prplos/.gitlab/testbed.yml

.testbed:
  stage: run
  image: "$CI_REGISTRY_IMAGE/$CI_DESIGNATED_BRANCH_SLUG/testbed:latest"
  timeout: 63 minutes
  rules:
    - if: '$CI_PIPELINE_SOURCE == "schedule"'
      when: never
    - when: on_success
  variables:
    DUT_WAN_INTERFACE: eth1
    TESTBED_MNG_INTERFACE: enp1s0
    TESTBED_LAN_INTERFACE: enp2s0
    TESTBED_WAN_INTERFACE: enp3s0
    TESTBED_TFTP_PATH: /var/lib/tftpboot
    TFTP_IMAGE_PATH: bin/targets/$DUT_TARGET/$DUT_SUBTARGET
    PACKAGE_DIR: bin/packages
    DUT_ARCH_PACKAGES_PATH: $PACKAGE_DIR/${DUT_ARCH}_${DUT_CPU_TYPE}_${DUT_CPU_SUBTYPE}
    TARGET_LAN_IP: 192.168.1.1
    TARGET_LAN_TEST_HOST: 192.168.1.2
    DUT_SLEEP_AFTER_BOOT: 90
    DUT_SSH_CONNECT_TIMEOUT: 30
    DUT_SSH_CONNECT_ATTEMPTS: 3
    DUT_TR181_WAN_INTERFACE: Device.Ethernet.Link.2.
    DUT_TR181_LAN_INTERFACE: Device.Ethernet.Link.3.
    TEST_SETUP_TEARDOWN_DURATION: 360
    CRAM_REMOTE_COMMAND: ssh root@$TARGET_LAN_IP
    CRAM_REMOTE_COPY: scp -o StrictHostKeyChecking=no -o UserKnownHostsFile=/dev/null
    CRAM_TEST_SUITE: |
      .gitlab/tests/cram/generic/acceleration-plan-components
      .gitlab/tests/cram/$DUT_BOARD/acceleration-plan-components
      .gitlab/tests/cram/generic
      .gitlab/tests/cram/$DUT_BOARD      
    CRAM_TEST_SUITE_POST: |
      .gitlab/tests/cram/post/generic      

  before_script:
    - source .gitlab/scripts/helpers.sh
    - section_start "testbed-setup" "Testbed Setup" "collapsed"

    - touch .run_failed

    - sudo ip link set $TESTBED_LAN_INTERFACE up 2> /dev/null
    - sudo ip link set $TESTBED_WAN_INTERFACE up 2> /dev/null
    - sleep 10

    - mkdir -p ~/.ssh; chmod 700 ~/.ssh
    - >
      if echo "$CI_RUNNER_DESCRIPTION" | grep -q testbed-01; then
        eval $(ssh-agent -s)
        echo "$TESTBED_SSH_PRIVATE_KEY" | tr -d '\r' | ssh-add -
        ssh-keyscan $TESTBED_UART_RELAY_HOST > ~/.ssh/known_hosts 2> /dev/null
        chmod 644 ~/.ssh/known_hosts
      fi      

    - |
      echo "Using firmware image $TFTP_IMAGE_PATH/$TFTP_IMAGE_FILENAME"
      cp "$TFTP_IMAGE_PATH/$TFTP_IMAGE_FILENAME" "$TESTBED_TFTP_PATH"

      if [ -n "$TFTP_IMAGE_UNPACK_COMMAND" ]; then
        echo "Running $TFTP_IMAGE_UNPACK_COMMAND"
        eval "$TFTP_IMAGE_UNPACK_COMMAND"
      fi      

    - >
      if echo "$CI_RUNNER_DESCRIPTION" | grep -q testbed-02; then
        export TB_CONFIG=".testbed/labgrid/testbed-02.yaml"
      fi      

    - section_end "testbed-setup"
    - section_start "dut-bringup" "DUT Bringup" "collapsed"

    - .gitlab/scripts/testbed-device.py --target $LABGRID_TARGET boot_into shell
    - >
      .gitlab/scripts/testbed-device.py
      --target $LABGRID_TARGET check_network
      --network lan
      --remote-host $TARGET_LAN_TEST_HOST      

    - sleep $DUT_SLEEP_AFTER_BOOT

    - section_end "dut-bringup"
    - section_start "dut-system-info" "DUT System Information" "collapsed"

    - echo "Waiting for SSH connection availability with ${DUT_SSH_CONNECT_TIMEOUT}sec timeout, with $DUT_SSH_CONNECT_ATTEMPTS attempts..."
    - time ssh -o BatchMode=yes -o StrictHostKeyChecking=no -o ConnectionAttempts=$DUT_SSH_CONNECT_ATTEMPTS -o ConnectTimeout=$DUT_SSH_CONNECT_TIMEOUT root@$TARGET_LAN_IP 'exit 0'

    - ssh-keyscan $TARGET_LAN_IP >> ~/.ssh/known_hosts 2> /dev/null
    - ssh root@$TARGET_LAN_IP 'logger -t CI -p local0.info "System accessible over SSH, $(uptime -p)"' || true
    - ssh root@$TARGET_LAN_IP "cat /var/log/messages" > syslog-$LABGRID_TARGET.txt 2> /dev/null || true
    - ssh root@$TARGET_LAN_IP "ubus call system board" | tee system-$LABGRID_TARGET.json

    - section_end "dut-system-info"

  script:
    - source .gitlab/scripts/helpers.sh
    - section_start "test-execution" "Test Execution" "expanded"

    - |
      TEST_CUTOFF_TIMEOUT=$((CI_JOB_TIMEOUT - DUT_SLEEP_AFTER_BOOT - TEST_SETUP_TEARDOWN_DURATION))
      echo "Job timeout: $CI_JOB_TIMEOUT seconds, test cutoff timeout: $TEST_CUTOFF_TIMEOUT seconds"      

    - |
      rm -fr .run_check_failed

      run_check() {
        "$@" || {
          local statuses=("${PIPESTATUS[@]}")
          for status in "${statuses[@]}"; do
            if [ "$status" -ne 0 ]; then
              log_error "run_check: $* | FAILED! (status: $status)" | tee -a .run_check_failed
            fi
          done
        }
      }

      run_check ssh root@$TARGET_LAN_IP 'f=$(logread | grep "procd.*init complete"); echo "${f:-Warning: init is not complete yet,} $(uptime -p)"'
      run_check ssh root@$TARGET_LAN_IP 'logger -t CI -p local0.info "Testing starts, $(uptime -p)"'
      run_check \
        timeout --kill-after=10s --verbose ${TEST_CUTOFF_TIMEOUT}s \
        python3 -m cram --xunit-file=cram-result-$LABGRID_TARGET.xunit.xml --verbose $CRAM_TEST_SUITE $CRAM_TEST_SUITE_EXTRA | tee cram-result-$LABGRID_TARGET.txt
      run_check python3 -m cram --xunit-file=cram-result-post-$LABGRID_TARGET.xunit.xml --verbose $CRAM_TEST_SUITE_POST | tee --append cram-result-$LABGRID_TARGET.txt
      run_check ssh root@$TARGET_LAN_IP 'logger -t CI -p local0.info "Testing finished, $(uptime -p)"'      

    - section_end "test-execution"

    - |
      [ -f .run_check_failed ] && cat .run_check_failed && exit 1
      rm .run_failed      

  after_script:
    - >-
      curl --fail \
        --output .gitlab/scripts/cram-xunit-merger.py \
        https://gitlab.com/prpl-foundation/tooling/cram-xunit-merger/-/raw/9bc14318ab88c2a024ad0ff8bf09c21d048e1e3b/cram-xunit-merger.py      

    - python3 .gitlab/scripts/cram-xunit-merger.py \
      --output "${CI_PROJECT_DIR}/cram-result-$LABGRID_TARGET.junit.xml" \
      "${CI_PROJECT_DIR}/cram-result-$LABGRID_TARGET.xunit.xml" \
      "${CI_PROJECT_DIR}/cram-result-post-$LABGRID_TARGET.xunit.xml" || true

    - >
      [ "$TMATE_ENABLE_SHELL" = "YES" ] && {
        echo "Starting tmate session..."
        tmate -n "${TMATE_SESSION_TOKEN}-$CI_JOB_ID" -k "$TMATE_API_KEY" -F new "ssh root@$TARGET_LAN_IP" > /tmp/tmate-session.log 2>&1
        true
      }      

    - mkdir -p ~/.ssh; chmod 700 ~/.ssh
    - >
      if echo "$CI_RUNNER_DESCRIPTION" | grep -q testbed-01; then
        eval $(ssh-agent -s)
        echo "$TESTBED_SSH_PRIVATE_KEY" | tr -d '\r' | ssh-add -
        ssh-keyscan $TESTBED_UART_RELAY_HOST > ~/.ssh/known_hosts 2> /dev/null
        chmod 644 ~/.ssh/known_hosts
      fi      

    - >
      if echo "$CI_RUNNER_DESCRIPTION" | grep -q testbed-02; then
        export TB_CONFIG=".testbed/labgrid/testbed-02.yaml"
      fi      

    - >
      ssh -o BatchMode=yes -o StrictHostKeyChecking=no -o ConnectTimeout=15 root@$TARGET_LAN_IP 'exit 0' || {
        .gitlab/scripts/testbed-device.py --target $LABGRID_TARGET console_dump_system_state
        .gitlab/scripts/testbed-device.py --target $LABGRID_TARGET console_recover_ssh_access
      } || true      

    - ssh-keyscan $TARGET_LAN_IP >> ~/.ssh/known_hosts 2> /dev/null
    - >
      ssh root@$TARGET_LAN_IP exit && {
        ssh root@$TARGET_LAN_IP ps > processes-$LABGRID_TARGET.txt
        ssh root@$TARGET_LAN_IP dmesg > dmesg-$LABGRID_TARGET.txt
        ssh root@$TARGET_LAN_IP "cat /var/log/messages" > syslog-$LABGRID_TARGET.txt
        ssh root@$TARGET_LAN_IP opkg list > opkg_list-$LABGRID_TARGET.txt
        script --quiet --command "ssh -t root@$TARGET_LAN_IP /bin/getDebugInformation --all"
        scp -r root@${TARGET_LAN_IP}:/tmp/debug-*_all.tar.gz . > /dev/null
        scp -r root@${TARGET_LAN_IP}:/etc etc > /dev/null
        scp -r root@${TARGET_LAN_IP}:/tmp/beerocks/logs prplmesh_beerocks_logs > /dev/null
      } || true      

    - .gitlab/scripts/testbed-device.py --target $LABGRID_TARGET power off
    - tar xzf debug-*_all.tar.gz || true
    - mv debug-*.txt debug-information-$LABGRID_TARGET.txt || true
    - ( [ $(tar -tf coredumps_all.tar.gz | wc -l) -eq 0 ] && rm coredumps_all.tar.gz || true ) 2>/dev/null
    - mv coredumps_all.tar.gz coredumps-$LABGRID_TARGET.tar.gz 2> /dev/null || true
    - mv console_$LABGRID_TARGET console_$LABGRID_TARGET.txt || true

  artifacts:
    expire_in: 1 month
    when: always
    paths:
      - .gitlab/tests/cram/**/*.t.err
      - etc
      - prplmesh_beerocks_logs
      - debug-information-$LABGRID_TARGET.txt
      - processes-$LABGRID_TARGET.txt
      - dmesg-$LABGRID_TARGET.txt
      - syslog-$LABGRID_TARGET.txt
      - system-$LABGRID_TARGET.json
      - console_$LABGRID_TARGET.txt
      - cram-result-*
      - opkg_list-$LABGRID_TARGET.txt
      - coredumps-$LABGRID_TARGET.tar.gz
    reports:
      junit: cram-result-$LABGRID_TARGET.junit.xml

.testbed true.cz:
  extends: .testbed
  variables:
    TESTBED_UART_RELAY_HOST: uart-relay.testbed.vpn.true.cz