SMART collection script for DataDog

dd-smart.py

#! /usr/bin/env python
# coding: utf8
"""
Publish metrics from `smartctl` output:

  * temp.DISK = celsius
  * 

Also publish these events when they occur:

  * SMART 5 - Reallocated_Sector_Count.
  * SMART 187 - Reported_Uncorrectable_Errors.
  * SMART 188 - Command_Timeout.
  * SMART 197 - Current_Pending_Sector_Count.
  * SMART 198 - Offline_Uncorrectable
"""

import sys, time, os, re, subprocess
from datadog.statsd import statsd


def output(disk):
    """Return raw output of `smartctl -x DISK`.

    Raise Exception when:
      * there is any error output
      * return status is not successful (ie not 0)
      * smartctl executable not found
      * DISK does not exist
    """
    if os.system('which smartctl > /dev/null') != 0:
        raise Exception("Failed to find `smartctl` executable")
    if not os.path.exists(disk):
        raise Exception("Disk does not exist: {}".format(disk))
    p = subprocess.Popen(['smartctl', '-x', disk],
                         stdout=subprocess.PIPE,
                         stderr=subprocess.PIPE)
    status = p.wait()
    err = p.stderr.read()
    if err or status != 0:
        raise Exception("Failed to execute `smartctl`: status={}, stderr={}".format(status, err))
    return (i.strip() for i in p.stdout.readlines())


def parsed(disk, lines):
    """Yield (name,value,line) for each numeric data point.
    """
    for line in lines:
        fields = line.split()
        if len(fields) < 3:
            continue
        if fields[1] == 'Temperature_Celsius':
            name = "temp.{}".format(disk)
            val = float(fields[7])
            statsd.gauge(name, val)
            yield (name, val, line)


for disk in sys.argv[1:]:
    for name, val, line in parsed(disk, output(disk)):
        # Print in datadog-friendly log format
        print("{} {} {}".format(name, int(time.time()), val))