This commit is contained in:
Alarig Le Lay 2020-06-14 16:07:49 +02:00
parent e5cdfee69d
commit 007ccd5b15
Signed by: alarig
GPG Key ID: 7AFE62C6DF8BCDEC
1 changed files with 283 additions and 174 deletions

457
check_zfs.py Executable file → Normal file
View File

@ -1,4 +1,6 @@
#!/usr/bin/python
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""Get zfs pool status."""
########################################################################
##
@ -10,8 +12,7 @@
## in Linux.
##
## Tested operating systems/ZFS versions:
## * Ubuntu 14.04 LTS, ZFS v5
## * CentOS 7, ZFS v5
## * See README.md
##
## This program is free software: you can redistribute it and/or modify
## it under the terms of the GNU General Public License as published by
@ -28,39 +29,47 @@
##
########################################################################
from sys import exit
import subprocess
import argparse
from array import *
from types import *
from os import geteuid
import logging
import sys
# from sys import exit
# from array import *
# from types import *
# import array
# import types
# from os import geteuid
##
# Commands to run
# CHANGE THESE IF YOU NEED TO
##
zpoolCommand='/sbin/zpool'
zfsCommand='/sbin/zfs'
zpoolCommand = '/sbin/zpool'
zfsCommand = '/sbin/zfs'
##
# Variables to print at the end
##
nagiosStatus=('OK','WARNING','CRITICAL','UNKNOWN')
stateNum=0
msg=''
perfdata=''
nagiosStatus = ('OK', 'WARNING', 'CRITICAL', 'UNKNOWN')
stateNum = 0
msg = ''
perfdata = ''
##
# Filled from command line arguments
##
checkCapacity=False
capWarnThreshold=50
capCritThreshold=80
checkFragmentation=False
fragWarnThreshold=50
fragCritThreshold=80
checkCapacity = False
capWarnThreshold = 50
capCritThreshold = 80
checkFragmentation = False
fragWarnThreshold = 50
fragCritThreshold = 80
def CheckArgBounds( valueArr, minVal, maxVal ):
logging.basicConfig(stream=sys.stdout, format='%(message)s', level=logging.WARN)
def CheckArgBounds(valueArr, minVal, maxVal):
"""Check value bounds."""
for value in valueArr:
if value < minVal:
return False
@ -68,26 +77,31 @@ def CheckArgBounds( valueArr, minVal, maxVal ):
return False
return True
def ConvertToGB( valueStr ):
def ConvertToGB(valueStr):
"""Convert to GB."""
value = valueStr[:-1]
value = value.replace(',', '.')
if valueStr.endswith('G'):
return float(value)
elif valueStr.endswith('T'):
gigs=float(value)*1024
gigs = float(value)*1024
return float(gigs)
elif valueStr.endswith('M'):
gigs=float(value) / 1024.0
gigs = float(value) / 1024.0
return float(gigs)
elif valueStr.endswith('K'):
gigs=float(value) / (1024.0 * 1024.0)
gigs = float(value) / (1024.0 * 1024.0)
return float(gigs)
def RaiseStateNum( stateNumIn, stateNum ):
def RaiseStateNum(stateNumIn, stateNum):
"""Raise state num."""
if stateNumIn > stateNum:
return stateNumIn
return stateNum
###################################################################################
##
# Parse command line args
@ -95,57 +109,55 @@ def RaiseStateNum( stateNumIn, stateNum ):
parser = argparse.ArgumentParser(
prog='check_zfs',
description='Check the ZFS pool specified by an argument.',
epilog='Note that monitor flags (e.g. capacity) require 2 arguments: warning threshold, and critical threshold')
parser.add_argument('--capacity', help="monitor utilization of zpool (%%, int [0-100])", type=int, nargs=2)
parser.add_argument('--fragmentation', help="monitor fragmentation of zpool (%%, int [0-100])", type=int, nargs=2)
epilog='Note that monitor flags (e.g. capacity) require 2 arguments:\
warning threshold, and critical threshold')
parser.add_argument('--capacity', help="monitor utilization of zpool (%%, int [0-100])",
type=int, nargs=2)
parser.add_argument('--fragmentation', help="monitor fragmentation of zpool (%%, int [0-100])",
type=int, nargs=2)
parser.add_argument('pool', help="name of the zpool to check", type=str)
args = parser.parse_args()
retVal = True
if args.capacity is not None:
checkCapacity=True
capWarnThreshold=args.capacity[0]
capCritThreshold=args.capacity[1]
capArr = array('i', [capWarnThreshold, capCritThreshold])
checkCapacity = True
capWarnThreshold = args.capacity[0]
capCritThreshold = args.capacity[1]
capArr = ['i', [capWarnThreshold, capCritThreshold]]
retVal = CheckArgBounds(capArr, 0, 100)
if retVal is False:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Capacity thresholds must be between 0 and 100 (as a percent)."
logging.warning("%s : Capacity thresholds must be between 0 and 100 (as a percent).",
nagiosStatus[stateNum])
parser.print_help()
exit(stateNum)
retVal = True
if args.fragmentation is not None:
checkFragmentation=True
fragWarnThreshold=args.fragmentation[0]
fragCritThreshold=args.fragmentation[1]
fragArr = array('i', [fragWarnThreshold, fragCritThreshold])
checkFragmentation = True
fragWarnThreshold = args.fragmentation[0]
fragCritThreshold = args.fragmentation[1]
fragArr = ['i', [fragWarnThreshold, fragCritThreshold]]
retVal = CheckArgBounds(fragArr, 0, 100)
if retVal is False:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Fragmentation thresholds must be between 0 and 100 (as a percent)."
logging.warning("%s : Fragmentation thresholds must be between 0 and 100 (as a percent).",
nagiosStatus[stateNum])
parser.print_help()
exit(stateNum)
###################################################################################
###################################################################################
##
# Verify that we're running as root. This should render redundant some checks
# below, but we'll leave them there in case of bugs and to make this more readable.
#if geteuid() != 0:
# stateNum = RaiseStateNum(3, stateNum)
# print nagiosStatus[stateNum] + ": process must be run as root. Did you for get sudo? If not, possible solution: add the following toyour visudo: nagios ALL=NOPASSWD: /sbin/zfs"
# exit(stateNum)
###################################################################################
##
# Get generic info about the ZFS environment
zfsEntries = []
fullCommand=['/usr/bin/sudo', '-n', zfsCommand, 'list']
fullCommand = ['/usr/bin/sudo', '-n', zfsCommand, 'list']
try:
childProcess = subprocess.Popen(fullCommand, stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
childProcess = subprocess.Popen(fullCommand, stdin=subprocess.PIPE, stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
except OSError:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": process must be run as root. Possible solution: add the following to your visudo: nagios ALL=NOPASSWD: /sbin/zfs"
logging.warning("%s : process must be run as root. Possible solution: add the following to your\
visudo: nagios ALL=NOPASSWD: /sbin/zfs", nagiosStatus[stateNum])
exit(stateNum)
zfsString = childProcess.communicate()[0]
@ -153,215 +165,312 @@ zfsRetval = childProcess.returncode
if zfsRetval is 1:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": process must be run as root. Possible solution: add the following to your visudo: nagios ALL=NOPASSWD: /sbin/zfs"
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zfs", nagiosStatus[stateNum])
exit(stateNum)
zfsLines = zfsString.splitlines()
for idx, line in enumerate(zfsLines):
if idx != 0:
zfsEntry=line.split()
zfsEntry = line.split()
zfsEntries.append(zfsEntry)
# Make sure the pool we specified is valid
validPool=False
validPool = False
for entry in zfsEntries:
if entry[0] == args.pool:
validPool=True
if entry[0].decode() == args.pool:
validPool = True
if not validPool:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Pool " + args.pool + " is invalid. Please select a valid pool."
logging.warning("%s : Pool %s is invalid. Please select a valid pool.",
nagiosStatus[stateNum], args.pool)
exit(stateNum)
###################################################################################
##
# Get info on zpool
fullCommand=['/usr/bin/sudo', '-n', zpoolCommand, 'list', args.pool]
fullCommand = ['/usr/bin/sudo', '-n', zpoolCommand, 'list', args.pool]
try:
childProcess = subprocess.Popen(fullCommand, stdin=subprocess.PIPE,
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
except OSError:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": process must be run as root. Possible solution: add the following to your visudo: nagios ALL=NOPASSWD: /sbin/zpool"
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zpool", nagiosStatus[stateNum])
exit(stateNum)
zpoolString = childProcess.communicate()[0]
zpoolRetval = childProcess.returncode
if zpoolRetval is 1:
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": process must be run as root. Possible solution: add the following to your visudo: nagios ALL=NOPASSWD: /sbin/zpool"
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zpool", nagiosStatus[stateNum])
exit(stateNum)
zpoolLines=zpoolString.splitlines()
zpoolMeta=zpoolLines[0].split()
zpoolMetaStr=','.join(zpoolMeta)
zpoolEntry=zpoolLines[1].split()
zpoolEntryStr=','.join(zpoolEntry)
zpoolLines = zpoolString.splitlines()
zpoolMeta = zpoolLines[0].decode().split()
zpoolMetaStr = ','.join(zpoolMeta)
zpoolEntry = zpoolLines[1].decode().split()
zpoolEntryStr = ','.join(zpoolEntry)
name=''
size=''
alloc=''
free=''
expandsz=''
frag=''
cap=''
dedup=''
health=''
altroot=''
name = ''
size = ''
alloc = ''
free = ''
expandsz = ''
frag = ''
cap = ''
dedup = ''
health = ''
altroot = ''
for idx, fieldName in enumerate(zpoolMeta):
if fieldName=='NAME':
name=zpoolEntry[idx]
elif fieldName=='SIZE':
size=zpoolEntry[idx]
elif fieldName=='ALLOC':
alloc=zpoolEntry[idx]
elif fieldName=='FREE':
free=zpoolEntry[idx]
elif fieldName=='EXPANDSZ':
expandsz=zpoolEntry[idx]
elif fieldName=='FRAG':
frag=zpoolEntry[idx]
elif fieldName=='CAP':
cap=zpoolEntry[idx]
elif fieldName=='DEDUP':
dedup=zpoolEntry[idx]
elif fieldName=='HEALTH':
health=zpoolEntry[idx]
elif fieldName=='ALTROOT':
altroot=zpoolEntry[idx]
if fieldName == 'NAME':
name = zpoolEntry[idx]
elif fieldName == 'SIZE':
size = zpoolEntry[idx]
elif fieldName == 'ALLOC':
alloc = zpoolEntry[idx]
elif fieldName == 'FREE':
free = zpoolEntry[idx]
elif fieldName == 'EXPANDSZ':
expandsz = zpoolEntry[idx]
elif fieldName == 'FRAG':
frag = zpoolEntry[idx]
elif fieldName == 'CAP':
cap = zpoolEntry[idx]
elif fieldName == 'DEDUP':
dedup = zpoolEntry[idx]
elif fieldName == 'HEALTH':
health = zpoolEntry[idx]
elif fieldName == 'ALTROOT':
altroot = zpoolEntry[idx]
if name=='':
if name == '':
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Missing required field in zpool output: NAME"
logging.warning("%s : Missing required field in zpool output: NAME", nagiosStatus[stateNum])
exit(stateNum)
if health=='':
if health == '':
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Missing required field in zpool output: HEALTH"
logging.warning("%s : Missing required field in zpool output: HEALTH", nagiosStatus[stateNum])
exit(stateNum)
if checkCapacity and cap=='':
if checkCapacity and cap == '':
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Cannot monitor capacity without zpool output: CAP. Outputs are" + zpoolMetaStr
logging.warning("%s Cannot monitor capacity without zpool output: CAP.\
Outputs are %s", nagiosStatus[stateNum], zpoolMetaStr)
exit(stateNum)
if checkFragmentation and frag=='':
if checkFragmentation and frag == '':
stateNum = RaiseStateNum(3, stateNum)
print nagiosStatus[stateNum] + ": Cannot monitor fragmentation without zpool output: FRAG. Outputs are " + zpoolMetaStr
logging.warning("%s : Cannot monitor fragmentation without zpool output: FRAG.\
Outputs are %s", nagiosStatus[stateNum], zpoolMetaStr)
exit(stateNum)
# Get compressratio on zpool
checkForCompression = ['/usr/bin/sudo', '-n', zfsCommand, 'get', 'compression', args.pool]
try:
childProcess = subprocess.Popen(checkForCompression, stdin=subprocess.PIPE,
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
except OSError:
stateNum = RaiseStateNum(3, stateNum)
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zpool", nagiosStatus[stateNum])
exit(stateNum)
zpoolString = childProcess.communicate()[0]
zpoolRetval = childProcess.returncode
if zpoolRetval is 1:
stateNum = RaiseStateNum(3, stateNum)
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zpool", nagiosStatus[stateNum])
exit(stateNum)
zpoolLines = zpoolString.splitlines()
zpoolMeta = zpoolLines[0].decode().split()
zpoolMetaStr = ','.join(zpoolMeta)
zpoolEntry = zpoolLines[1].decode().split()
zpoolEntryStr = ','.join(zpoolEntry)
compressName = ''
compressValue = ''
compressRatioName = ''
compressRatioValue = ''
for idx, fieldName in enumerate(zpoolMeta):
if fieldName == 'NAME':
compressName = zpoolEntry[idx]
elif fieldName == 'VALUE':
compressValue = zpoolEntry[idx]
if compressName == '':
stateNum = RaiseStateNum(3, stateNum)
logging.warning("%s : Missing required field in zpool output: NAME", nagiosStatus[stateNum])
exit(stateNum)
if compressValue == 'on':
getCompressRatioCommand = ['/usr/bin/sudo', '-n', zfsCommand, 'get', 'compressratio', args.pool]
try:
childProcess = subprocess.Popen(getCompressRatioCommand, stdin=subprocess.PIPE,
stdout=subprocess.PIPE, stderr=subprocess.PIPE)
except OSError:
stateNum = RaiseStateNum(3, stateNum)
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zpool", nagiosStatus[stateNum])
exit(stateNum)
zpoolString = childProcess.communicate()[0]
zpoolRetval = childProcess.returncode
if zpoolRetval is 1:
stateNum = RaiseStateNum(3, stateNum)
logging.warning("%s : process must be run as root. Possible solution: add the following to\
your visudo: nagios ALL=NOPASSWD: /sbin/zpool", nagiosStatus[stateNum])
exit(stateNum)
zpoolLines = zpoolString.splitlines()
zpoolMeta = zpoolLines[0].decode().split()
zpoolMetaStr = ','.join(zpoolMeta)
zpoolEntry = zpoolLines[1].decode().split()
zpoolEntryStr = ','.join(zpoolEntry)
for idx, fieldName in enumerate(zpoolMeta):
if fieldName == 'NAME':
compressRatioName = zpoolEntry[idx]
elif fieldName == 'VALUE':
compressRatioValue = zpoolEntry[idx]
###################################################################################
##
# OK, finally in the actual status checking of the zpool
# Let's build up our perfdata, regardless of what we're checking
fragPercent=''
if frag!='':
fragPercent=frag.replace("%", "")
fragPerfStr="frag="+str(fragPercent)+"%;"
fragPercent = ''
if frag != '':
fragPercent = frag.replace("%", "")
fragPerfStr = "frag="+str(fragPercent)+"%;"
if checkFragmentation:
fragPerfStr=fragPerfStr+str(fragWarnThreshold)+";"+str(fragCritThreshold)+";"
fragPerfStr = fragPerfStr+str(fragWarnThreshold)+";"+str(fragCritThreshold)+";"
else:
fragPerfStr+=(";;");
perfdata+=(fragPerfStr)
perfdata+=" "
fragPerfStr += (";;")
perfdata += (fragPerfStr)
perfdata += " "
capPercent=''
if cap!='':
capPercent=cap.replace("%", "")
capPerfStr="cap="+str(capPercent)+"%;"
capPercent = ''
if cap != '':
capPercent = cap.replace("%", "")
capPerfStr = "cap="+str(capPercent)+"%;"
if checkCapacity:
capPerfStr=capPerfStr+str(capWarnThreshold)+";"+str(capCritThreshold)+";"
capPerfStr = capPerfStr+str(capWarnThreshold)+";"+str(capCritThreshold)+";"
else:
capPerfStr+=(";;");
perfdata+=(capPerfStr)
perfdata+=" "
capPerfStr += (";;")
perfdata += (capPerfStr)
perfdata += " "
# Perfdata for dedup & compression factor
if dedup != '':
dedup_no_x = dedup.rstrip('x')
perfdata += "dedup="+str(dedup_no_x)
perfdata += " "
if compressRatioValue != '':
compressRatioNoX = compressRatioValue.rstrip('x')
perfdata += "compress_ratio="+str(compressRatioNoX)
perfdata += " "
# Sizes can be in K, M, G, or T (maybe P, but I'm not doing this yet)
if size!='':
if size != '':
sizeGB = ConvertToGB(size)
perfdata+="size="+str(sizeGB)+"GB;;;"
perfdata+=" "
perfdata += "size="+str(sizeGB)+"GB;;;"
perfdata += " "
if alloc!='':
if alloc != '':
allocGB = ConvertToGB(alloc)
perfdata+="alloc="+str(allocGB)+"GB;;;"
perfdata+=" "
perfdata += "alloc="+str(allocGB)+"GB;;;"
perfdata += " "
if free!='':
if free != '':
freeGB = ConvertToGB(free)
perfdata+="free="+str(freeGB)+"GB;;;"
perfdata+=" "
perfdata += "free="+str(freeGB)+"GB;;;"
perfdata += " "
##
# Do mandatory checks
healthNum=-1
if health=='ONLINE':
healthNum=0
elif health=='OFFLINE':
healthNum = -1
if health == 'ONLINE':
healthNum = 0
elif health == 'OFFLINE':
stateNum = RaiseStateNum(1, stateNum)
healthNum=1
elif health=='REMOVED':
healthNum = 1
elif health == 'REMOVED':
stateNum = RaiseStateNum(1, stateNum)
healthNum=2
elif health=='UNAVAIL':
healthNum = 2
elif health == 'UNAVAIL':
stateNum = RaiseStateNum(1, stateNum)
healthNum=3
elif health=='DEGRADED':
healthNum = 3
elif health == 'DEGRADED':
stateNum = RaiseStateNum(2, stateNum)
healthNum=4
elif health=='FAULTED':
healthNum = 4
elif health == 'FAULTED':
stateNum = RaiseStateNum(2, stateNum)
healthNum=5
perfdata+="health="+str(healthNum)+";1;3;"
perfdata+=" "
healthNum = 5
perfdata += "health="+str(healthNum)+";1;3;"
perfdata += " "
##
# Initial part of msg
msg="POOL: "+str(name)
healthMsgFilled=False
msg = "POOL: " + str(name)
healthMsgFilled = False
if healthNum > 0:
msg+=", STATUS: "+str(health)
healthMsgFilled=True
msg += ", STATUS: "+str(health)
healthMsgFilled = True
##
# Do optional checks
fragMsgFilled=False
capMsgFilled=False
if checkFragmentation and fragPercent!='':
if int(fragPercent) > int(fragCritThreshold):
fragMsgFilled=True
stateNum = RaiseStateNum(2, stateNum)
msg+=", FRAG CRIT: "+str(frag)
elif int(fragPercent) > int(fragWarnThreshold):
fragMsgFilled=True
stateNum = RaiseStateNum(1, stateNum)
msg+=", FRAG WARN: "+str(frag)
if checkCapacity and capPercent!='':
fragMsgFilled = False
capMsgFilled = False
if checkFragmentation and fragPercent != '':
if fragPercent.isdigit() is True:
if int(fragPercent) > int(fragCritThreshold):
fragMsgFilled = True
stateNum = RaiseStateNum(2, stateNum)
msg += ", FRAG CRIT: " + str(frag)
elif int(fragPercent) > int(fragWarnThreshold):
fragMsgFilled = True
stateNum = RaiseStateNum(1, stateNum)
msg += ", FRAG WARN: " + str(frag)
if checkCapacity and capPercent != '':
if int(capPercent) > int(capCritThreshold):
capMsgFilled=True
capMsgFilled = True
stateNum = RaiseStateNum(2, stateNum)
msg+=", CAP CRIT: "+str(cap)
msg += ", CAP CRIT: " + str(cap)
elif int(capPercent) > int(capWarnThreshold):
capMsgFilled=True
capMsgFilled = True
stateNum = RaiseStateNum(1, stateNum)
msg+=", CAP WARN: "+str(cap)
msg += ", CAP WARN: "+str(cap)
##
# Build up rest of message
if not healthMsgFilled:
msg+=", STATUS: "+str(health)
if size!='':
msg+=", SIZE: "+str(size)
if alloc!='':
msg+=", ALLOC: "+str(alloc)
if free!='':
msg+=", FREE: "+str(free)
if frag!='' and not fragMsgFilled:
msg+=", FRAG: "+str(frag)
if cap!='' and not capMsgFilled:
msg+=", CAP: "+str(cap)
msg += ", STATUS: " + str(health)
if size != '':
msg += ", SIZE: " + str(size)
if alloc != '':
msg += ", ALLOC: " + str(alloc)
if free != '':
msg += ", FREE: " + str(free)
if dedup != '':
msg += ", DEDUP: " + str(dedup)
if compressRatioValue != '':
msg += ", COMPRESS: " + str(compressRatioValue)
if frag != '' and not fragMsgFilled:
msg += ", FRAG: " + str(frag)
if cap != '' and not capMsgFilled:
msg += ", CAP: " + str(cap)
##
# Print our output and return
print nagiosStatus[stateNum]+": "+msg+" | "+perfdata
logging.warning("%s: %s | %s", nagiosStatus[stateNum], msg, perfdata)
exit(stateNum)