check_smart_attr.py

#!/usr/bin/python
# -*- coding: utf-8-*-
####################################################################
# FILENAME: check_smart_attr.py
# PROJECT: Misc Scripts
# DESCRIPTION: Nagios health check to monitor a SMART attribute
#              and alert against it.
#
#
# $Id$
####################################################################
# (C)2011 DigiTar, All Rights Reserved
####################################################################

import subprocess, sys, re
from argparse import ArgumentParser

EXIT_OK = 0
EXIT_WARNING = 1
EXIT_CRITICAL = 2
EXIT_UNKNOWN = 3



re_id = re.compile(r"^(?P<id>\d+)\s+(?P<attr_name>[-A-Za-z_0-9]+)\s+(?P<flag>0x\d+)\s+(?P<value>\d+)\s+(?P<worst>\d+)\s+(?P<thresh>\d+)\s+(?P<type>[-A-Za-z_]+)\s+(?P<updated>[A-Za-z]+)\s+(?P<when_failed>[^\s]+)\s+(?P<raw_value>[^\s]+)$")

parser = ArgumentParser()
parser.add_argument("--disk", dest="disk", type=str, required=True)
parser.add_argument("--disk_type", dest="disk_type", type=str, default="ata")
parser.add_argument("--smartctlpath", dest="spath", default="/usr/sbin/smartctl")
parser.add_argument("--attrid", dest="attribute_id", type=int, required=True)
parser.add_argument("--critical", dest="critical_threshold", type=int, required=True)
parser.add_argument("--warning", dest="warning_threshold", type=int, required=True)
parser.add_argument("--descending", dest="descending_values", action="store_const", const=True, default=False)


if __name__ == "__main__":
    args = parser.parse_args()
    
    try:
        output = subprocess.Popen([args.spath,
                                   "-d",
                                   args.disk_type,
                                   "-A",
                                   args.disk],
                                  stdout=subprocess.PIPE).communicate()[0]
    except subprocess.CalledProcessError, e:
        print "Error running check: " + e.output
        sys.exit(EXIT_UNKNOWN)
    
    for line in output.split("\n"):
        match = re_id.match(line)
        if match:
            parsed_match = match.groupdict()
            if int(parsed_match["id"]) == args.attribute_id:
                value = int(parsed_match["value"])
                if args.descending_values:
                    if value < args.critical_threshold:
                        print "CRITICAL: %s - %s value is %d. Threshold: %d" % (args.disk,
                                                                                parsed_match["attr_name"],
                                                                                value,
                                                                                args.critical_threshold)
                        sys.exit(EXIT_CRITICAL)
                    elif value < args.warning_threshold:
                        print "WARNING: %s - %s value is %d. Threshold: %d" % (args.disk,
                                                                               parsed_match["attr_name"],
                                                                               value,
                                                                               args.warning_threshold)
                        sys.exit(EXIT_WARNING)
                    else:
                        print "OK: %s - %s value is %d" % (args.disk,
                                                           parsed_match["attr_name"],
                                                           value)
                        sys.exit(EXIT_OK)
                else:
                    if value > args.critical_threshold:
                        print "CRITICAL: %s - %s value is %d. Threshold: %d" % (args.disk,
                                                                                parsed_match["attr_name"],
                                                                                value,
                                                                                args.critical_threshold)
                        sys.exit(EXIT_CRITICAL)
                    elif value > args.warning_threshold:
                        print "WARNING: %s - %s value is %d. Threshold: %d" % (args.disk,
                                                                               parsed_match["attr_name"],
                                                                               value,
                                                                               args.warning_threshold)
                        sys.exit(EXIT_WARNING)
                    else:
                        print "OK: %s - %s value is %d" % (args.disk,
                                                           parsed_match["attr_name"],
                                                           value)
                        sys.exit(EXIT_OK)
    
    
    # No matching ID found kemosabe...kick out
    print "UNKNOWN: Could not find SMART ID %d in device %s output." % (args.attribute_id, args.disk)
    sys.exit(EXIT_UNKNOWN)