send-boot-metrics.conf

# Copyright (c) 2013 The Chromium OS Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.

description   "Report once-per-boot UMA statistics"
author        "chromium-os-dev@chromium.org"

start on started system-services

script
  # Every time we shut down, we save a log of all the bootstat event
  # files.  To prevent unbounded growth, when we get back here, we
  # remove the data for all but the 50 most recent boots.
  BOOT_LOG_ROTATE_COUNT=50
  rm -rf $(ls -dt /var/log/metrics/shutdown.[0-9]* |
               sed "1,$BOOT_LOG_ROTATE_COUNT d")

  GET_FIRMWARE_TIME='
    /Initial usec timer/ {
      sub(".*Initial usec timer ", "")
      initial_usec = $0
      exit
    }
    END {
      printf "%.2f\n", initial_usec / 1000000
    }
  '

  FIRMWARE_TIME=$(dmesg | awk "$GET_FIRMWARE_TIME")

  # TODO(jrbarnette): The firmware-boot-time file is used by tests
  # for two purposes:
  #  1) by boot time tests, to estimate total boot time from power
  #     on (not just from kernel startup), and
  #  2) by firmware tests, to estimate how long the dev mode warning
  #     screen is presented.
  #
  # Probably there should be another way for autotests to get boot
  # time, so the code to create this file can be removed.
  echo $FIRMWARE_TIME >/tmp/firmware-boot-time

  # usage:
  #   report_disk_metrics <read-sectors> <write-sectors>
  report_disk_metrics() {
    metrics_client Platform.BootSectorsRead $1 1 1000000 50
    metrics_client Platform.BootSectorsWritten $2 1 10000 50
  }

  report_disk_metrics $(
    bootstat_get_last boot-complete read-sectors write-sectors)

  send_metrics_on_resume -b

  # Report disk health from SMART (S.M.A.R.T.) parameters.
  # usage:
  #   report_smart_metrics <uncorrectable-errors> <transfer-errors>
  report_smart_metrics() {
    metrics_client Platform.SmartUncorrectableErrors $1 1 1000000 20
    metrics_client Platform.SmartTransferErrors      $2 1 1000000 20
  }

  GET_SMART_ERRORS='
    BEGIN { uncorr = 0; txfer = 0 }
    $2 == "Reported_Uncorrect" { uncorr = $10 }
    $2 == "UDMA_CRC_Error_Count" { txfer = $10 }
    END { print uncorr, txfer }
  '
  report_smart_metrics $(
    smartctl -A $(rootdev -s -d) | awk "$GET_SMART_ERRORS")

  # Check if the EC is in TPSRESET state and if so report it, but only once
  # (by design, the EC stays in that state forever).
  EC_RESET_FILE=/home/chronos/.ec.tpsreset
  if ectool chipinfo | grep -q TPSRESET &&
    test ! -f $EC_RESET_FILE
  then
    metrics_client -v Ec.TpsReset
    su chronos -c "touch $EC_RESET_FILE"
  fi

  POWERWASH_COMPLETE="/mnt/stateful_partition/unencrypted/.powerwash_completed"
  POWERWASH_COUNT="/mnt/stateful_partition/unencrypted/preserve/powerwash_count"
  # Report powerwash stats if a powerwash recently occurred.
  if [ -f $POWERWASH_COMPLETE -a -f /home/chronos/.oobe_completed ]; then
    # Upload the fact powerwash occurred and the number of times this device
    # has been powerwashed.
    COUNT=$(head -1 $POWERWASH_COUNT | cut -c1-4)
    if [ $(expr "$COUNT" : "^[0-9][0-9]*$") -ne 0 ]; then
      metrics_client Installer.PowerwashCount $COUNT 1 1000 20
    fi
    rm -f $POWERWASH_COMPLETE
  fi

  CHROME_EXEC_TIME=$(bootstat_get_last chrome-exec time)
  BOOT_COMPLETE_TIME=$(bootstat_get_last boot-complete time)
  PRE_STARTUP_TIME=$(bootstat_get_last pre-startup time)
  TOTAL_TIME=$(echo $FIRMWARE_TIME $BOOT_COMPLETE_TIME | awk '{print $1 + $2}')
  SYSTEM_TIME=$(echo $CHROME_EXEC_TIME $PRE_STARTUP_TIME |
             awk '{print $1 - $2}')
  CHROME_TIME=$(echo $BOOT_COMPLETE_TIME $CHROME_EXEC_TIME |
             awk '{print $1 - $2}')
  metrics_client -t BootTime.Total $TOTAL_TIME 1 10000 50
  metrics_client -t BootTime.Firmware $FIRMWARE_TIME 1 10000 50
  metrics_client -t BootTime.Kernel $PRE_STARTUP_TIME 1 10000 50
  metrics_client -t BootTime.System $SYSTEM_TIME 1 10000 50
  metrics_client -t BootTime.Chrome $CHROME_TIME 1 10000 50
end script