat <"\${tmphosts}" >/etc/hosts \
                && rm -f "\${tmphosts}" \
                && getent hosts api.${{ matrix.subdomain }}.${{ matrix.dns_tld }} | grep 127.0.1.1

        # cloud-init runs as root
        # (e.g.) https://cloudinit.readthedocs.io/en/latest/reference/merging.html#example-cloud-config
        runcmd:
          - '/root/getting-started.sh'  # FIXME: this may run before the script is written
        EOF

        cloud-init schema -c user-data.yml

        # shellcheck disable=SC2046,SC2043
        for subnet_id in ${{ env.AWS_VPC_SUBNET_IDS }}; do
            # spot, on-demand
            for market_type in ${{ vars.MARKET_TYPES || 'spot' }}; do
                for instance_type in ${AWS_EC2_INSTANCE_TYPES}; do
                    # https://docs.aws.amazon.com/cli/latest/reference/ec2/run-instances.html
                    response="$(aws ec2 run-instances \
                      $([[ -n '${{ matrix.ami }}' ]] && echo '--image-id ${{ matrix.ami }}') \
                      --launch-template 'LaunchTemplateId=${{ env.AWS_EC2_LAUNCH_TEMPLATE }},Version=${{ matrix.launch_template_version }}' \
                      --instance-type "${instance_type}" \
                      $([[ "$market_type" =~ spot ]] && echo '--instance-market-options MarketType=spot') \
                      --security-group-ids '${{ env.AWS_VPC_SECURITY_GROUP_IDS }}' \
                      --subnet-id "${subnet_id}" \
                      --key-name '${{ steps.generate-key-pair.outputs.key_name }}' \
                      --associate-public-ip-address \
                      --user-data file://user-data.yml \
                      --tag-specifications "ResourceType=instance,Tags=[{Key=Name,Value=open-balena-tests},{Key=matrix.target,Value=${{ matrix.target }}},{Key=MarketType,Value=${market_type}},{Key=Owner,Value=${{ env.AWS_IAM_USERNAME }}},{Key=GITHUB_SHA,Value=${GITHUB_SHA}-tests},{Key=GITHUB_RUN_ID,Value=${GITHUB_RUN_ID}-tests},{Key=GITHUB_RUN_NUMBER,Value=${GITHUB_RUN_NUMBER}-tests},{Key=GITHUB_RUN_ATTEMPT,Value=${GITHUB_RUN_ATTEMPT}-tests}]" || true)"

                    [[ -n $response ]] && break
                done
                [[ -n $response ]] && break
            done
            [[ -n $response ]] && break
        done

        [[ -z $response ]] && exit 1

        instance_id="$(echo "${response}" | jq -r '.Instances[].InstanceId')"
        echo "instance_id=${instance_id}" >>"${GITHUB_OUTPUT}"

        aws ec2 wait instance-running --instance-ids "${instance_id}"
        with_backoff aws ec2 wait instance-status-ok --instance-ids "${instance_id}"

      env:
        ATTEMPTS: 2
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}
        COMMIT: ${{ github.event.pull_request.head.sha || github.event.head_commit.id || github.event.pull_request.head.ref }}

    - name: SUT&DUT (Ubuntu/compose)
      if: matrix.target == 'compose-private-pki'
      timeout-minutes: 30
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        function log_output() {
            rm -f "{HOME}/.ssh/config"

            aws ssm list-command-invocations \
              --details \
              --output text \
              --command-id "${cid}" || true

            aws logs describe-log-streams \
              --log-group-name open-balena-tests \
              --log-stream-name-prefix "${cid}" || true

            aws logs put-retention-policy \
              --log-group-name open-balena-tests \
              --retention-in-days "${{ env.AWS_LOGS_RETENTION }}" || true

            aws ec2 describe-instances --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} \
              | jq -r .Reservations[].Instances[].StateReason
        }
        trap 'log_output' EXIT

        # https://docs.aws.amazon.com/systems-manager/latest/userguide/session-manager-getting-started-enable-ssh-connections.html
        cat << EOF > "${HOME}/.ssh/config"
        host i-*
            StrictHostKeyChecking no
            UserKnownHostsFile /dev/null
            ProxyCommand sh -c "aws ssm start-session --target %h --document-name AWS-StartSSHSession --parameters 'portNumber=%p'"
        EOF

        # docs/getting-started.md
        cmds="set -ax \
          && cloud-init status --wait --long && cat </var/log/cloud-init-output.log \
          && sudo -u balena git clone https://token:${{ secrets.GITHUB_TOKEN }}@github.com/${{ github.repository }}.git /home/balena/open-balena \
          && sudo -u balena git config --global --add safe.directory /home/balena/open-balena \
          && cd /home/balena/open-balena \
          && sudo -u balena git checkout ${COMMIT} \
          && . /root/.env \
          && sudo -u balena touch .env \
          && docker compose version \
          && docker compose config --no-interpolate >${COMMIT} \
          && cat <${COMMIT} >docker-compose.yml \
          && sudo -u balena --preserve-env=DNS_TLD\,VERBOSE\,PRODUCTION_MODE make up \
          && sudo -u balena make self-signed \
          && sudo -u balena make verify \
          && sudo -u balena make restart \
          && docker compose wait dut"

        # AWS-RunShellScript runs as root
        result="$(aws ssm send-command \
          --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} \
          --document-name AWS-RunShellScript \
          --comment "open-balena-tests@${{ matrix.target }}" \
          --parameters commands=["${cmds}"] \
          --cloud-watch-output-config '{"CloudWatchLogGroupName":"open-balena-tests","CloudWatchOutputEnabled":true}')"

        echo "${result}" | jq -re
        cid="$(echo "${result}" | jq -r .Command.CommandId)"
        iid="$(echo "${result}" | jq -r .Command.InstanceIds[0])"
        { [[ -n "$cid" ]] && [[ -n "$iid" ]]; } || false

        # https://docs.github.com/en/actions/using-workflows/workflow-commands-for-github-actions#grouping-log-lines
        CYAN='\033[0;36m'; NC='\033[0m'; echo -e "::group::${CYAN}open-balena-tests${NC}"
        until [[ "$(aws ssm list-command-invocations --command-id "${cid}" \
          | jq -re '.CommandInvocations[].Status')" =~ InProgress ]]; do
            echo '::info::starting...'
            sleep $(((RANDOM%5) + 5))s
        done
        echo '::info::command started'

        while [[ $(aws logs describe-log-streams \
          --log-group-name open-balena-tests \
          --log-stream-name-prefix "${cid}" | jq -r '.logStreams|length') -le 0 ]]; do
            echo '::info::waiting for logs...'
            sleep $(((RANDOM%5) + 5))s
        done
        echo '::info::logs started'

        until [[ "$(docker compose ls --format json | jq -re '.[] | select(.Status | startswith("running")).Name')" =~ open-balena ]]; do
            echo '::info::waiting for composition...'
            with_backoff docker compose ls
            sleep $(((RANDOM%5) + 5))s

            aws ec2 wait instance-running --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} || exit 1
        done
        echo '::info::composition started'

        touch .env
        for service in sut dut; do
        until [[ "$(docker compose ps --services "${service}" --status running)" =~ ${service} ]]; do
            echo "::info::waiting for ${service}..."
            with_backoff docker compose ps
            sleep $(((RANDOM%5) + 5))s

            aws ec2 wait instance-running --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} || exit 1
        done
        echo "::info::${service} started"
        done

        echo '::info::settling down...'
        sleep $(((RANDOM%15) + 15))s

        while [[ "$(aws ssm list-command-invocations --command-id "${cid}" \
          | jq -re '.CommandInvocations[].Status')" =~ InProgress ]]; do
            with_backoff docker compose ls && with_backoff docker compose ps
            with_backoff docker compose logs --follow --timestamps sut
            echo '::info::still running...'
            sleep $(((RANDOM%1) + 1))s

            aws ec2 wait instance-running --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} || exit 1
            aws ec2 wait instance-status-ok --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }} || exit 1
        done

        aws ssm wait command-executed --command-id "${cid}" --instance-id "${iid}"
        echo '::info::command finished'
        echo "::endgroup::"

        if ! [[ "$(aws ssm list-command-invocations --command-id "${cid}" \
          | jq -r '.CommandInvocations[].Status')" =~ Success ]]; then
            false
        fi

      env:
        ATTEMPTS: 2
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}
        DOCKER_HOST: ssh://ubuntu@${{ steps.ubuntu-sut.outputs.instance_id }}:22
        COMMIT: ${{ github.event.pull_request.head.sha || github.event.head_commit.id || github.event.pull_request.head.ref }}

    - name: remove balenaCloud SSH key
      if: always() && matrix.target == 'balena-public-pki'
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        with_backoff balena keys | grep ${{ steps.provision-ssh-key.outputs.key_id }} \
          | awk '{print $1}' | xargs --no-run-if-empty balena key rm --yes

    - name: remove AWS/EC2 key-pair
      if: always() && matrix.target == 'compose-private-pki'
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        aws ec2 delete-key-pair --key-name ${{ steps.generate-key-pair.outputs.key_name }}

    - name: delete balenaOS test device
      if: always() && matrix.target == 'balena-public-pki'
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        with_backoff balena login --token '${{ secrets.BALENA_API_KEY }}'

        with_backoff balena device rm ${{ steps.register-test-device.outputs.balena_device_uuid }} --yes

      env:
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}

    # always destroy test EC2 instances even if the workflow is cancelled
    - name: destroy AWS test device(s)
      if: always() && matrix.target == 'balena-public-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        if [[ -n '${{ steps.balena-sut.outputs.instance_id }}' ]]; then
            with_backoff aws ec2 terminate-instances \
              --instance-ids ${{ steps.balena-sut.outputs.instance_id }}
        fi

      env:
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}

    # always destroy test EC2 instances even if the workflow is cancelled
    - name: destroy AWS test device(s)
      if: always() && matrix.target == 'compose-private-pki'
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        if [[ -n '${{ steps.ubuntu-sut.outputs.instance_id }}' ]]; then
            with_backoff aws ec2 terminate-instances \
              --instance-ids ${{ steps.ubuntu-sut.outputs.instance_id }}
        fi

      env:
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}

    # always destroy stale test EC2 instances
    - name: destroy stale AWS test device(s)
      if: always()
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        stale_instances=$(mktemp)
        aws ec2 describe-instances --filters \
          Name=tag:Name,Values=open-balena-tests \
          Name=instance-state-name,Values=running \
          | jq -re '.Reservations[].Instances[].InstanceId + " " + .Reservations[].Instances[].LaunchTime' > "${stale_instances}" || true

        if test -s "${stale_instances}"; then
            while IFS= read -r line; do
                instance_id="$(echo "${line}" | awk '{print $1}')"
                launch_time="$(echo "${line}" | awk '{print $2}')"
                now="$(date +%s)"
                then="$(date --date "${launch_time}" +%s)"
                days_since_launch="$(( (now - then) / 86400 ))"
                if [[ -n "$days_since_launch" ]] && [[ $days_since_launch -ge 1 ]]; then
                    with_backoff aws ec2 terminate-instances --instance-ids "${instance_id}"
                fi
            done <"${stale_instances}"
            rm -f "${stale_instances}"
        fi

      env:
        AWS_DEFAULT_REGION: ${{ vars.AWS_REGION || 'us-east-1' }}

    # remove orphaned ACME DNS-01 validation records
    # https://letsencrypt.org/docs/challenge-types/#dns-01-challenge
    # FIXME: clean up older _acme-challenge.auto TXT records
    - name: cleanup Cloudflare DNS
      if: always() && matrix.target == 'balena-public-pki'
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE }}' =~ on|On|Yes|yes|true|True ]] && set -x

        if [[ -n '${{ steps.register-test-device.outputs.balena_device_uuid }}' ]]; then
            match="${{ steps.register-test-device.outputs.balena_device_uuid }}.${{ matrix.subdomain }}"

            zone_id="$(curl --silent --retry ${{ env.RETRY }} \
              "https://api.cloudflare.com/client/v4/zones?name=${{ matrix.dns_tld }}" \
              -H 'Authorization: Bearer ${{ secrets.CLOUDFLARE_API_TOKEN }}' | jq -r '.result[].id')"

            for record in $(curl --silent --retry ${{ env.RETRY }} \
              "https://api.cloudflare.com/client/v4/zones/${zone_id}/dns_records" \
              -H 'Authorization: Bearer ${{ secrets.CLOUDFLARE_API_TOKEN }}' \
              | jq -r --arg match "${match}" '.result[] | select(((.type=="TXT") and (.name | contains($match))))' \
              | base64); do

                json="$(echo "${record}" | base64 -d | jq -r)"
                id="$(echo "${json}" | jq -r .id)"
                name="$(echo "${json}" | jq -r .name)"

                if [[ -n $id ]] && [[ -n $name ]]; then
                    echo "::warning::Orphaned DNS record ${name} (${id})..."

                    if [[ -z $DRY_RUN ]]; then
                        curl -X DELETE --silent --retry ${{ env.RETRY }} \
                          "https://api.cloudflare.com/client/v4/zones/${zone_id}/dns_records/${id}" \
                          -H 'Authorization: Bearer ${{ secrets.CLOUDFLARE_API_TOKEN }}'
                    fi
                fi
            done
        fi

      env:
        DRY_RUN: false