#!/usr/bin/python
# -*- coding: latin-1 -*-
#
# check_cpq
# script di verifica tramite le MIB HP dello stato di saluta dei server
#
# Version: 1.0 - angeloxx@angeloxx.it
#                Prima stesura
# Version: 1.1 - angeloxx@angeloxx.it
#                Aggiunta della funzionalità "HEALTH"
#
# Query results can be:
#    1 -> other     Status detection is not supported by this system or driver.
#    2 -> ok        Device is operating properly
#    3 -> degraded  A redundant device is not operating properly.
#    4 -> failed    A non-redundant device is not operating properly.
#
# Device FAN,PSU,TEMP starts from 1 to X
# Device CPU,LDEV,PDEV starts from 0 to X
#

import sys, threading, StringIO
from optparse import OptionParser
from pysnmp import asn1, v1, v2c, role


NAGIOS_OK = 0
NAGIOS_WARNING = 1
NAGIOS_CRITICAL = 2
NAGIOS_UNKNOWN = 3

uso = "%prog [-c community] -s host -d (FAN|PSU|TEMP|LDEV|PDEV|CPU|HEALTH)"
parser = OptionParser(uso)
parser.add_option("-c", "--community",      dest="community",   help="SNMP community", default="public")
parser.add_option("-s", "--server",         dest="host",        help="remote host", default='')
parser.add_option("-d", "--device",         dest="device",      help="monitored device", default='PSU')

(options, args) = parser.parse_args()
if not options.host:
    parser.error("option host is mandatory")
    sys.exit(NAGIOS_UNKNOWN)

#if options.device == 'HEALTH':
# CPQHLTH-MIB::cpqHeEventLogEntrySeverity.0 = INTEGER: informational(2)
# CPQHLTH-MIB::cpqHeEventLogErrorDesc.7 = STRING: "System Power Supplies Not Redundant"
qry = '.1.3.6.1.4.1.232.6.2.11.3.1.2.'
deviceid = 0
ishealth = True

if options.device == 'PSU':
    # CPQHLTH-MIB::cpqHeFltTolPowerSupplyCondition.0.X = INTEGER: ok(2) 
    qry = '.1.3.6.1.4.1.232.6.2.9.3.1.4.0.' 
    deviceid = 1
    ishealth = False

if options.device == 'TEMP':
    # CPQHLTH-MIB::cpqHeTemperatureCondition.1.X = INTEGER: ok(2)
    qry = '.1.3.6.1.4.1.232.6.2.6.8.1.6.1.'
    deviceid = 1
    ishealth = False
    
if options.device == 'FAN':
    # CPQHLTH-MIB::cpqHeFltTolFanCondition.1.X = INTEGER: ok(2)
    qry = '.1.3.6.1.4.1.232.6.2.6.7.1.9.1.'
    deviceid = 1
    ishealth = False

if options.device == 'CPU':
    # CPQSTDEQ-MIB::cpqSeCpuStatus.X = INTEGER: ok(2)
    qry = '.1.3.6.1.4.1.232.1.2.2.1.1.6.'
    deviceid = 0
    ishealth = False

if options.device == 'LDEV':
    # CPQIDA-MIB::cpqDaLogDrvStatus.1.X = INTEGER: ok(2)
    qry = '.1.3.6.1.4.1.232.3.2.3.1.1.4.1.'
    deviceid = 0
    ishealth = False

if options.device == 'PDEV':
    # CPQIDA-MIB::cpqDaPhyDrvStatus.1.X = INTEGER: ok(2)
    qry = '1.3.6.1.4.1.232.3.2.5.1.1.6.1.'
    deviceid = 0
    ishealth = False


client = role.manager((options.host, 161))
req = eval('v1').GETREQUEST()
rsp = eval('v1').GETRESPONSE()

# Cache degli stati
statuses = ""
status = NAGIOS_OK
response = "OK"
try:
    walk = True
   
    while (walk):
        (answer, src) = client.send_and_receive(req.encode(community=options.community, encoded_oids=map(asn1.OBJECTID().encode, [qry + str(deviceid)])))
        # Decode SNMP response
        rsp.decode(answer)
        vals = map(lambda x: x[0](), map(asn1.decode, rsp['encoded_vals']))
            
        if not ishealth:
            if vals[0] == 1:
                statuses = "%s [%s is %s]" % (statuses, deviceid, 'unknown')
            elif vals[0] == 2:
                statuses = "%s [%s is %s]" % (statuses, deviceid, 'ok')
            elif vals[0] == 3:
                statuses = "%s [%s is %s]" % (statuses, deviceid, 'degraded')
                # cambia lo stato, sempre che sia OK e non peggio
                if (status == NAGIOS_OK):
                    status = NAGIOS_WARNING
                    response = "WARN"
            elif vals[0] == 4:
                statuses = "%s [%s is %s]" % (statuses, deviceid, 'failed')
                status = NAGIOS_CRITICAL
                response = "CRIT"
            
            if rsp['error_status'] and deviceid >10:
                # fine della trottata, il walk è terminato perché è stato restituito penso ''
                # a meno che il deviceid sia simbolicamente inferiore a 10 perché per i dischi
                # pare che gli oggetti dei dischi non presenti non vengano restituiti (era meglio un unknown)
                # quindi... non è detto che la cavalcata abbia veramente fine
                walk = False
        else:
            # Possibili valori:
            #   informational (2)
            #   infoWithAlert (3)
            #   repaired (6)
            #   caution (9)
            #   critical (15)
            # se i valori sono 9 o 15 allora preoccupiamoci e leggiamo da
            #  .1.3.6.1.4.1.232.6.2.11.3.1.8.X la descrizione dell'errore
            
            
            if rsp['error_status'] and deviceid > 0:
                walk = False
                
            if vals[0] == 9 or vals[0] == 15:
                (answer, src) = client.send_and_receive(req.encode(community=options.community, encoded_oids=map(asn1.OBJECTID().encode, ['.1.3.6.1.4.1.232.6.2.11.3.1.8.' + str(deviceid)])))
                rsp.decode(answer)
                desc = map(lambda x: x[0](), map(asn1.decode, rsp['encoded_vals']))
                statuses = "%s [%s]" % (statuses, desc[0])
                status = NAGIOS_CRITICAL
                response = "CRIT"
            
            
        deviceid = deviceid + 1
        
except role.NetworkError:
    print "KO: Unable to contact or resolve name of remote HP server (%s)" % options.host
    sys.exit(NAGIOS_CRITICAL)


if statuses == "":
    statuses = "[no additional information]"

print "%s: %s status is %s" % (response,options.host,statuses)
sys.exit(status)