API tools faq
paste
Advertisement
Guest User

smartmon.sh

a guest
Aug 28th, 2021
105
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Bash 13.71 KB | None | 0 0
raw download clone embed print report
  1. #!/usr/bin/env bash
  2. # Script informed by the collectd monitoring script for smartmontools (using smartctl)
  3. # by Samuel B. <samuel_._behan_(at)_dob_._sk> (c) 2012
  4. # source at: http://devel.dob.sk/collectd-scripts/
  5.  
  6. # TODO: This probably needs to be a little more complex.  The raw numbers can have more
  7. #       data in them than you'd think.
  8. #       http://arstechnica.com/civis/viewtopic.php?p=22062211
  9.  
  10. # Formatting done via shfmt -i 2
  11. # https://github.com/mvdan/sh
  12.  
  13. SMARTCTL=`which smartctl`
  14. #FORCED_DEVICE_LIST=$(cat << EOF
  15. #/dev/sg3|scsi
  16. #/dev/sg4|sat
  17. #/dev/sg5|sat
  18. #/dev/sg6|sat
  19. #/dev/sg7|scsi
  20. #/dev/sg8|sat
  21. #/dev/sg9|sat
  22. #/dev/sdc|sat
  23. #EOF
  24. #)
  25.  
  26. parse_smartctl_attributes_awk="$(
  27.  cat <<'SMARTCTLAWK'
  28. $1 ~ /^ *[0-9]+$/ && $2 ~ /^[a-zA-Z0-9_-]+$/ {
  29.  gsub(/-/, "_");
  30.  printf "%s_value{%s,smart_id=\"%s\"} %d\n", tolower($2), labels, $1, $4
  31.  printf "%s_worst{%s,smart_id=\"%s\"} %d\n", tolower($2), labels, $1, $5
  32.  printf "%s_threshold{%s,smart_id=\"%s\"} %d\n", tolower($2), labels, $1, $6
  33.  printf "%s_raw_value{%s,smart_id=\"%s\"} %e\n", tolower($2), labels, $1, $10
  34. }
  35. SMARTCTLAWK
  36. )"
  37.  
  38. smartmon_attrs="$(
  39.  cat <<'SMARTMONATTRS'
  40. airflow_temperature_cel
  41. command_timeout
  42. current_pending_sector
  43. end_to_end_error
  44. erase_fail_count
  45. g_sense_error_rate
  46. hardware_ecc_recovered
  47. host_reads_mib
  48. host_reads_32mib
  49. host_writes_mib
  50. host_writes_32mib
  51. load_cycle_count
  52. media_wearout_indicator
  53. multi_zone_error_rate
  54. wear_leveling_count
  55. nand_writes_1gib
  56. offline_uncorrectable
  57. percent_lifetime_remain
  58. power_cycle_count
  59. power_off_retract_count
  60. power_on_hours
  61. program_fail_count
  62. raw_read_error_rate
  63. reallocated_event_count
  64. reallocated_sector_ct
  65. reallocate_nand_blk_cnt
  66. reported_uncorrect
  67. sata_downshift_count
  68. seek_error_rate
  69. spin_retry_count
  70. spin_up_time
  71. start_stop_count
  72. temperature_case
  73. temperature_celsius
  74. temperature_internal
  75. total_lbas_read
  76. total_lbas_written
  77. total_host_sector_write
  78. udma_crc_error_count
  79. unsafe_shutdown_count
  80. workld_host_reads_perc
  81. workld_media_wear_indic
  82. workload_minutes
  83. throughput_performance
  84. seek_time_performance
  85. helium_level
  86. unknown_attribute
  87. SMARTMONATTRS
  88. )"
  89. smartmon_attrs="$(echo ${smartmon_attrs} | xargs | tr ' ' '|')"
  90.  
  91. parse_smartctl_attributes() {
  92.   local disk="$1"
  93.   local disk_type="$2"
  94.   local name="$3"
  95.   local serial_number="$4"
  96.   local labels="disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\",serial_number=\"${serial_number}\""
  97.   local vars="$(echo "${smartmon_attrs}" | xargs | tr ' ' '|')"
  98.   sed 's/^ \+//g' |
  99.     awk -v labels="${labels}" "${parse_smartctl_attributes_awk}" 2>/dev/null |
  100.     grep -E "(${smartmon_attrs})"
  101. }
  102.  
  103. parse_smartctl_scsi_attributes() {
  104.   local disk="$1"
  105.   local disk_type="$2"
  106.   local name="$3"
  107.   local labels="disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\""
  108.   while read line; do
  109.     attr_type="$(echo "${line}" | tr '=' ':' | cut -f1 -d: | sed 's/^ \+//g' | tr ' ' '_')"
  110.     attr_value="$(echo "${line}" | tr '=' ':' | cut -f2 -d: | sed 's/^ \+//g')"
  111.     case "${attr_type}" in
  112.     number_of_hours_powered_up_) power_on="$(echo "${attr_value}" | awk '{ printf "%e\n", $1 }')" ;;
  113.     Current_Drive_Temperature) temp_cel="$(echo ${attr_value} | cut -f1 -d' ' | awk '{ printf "%e\n", $1 }')" ;;
  114.     Blocks_read_from_cache_and_sent_to_initiator_) lbas_read="$(echo ${attr_value} | awk '{ printf "%e\n", $1 }')" ;;
  115.     Accumulated_start-stop_cycles) power_cycle="$(echo ${attr_value} | awk '{ printf "%e\n", $1 }')" ;;
  116.     Elements_in_grown_defect_list) grown_defects="$(echo ${attr_value} | awk '{ printf "%e\n", $1 }')" ;;
  117.     Non-medium_error_count) non_medium="$(echo ${attr_value} | awk '{ printf "%e\n", $1 }')" ;;
  118.     read) read_uncorrected="$(echo ${attr_value} | awk '{ printf "%e\n", $7 }')" ;;
  119.     write) write_uncorrected="$(echo ${attr_value} | awk '{ printf "%e\n", $7 }')" ;;
  120.     verify) verify_uncorrected="$(echo ${attr_value} | awk '{ printf "%e\n", $7 }')" ;;
  121.     esac
  122.   done
  123.   [ ! -z "$power_on" ] && echo "power_on_hours_raw_value{${labels},smart_id=\"9\"} ${power_on}"
  124.   [ ! -z "$temp_cel" ] && echo "temperature_celsius_raw_value{${labels},smart_id=\"194\"} ${temp_cel}"
  125.   [ ! -z "$lbas_read" ] && echo "total_lbas_read_raw_value{${labels},smart_id=\"242\"} ${lbas_read}"
  126.   [ ! -z "$power_cycle" ] && echo "power_cycle_count_raw_value{${labels},smart_id=\"12\"} ${power_cycle}"
  127.   [ ! -z "$grown_defects" ] && echo "sas_grown_defects_count_raw_value{${labels},smart_id=\"0\"} ${grown_defects}"
  128.   [ ! -z "$non_medium" ] && echo "sas_non_medium_errors_count_raw_value{${labels},smart_id=\"0\"} ${non_medium}"
  129.   [ ! -z "$read_uncorrected" ] && echo "sas_read_uncorrected_errors_count_raw_value{${labels},smart_id=\"0\"} ${read_uncorrected}"
  130.   [ ! -z "$write_uncorrected" ] && echo "sas_write_uncorrected_errors_count_raw_value{${labels},smart_id=\"0\"} ${write_uncorrected}"
  131.   [ ! -z "$verify_uncorrected" ] && echo "sas_verify_uncorrected_errors_count_raw_value{${labels},smart_id=\"0\"} ${verify_uncorrected}"
  132. }
  133.  
  134. parse_smartctl_nvme_attributes() {
  135.   local disk="$1"
  136.   local disk_type="$2"
  137.   local name="$3"
  138.   local serial_number="$4"
  139.   local labels="disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\",serial_number=\"${serial_number}\""
  140.  
  141.   while read line; do
  142.     attr_type="$(echo "${line}" | tr '=' ':' | cut -f1 -d: | sed 's/^ \+//g' | tr ' ' '_')"
  143.     attr_value="$(echo "${line}" | tr '=' ':' | cut -f2 -d: | sed 's/^ \+//g')"
  144.     case "${attr_type}" in
  145.     Temperature) temp_cel="$(echo ${attr_value} | cut -f1 -d' ' | awk '{ printf "%e\n", $1 }')" ;;
  146.     Percentage_Used) wear_level="$(echo "${attr_value}" | awk '{ printf "%e\n", $1 }')" ;;
  147.     Host_Read_Commands) host_read_commands="$(echo ${attr_value} | awk '/[0-9]/{gsub ( /,/,"" ) } ; 1' | awk '{ printf "%e\n", $1 }')" ;;
  148.     Host_Write_Commands) host_write_commands="$(echo ${attr_value} | awk '/[0-9]/{gsub ( /,/,"" ) } ; 1' | awk '{ printf "%e\n", $1 }')" ;;
  149.     Data_Units_Read) data_read="$(echo ${attr_value}| sed -E 's/\s+.+$//g' | tr -d ','| awk '{printf "%e\n", $1}')" ;;
  150.     Data_Units_Written) data_written="$(echo ${attr_value}| sed -E 's/\s+.+$//g' | tr -d ','| awk '{printf "%e\n", $1}')" ;;
  151.     Power_Cycles) power_cycles="$(echo ${attr_value} | awk '/[0-9]/{gsub ( /,/,"" ) } ; 1' | awk '{ printf "%e\n", $1 }')" ;;
  152.     Power_On_Hours) power_on="$(echo ${attr_value} | awk '/[0-9]/{gsub ( /,/,"" ) } ; 1' | awk '{ printf "%e\n", $1 }')" ;;
  153.     Unsafe_Shutdowns) unsafe_shutdown_count="$(echo ${attr_value} | awk '/[0-9]/{gsub ( /,/,"" ) } ; 1' | awk '{ printf "%e\n", $1 }')" ;;
  154.     esac
  155.   done
  156.   [ ! -z "$temp_cel" ] && echo "temperature_celsius_raw_value{${labels},smart_id=\"194\"} ${temp_cel}"
  157.   [ ! -z "$wear_level" ] && echo "wear_level_raw_value{${labels},smart_id=\"0\"} ${wear_level}"
  158.   [ ! -z "$host_read_commands" ] && echo "host_read_commands_raw_value{${labels},smart_id=\"0\"} ${host_read_commands}"
  159.   [ ! -z "$host_write_commands" ] && echo "host_write_commands_raw_value{${labels},smart_id=\"0\"} ${host_write_commands}"
  160.   [ ! -z "$data_read" ] && echo "data_units_read_raw_value{${labels},smart_id=\"0\"} ${data_read}"
  161.   [ ! -z "$data_written" ] && echo "data_units_written_raw_value{${labels},smart_id=\"0\"} ${data_written}"
  162.   [ ! -z "$power_cycles" ] && echo "power_cycle_count_raw_value{${labels},smart_id=\"12\"} ${power_cycles}"
  163.   [ ! -z "$power_on" ] && echo "power_on_hours_raw_value{${labels},smart_id=\"9\"} ${power_on}"
  164.   [ ! -z "$unsafe_shutdown_count" ] && echo "unsafe_shutdowns_raw_value{${labels},smart_id=\"9\"} ${unsafe_shutdown_count}"
  165. }
  166.  
  167. parse_smartctl_info() {
  168.   local -i smart_available=0 smart_enabled=0 smart_healthy=0
  169.   local disk="$1" disk_type="$2" name="$3"
  170.   local model_family='N/A' device_model='N/A' size='N/A' serial_number='N/A' fw_version='N/A' vendor='N/A' product='N/A' revision='N/A' lun_id='N/A'
  171.   while read line; do
  172.     info_type="$(echo "${line}" | cut -f1 -d: | tr ' ' '_')"
  173.     info_value="$(echo "${line}" | cut -f2- -d: | sed 's/^ \+//g' | sed 's/"/\\"/')"
  174.     case "${info_type}" in
  175.     Model_Family) model_family="${info_value}" ;;
  176.     Device_Model) device_model="${info_value}" ;;
  177.     Model_Number) device_model="${info_value}" ;;
  178.     Serial_[Nn]umber) serial_number="${info_value}" ;;
  179.     Firmware_Version) fw_version="${info_value}" ;;
  180.     User_Capacity) size="$(echo ${info_value}| sed -E 's/\s+.+$//g' | tr -d ','| awk '{printf "%d GB\n", $1/1024/1024/1024}')" ;;
  181.     Namespace_1_Size/Capacity) size="$(echo ${info_value}| sed -E 's/\s+.+$//g' | tr -d ','| awk '{printf "%d GB\n", $1/1024/1024/1024}')" ;;
  182.     Vendor) vendor="${info_value}" ;;
  183.     Product) product="${info_value}" ;;
  184.     Revision) revision="${info_value}" ;;
  185.     Logical_Unit_id) lun_id="${info_value}" ;;
  186.     esac
  187.     if [[ "${info_type}" == 'SMART_support_is' ]]; then
  188.       val=$(echo $info_value | awk '{$1=$1k;print}')
  189.       case "${val:0:7}" in
  190.       Enabled) smart_enabled=1 ;;
  191.       Availab) smart_available=1 ;;
  192.       Unavail) smart_available=0 ;;
  193.       esac
  194.     fi
  195.     if [[ "${info_type}" == 'SMART_overall-health_self-assessment_test_result' ]]; then
  196.       info_value=`echo ${info_value}| tr -d ' '`
  197.       case "${info_value}" in
  198.       PASSED) smart_healthy=1 ;;
  199.       esac
  200.     elif [[ "${info_type}" == 'SMART_Health_Status' ]]; then
  201.       info_value=`echo ${info_value}| tr -d ' '`
  202.       case "${info_value}" in
  203.       OK) smart_healthy=1 ;;
  204.       esac
  205.     fi
  206.   done
  207.   if [[ $device_model == 'N/A' ]] && ([[ $vendor != 'N/A' ]] || [[ $product != 'N/A'  ]])
  208.     then
  209.     device_model="${vendor} $product"
  210.   fi
  211.  
  212.  
  213.   disk=$(echo $disk | awk '{$1=$1k;print}')
  214.   name=$(echo $name | awk '{$1=$1k;print}')
  215.   vendor=$(echo $vendor | awk '{$1=$1k;print}')
  216.   model_family=$(echo $model_family | awk '{$1=$1k;print}')
  217.   device_model=$(echo $device_model | awk '{$1=$1k;print}')
  218.   serial_number=$(echo $serial_number | awk '{$1=$1k;print}')
  219.   fw_version=$(echo $fw_version | awk '{$1=$1k;print}')
  220.  
  221.   echo "device_info{disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\",vendor=\"${vendor}\",product=\"${product}\",revision=\"${revision}\",lun_id=\"${lun_id}\",model_family=\"${model_family}\",device_model=\"${device_model}\",serial_number=\"${serial_number}\",size=\"${size}\",firmware_version=\"${fw_version}\",smart_healthy=\"${smart_healthy}\"} 1"
  222.   echo "device_smart_available{disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\",serial_number=\"${serial_number}\"} ${smart_available}"
  223.   echo "device_smart_enabled{disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\",serial_number=\"${serial_number}\"} ${smart_enabled}"
  224.   echo "device_smart_healthy{disk=\"${disk}\",type=\"${disk_type}\",name=\"${name}\",serial_number=\"${serial_number}\"} ${smart_healthy}"
  225. }
  226.  
  227. output_format_awk="$(
  228.  cat <<'OUTPUTAWK'
  229. BEGIN { v = "" }
  230. v != $1 {
  231.  print "# HELP smartmon_" $1 " SMART metric " $1;
  232.   print "# TYPE smartmon_" $1 " gauge";
  233.   v = $1
  234. }
  235. {print "smartmon_" $0}
  236. OUTPUTAWK
  237. )"
  238.  
  239. format_output() {
  240.  sort |
  241.    awk -F'{' "${output_format_awk}"
  242. }
  243.  
  244. smartctl_version="$($SMARTCTL -V | head -n1 | awk '$1 == "smartctl" {print $2}')"
  245.  
  246. echo "smartctl_version{version=\"${smartctl_version}\"} 1" | format_output
  247.  
  248. if [[ "$(expr "${smartctl_version}" : '\([0-9]*\)\..*')" -lt 6 ]]; then
  249.  exit
  250. fi
  251.  
  252. device_list=
  253. if [[ -z $FORCED_DEVICE_LIST ]]
  254.  then
  255.  device_list="$($SMARTCTL --scan-open | awk '/^\/dev/{print $1 "|" $3}')"
  256. else
  257.  device_list=$FORCED_DEVICE_LIST
  258. fi
  259.  
  260. for device in ${device_list}; do
  261.  disk="$(echo ${device} | cut -f1 -d'|')"
  262.  type="$(echo ${device} | cut -f2 -d'|')"
  263.  disk=$(echo $disk | awk '{$1=$1k;print}')
  264.  type=$(echo $type | awk '{$1=$1k;print}')
  265.  
  266.  serial_number=`$SMARTCTL -i -d "${type}" "${disk}" | awk -F ':' '/[Ss]erial [Nn]umber/ {print $2}'| sed -E 's/^\s+//g'`
  267.  serial_number=$(echo $serial_number | awk '{$1=$1k;print}')
  268.  
  269.  active=1
  270.  # Check if the device is in a low-power mode
  271.  $SMARTCTL -n standby -d "${type}" "${disk}" > /dev/null || active=0
  272.  echo "device_active{disk=\"${disk}\",type=\"${type}\",serial_number=\"${serial_number}\"}" "${active}"
  273.  # Skip further metrics to prevent the disk from spinning up
  274.  test ${active} -eq 0 && continue
  275.  # Get Device name label
  276.  name=""
  277.  case ${type} in
  278.    scsi)
  279.      vendor=`$SMARTCTL -i -d "${type}" "${disk}" | awk -F ':' '/Vendor/ {print $2}'| sed -E 's/^\s+//g'`
  280.      product=`$SMARTCTL -i -d "${type}" "${disk}" | awk -F ':' '/Product/ {print $2}'| sed -E 's/^\s+//g'`
  281.      name="${vendor} ${product}" ;;
  282.     nvme)
  283.      name=`$SMARTCTL -i -d "${type}" "${disk}" | awk -F ':' '/Model Number/ {print $2}'| sed -E 's/^\s+//g'` ;;
  284.    *)
  285.      name=`$SMARTCTL -i -d "${type}" "${disk}" | awk -F ':' '/Device Model/ {print $2}'| sed -E 's/^\s+//g'` ;;
  286.  esac
  287.  
  288.  
  289.  name=$(echo $name | awk '{$1=$1k;print}')
  290.  echo "smartctl_run{disk=\"${disk}\",type=\"${type}\",name=\"${name}\",serial_number=\"${serial_number}\"}" "$(TZ=UTC date '+%s')"
  291.  # Get the SMART information and health
  292.  $SMARTCTL -i -H -d "${type}" "${disk}" | parse_smartctl_info "${disk}" "${type}" "${name}"
  293.  # Get the SMART attributes
  294.  case ${type} in
  295.    sat) $SMARTCTL -A -d "${type}" "${disk}" | parse_smartctl_attributes "${disk}" "${type}" "${name}" "${serial_number}" ;;
  296.    nvme) $SMARTCTL -A -d "${type}" "${disk}" | parse_smartctl_nvme_attributes "${disk}" "${type}" "${name}" "${serial_number}" ;;
  297.    atacam) $SMARTCTL -A -d "${type}" "${disk}" | parse_smartctl_attributes "${disk}" "${type}" "${name}" "${serial_number}" ;;
  298.    sat+megaraid*) $SMARTCTL -A -d "${type}" "${disk}" | parse_smartctl_attributes "${disk}" "${type}" "${name}" "${serial_number}" ;;
  299.    scsi) $SMARTCTL -a -d "${type}" "${disk}" | parse_smartctl_scsi_attributes "${disk}" "${type}" "${name}" ;;
  300.    megaraid*) $SMARTCTL -A -d "${type}" "${disk}" | parse_smartctl_scsi_attributes "${disk}" "${type}" "${name}" ;;
  301.    *)
  302.      continue
  303.      ;;
  304.  esac
  305. done | format_output
  306.  
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!