---
name: openBalena tests

on:
  workflow_call:
    inputs:
      environment:
        description: "balenaCloud environment"
        required: true
        type: string
      fleet:
        description: "balenaCloud fleet"
        required: true
        type: string
      dns_tld:
        description: "domain name to use for issuing SSL certificates"
        required: true
        type: string

# 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:
  actions: read
  checks: read
  contents: read
  deployments: read
  id-token: write # AWS GitHub OIDC required: write
  issues: read
  discussions: read
  packages: read
  pages: read
  pull-requests: read
  repository-projects: read
  security-events: read
  statuses: read

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_TYPE: c6a.2xlarge
  AWS_EC2_LAUNCH_TEMPLATE: lt-02e10a4f66261319d
  AWS_EC2_LT_VERSION: 2
  AWS_IAM_USERNAME: balena-tests-iam-User-1GXO3XP12N6LL
  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
  SUBDOMAIN: auto

jobs:
  test:
    runs-on: ["self-hosted", "X64", "distro:jammy"]  # tests require socat v1.7.4
    timeout-minutes: 60
    strategy:
      fail-fast: true

    steps:
    - uses: actions/checkout@b80ff79f1755d06ba70441c368a6fe801f5f3a62
      with:
        # FIXME: remove once balenaBlocks/balenaVirt is a thing
        submodules: true

    - uses: aws-actions/configure-aws-credentials@bd0758102444af2a09b9e47a2c93d0f091c1252d
      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://github.com/pdcastro/ssh-uuid#why
    # https://github.com/pdcastro/ssh-uuid#linux-debian-ubuntu-others
    - name: install additional dependencies
      shell: bash
      run: |
        set -ue
        echo '::notice::install additional dependencies'

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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: (pre)register test device
      id: register-test-device
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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
        balena device register '${{ inputs.fleet }}' --uuid "${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)"

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

        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 device to draft release
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -uae

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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="$(with_backoff balena releases '${{ inputs.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 test device environment
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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 '${{ env.SUBDOMAIN }}.${{ inputs.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.${{ env.SUBDOMAIN }}.${{ inputs.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.${{ env.SUBDOMAIN }}.${{ inputs.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

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

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

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

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

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

        with_backoff balena env add REGISTRY2_TOKEN_AUTH_REALM 'https://api.${{ env.SUBDOMAIN }}.${{ inputs.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.${{ env.SUBDOMAIN }}.${{ inputs.dns_tld }}' \
          --device '${{ steps.register-test-device.outputs.balena_device_uuid }}'

        with_backoff balena env add WEBRESOURCES_S3_HOST 's3.${{ env.SUBDOMAIN }}.${{ inputs.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@${{ env.SUBDOMAIN }}.${{ inputs.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 test device secrets
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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 ephemeral test device
      id: provision-test-device
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

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

        for subnet_id in ${{ env.AWS_VPC_SUBNET_IDS }}; do
            # spot, on-demand
            for market_type in ${{ vars.MARKET_TYPES || 'spot' }}; do
                # https://docs.aws.amazon.com/cli/latest/reference/ec2/run-instances.html
                response="$(aws ec2 run-instances \
                  --launch-template 'LaunchTemplateId=${{ env.AWS_EC2_LAUNCH_TEMPLATE }},Version=${{ env.AWS_EC2_LT_VERSION }}' \
                  --instance-type '${{ env.AWS_EC2_INSTANCE_TYPE }}' \
                  $([[ $market_type =~ spot ]] && echo '--instance-market-options MarketType=spot') \
                  --security-group-ids '${{ env.AWS_VPC_SECURITY_GROUP_IDS }}' \
                  --subnet-id "${subnet_id}" \
                  --associate-public-ip-address \
                  --user-data file://config.json \
                  --tag-specifications "ResourceType=instance,Tags=[{Key=Name,Value=balena-tests},{Key=MarketType,Value=${market_type}},{Key=Owner,Value=${{ env.AWS_IAM_USERNAME }}},{Key=GITHUB_SHA,Value=${GITHUB_SHA}-tests}]" || true)"

                [[ -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}"

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

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

      env:
        AWS_ACCESS_KEY_ID: ${{ env.AWS_ACCESS_KEY_ID }}
        AWS_SECRET_ACCESS_KEY: ${{ env.AWS_SECRET_ACCESS_KEY }}
        AWS_DEFAULT_REGION: ${{ env.AWS_REGION }}

    - name: provision SSH key
      id: provision-ssh-key
      # wait for cloud-config
      # https://github.com/balena-os/cloud-config
      timeout-minutes: 5
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

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

        source src/balena-tests/functions

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

        if ! [[ -e "${HOME}/.ssh/id_rsa" ]]; then
            ssh-keygen -N '' \
              -C "$(balena whoami | grep EMAIL | cut -c11-)" \
              -f "${HOME}/.ssh/id_rsa"
        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

        pgrep ssh-agent || ssh-agent -a "${SSH_AUTH_SOCK}"

        ssh-add "${HOME}/.ssh/id_rsa"

        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"
        done

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

      env:
        SSH_AUTH_SOCK: /tmp/ssh_agent.sock

    - name: wait for application
      timeout-minutes: 10
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

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

        source src/balena-tests/functions

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

        balena whoami && ssh-add -l

        while [[ "$(curl -X POST --silent --retry ${{ env.RETRY }} --fail \
          'https://api.${{ inputs.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"
        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"
        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"
        done

      env:
        SSH_AUTH_SOCK: /tmp/ssh_agent.sock

    # (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 components
      timeout-minutes: 10
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      run: |
        set -ue

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

        source src/balena-tests/functions

        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"
        done

      env:
        SSH_AUTH_SOCK: /tmp/ssh_agent.sock

    - name: SUT&DUT
      if: ${{ github.event_name == 'pull_request' && github.event.action != 'closed'}}
      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 || 'false' }}' =~ on|On|Yes|yes|true|True ]] && set -x

        source src/balena-tests/functions

        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 ${{ env.RETRY }} --fail \
              'https://api.${{ inputs.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"
        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 ${{ env.RETRY }} --fail \
              'https://api.${{ inputs.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"
        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:
        SSH_AUTH_SOCK: /tmp/ssh_agent.sock
        ATTEMPTS: 2

    - name: remove SSH key
      if: always()
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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 balena key rm --yes

        pgrep ssh-agent && (pgrep ssh-agent | xargs kill)

        rm -f /tmp/ssh_agent.sock

    - name: destroy balena test device
      if: always()
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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_ACCESS_KEY_ID: ${{ env.AWS_ACCESS_KEY_ID }}
        AWS_SECRET_ACCESS_KEY: ${{ env.AWS_SECRET_ACCESS_KEY }}
        AWS_DEFAULT_REGION: ${{ env.AWS_REGION }}

    # always destroy test EC2 instances even if the workflow is cancelled
    - name: destroy AWS test device
      if: always()
      run: |
        set -ue

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

        source src/balena-tests/functions

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

        with_backoff aws ec2 describe-instances --filters Name=tag:GITHUB_SHA,Values=${GITHUB_SHA}-tests \
          | jq -r .Reservations[].Instances[].InstanceId \
          | xargs aws ec2 terminate-instances --instance-ids

        stale_instances=$(mktemp)
        aws ec2 describe-instances --filters \
          Name=tag:Name,Values=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_ACCESS_KEY_ID: ${{ env.AWS_ACCESS_KEY_ID }}
        AWS_SECRET_ACCESS_KEY: ${{ env.AWS_SECRET_ACCESS_KEY }}
        AWS_DEFAULT_REGION: ${{ env.AWS_REGION }}

    # 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-dns-records
      if: always()
      continue-on-error: true
      run: |
        set -ue

        [[ '${{ vars.VERBOSE || 'false' }}' =~ 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 }}.${{ env.SUBDOMAIN }}"

            zone_id="$(curl --silent --retry ${{ env.RETRY }} \
              "https://api.cloudflare.com/client/v4/zones?name=${{ inputs.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: true