Check cisco mds log files on remote switches

check_mds_log.py

#!/usr/bin/python
from datetime import datetime, timedelta
import logging
import subprocess
import sys
import re


logging.basicConfig(format='%(levelname)s: %(message)s', level=logging.WARN)
#                    '(year )  (Month)   (day    )  (timestamp        )  (sch)    (err):  (ed)
splitre = re.compile('(\d{4})\s(\w{3})\s+(\d{1,2})\s(\d{2}:\d{2}:\d{2})\s(\w+)\s\%(.+?):\s(.+)')

# Error message and handlers -------------------------------------------------------------------------------------------

def _ignore(reason=None):
    logging.debug(reason)
    return 0

def _warn(reason):
    logging.warning(reason)
    return 1

def _alert(reason):
    logging.critical(reason)
    return 2

# Special handlers

def _loginwarn(reason):
    if 'Authentication failed for user root' in reason:
        return _alert(reason)
    return _warn(reason)

error_handles = {
    'PORT-5-IF_DOWN_LINK_FAILURE': _alert,
    'DEVICE-ALIAS-3-MERGE_FAILED': _warn,
    'ETHPORT-5-IF_DOWN_LINK_FAILURE': _warn,
    'FCDOMAIN-2-EPORT_ISOLATED': _warn,
    'KERN-1-SYSTEM_MSG': _warn,
    'MODULE-2-MOD_NOT_ALIVE': _alert,
    'PLATFORM-3-MOD_PWRFAIL_MULTI': _alert,
    'PORT-3-IF_UNSUPPORTED_TRANSCEIVER': _warn,
    'PORT-5-IF_DOWN_ETH_LINK_DOWN': _alert,
    'PORT-5-IF_DOWN_PEER_RESET': _warn,
    'PORT-5-IF_DOWN_PORT_CHANNEL_MEMBERS_DOWN': _warn,
    'SYSMGR-2-SERVICE_CRASHED': _alert,
    'ZONE-2-ZS_MERGE_FULL_DATABASE_MISMATCH': _alert,
    'AUTHPRIV-3-SYSTEM_MSG': _loginwarn,
}

# Log handlers ---------------------------------------------------------------------------------------------------------

def process_line(line):
    """Split our log file entry into time, host, error code and error message
    For each error code we look for it in our known dictionary, if it is found
    we process it with the key-value method defined in error_handles

    'last message repeated' is an edge case that we try to handle gracefully
    """
    m = splitre.match(line)
    if m:
        # 0-3 time, 4 host, 5 err code, 6 error message
        hndlr = error_handles.get(m.group(6))
        if hndlr:
            return hndlr(m.group(7))
        else:
            return _ignore(m.group(7))
    else:
        if 'last message repeated' in line:
            return 0
        logging.critical(line)
        return 2

def format_request(host):
    """Query cisco switch for log files beginning 1 hour ago"""
    lh = datetime.today() - timedelta(hours = 1)
    return ["ssh", "-q", host, "show", "logging", "logfile", "start-time", lh.strftime('%Y %b %d %H:%M:%S')]

def request_logs(host):
    """Start a ssh connection to the desired host
    Each line of output is sent through process_line,
    the return values are then used to determine the
    exit value for the script

    If no values are returned the log was probably empty
    so we exit with OK / 0
    """
    process = subprocess.Popen(format_request(host), stdout=subprocess.PIPE)
    (output, errors) = process.communicate()
    if errors:
        logging.critical(errors)
        sys.exit(1)
    lines = output.splitlines()
    retvals = map(process_line, lines)
    if retvals and max(retvals) > 0:
        sys.exit(max(retvals))
    print "OK: no errors reported"
    sys.exit(0)



if __name__ == '__main__':
    if len(sys.argv) == 2:
        request_logs(sys.argv[1])
    else:
        logging.critical("No host given. Usage: %s <host>" % sys.argv[0])
        sys.exit(1)