Nagios plugin to monitor the SMART status of hard disks

check_smart_status.py

#!/usr/bin/env python
'''Report the SMART status of hard disks using a log file.'''

import argparse
import subprocess
import sys

# Nagios status codes
OK, WARNING, CRITICAL, UNKNOWN = range(4)

# Bits in the return code of smartctl
BITS = (
    ('smartctl error', 3),
    ('Device open failed', 2),
    ('SMART or ATA command failed', 1),
    ('Disk failing', 1),
    ('Prefail attributes over threshold', 1),
    ('Attributes over threshold in the past', 1),
    ('Error log contains SMART errors', 1),
    ('Self-test lof contains errors', 1),
)

def main(logfile):
    with open(logfile, 'r') as f:
        last_line = f.readlines()[-1][:-1]

    timestamp, devices = last_line.split(';', 1)
    devices = devices.split(';')

    total_status = 0
    output = []
    for device in devices:
        name, code = device.split('::')
        status = 0
        status_string = []
        for i in range(8):
            if (int(code) & (1 << i)):
                meaning, c = BITS[i]
                status = max(status, c)
                status_string.append(meaning)

        if status > 0:
            status_string = ', '.join(status_string) + ' (%s)' % bin(int(code))
        else:
            status_string = 'OK'

        total_status = max(total_status, status)
        output.append('%s: %s' % (name, status_string))

    if total_status == 0:
        s = 'OK'
    elif total_status == 1:
        s = 'WARNING'
    elif total_status == 2:
        s = 'CRITICAL'
    else:
        s = 'UNKNOWN'

    output = 'SMART STATUS %s - ' % s + ', '.join(output)
    print output
    return total_status


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--logfile', '-f',
                        default='/var/log/smart_status.log',
                        help='Path to SMART status log file')
    args = parser.parse_args()

    try:
        status = main(args.logfile)
    except Exception as e:
        print 'SMART STATUS UNKNOWN - Python exception %s' % str(e)
        status = UNKNOWN

    sys.exit(status)

smart_watch.py

#!/usr/bin/env python
'''Record the status of hard disks using SMART data, writing to a log file.
Run this regularly with cron.
'''

import argparse
import subprocess
import sys

def check_smart(device):
    args = ['/usr/sbin/smartctl', '--all', '-q', 'silent'] + device.split()
    return subprocess.call(args)


def main(logfile, devices):
    date_cmd = subprocess.Popen(['date', '+%s'], stdout=subprocess.PIPE)
    now = date_cmd.communicate()[0].rstrip()
    line = [now]
    for device in devices:
        code = check_smart(device)

        # Prettify devices on RAID controllers/HBAs
        if '-d' in device:
            device = device.split()[1]

        s = '%s::%i' % (device, code)
        line.append(s)

    line = ';'.join(line) + '\n'

    with open(logfile, 'a') as f:
        f.write(line)


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument('--logfile', '-f',
                        default='/var/log/smart_status.log',
                        help='Path to SMART status log file')
    parser.add_argument('--device', '-x', action='append',
                        default=[],
                        help='Add a device to monitor')
    args = parser.parse_args()

    if len(args.device) == 0:
        sys.stderr.write('No devices specified.\n')
        parser.print_help()
        sys.exit(1)

    main(args.logfile, args.device)