def check_memory(c):
'''Collect memory percent, used, free, available'''
# svmem(total=2749374464, available=1501151232, percent=45.4, used=979968000,
# free=736043008, active=1145720832, inactive=590102528, buffers=107663360,
# cached=925700096, shared=86171648)
memory = psutil.virtual_memory()
m1 = CheckItem('memory_percent',memory.percent,"Memory used (percent)", unit = '%')
c.add_item(m1)
m2 = CheckItem('memory_used',memory.used,"Memory used (bytes)", unit='bytes')
h_used = m2.human()
c.add_item(m2)
m3 = CheckItem('memory_available',memory.available,"Memory available (bytes)", unit='bytes')
h_avail = m3.human()
c.add_item(m3)
m4 = CheckItem('memory_total',memory.total,"Memory total (bytes)", unit='bytes')
h_total = m4.human()
c.add_item(m4)
c.add_message("used {} % - used {} - avail {} - total {}".format(memory.percent, h_used, h_avail, h_total) )
return c
def check_url(c):
'''Get a URL and check response code and pattern in the response body.'''
# OUTPUT:
# cmt_url
# cmt_url_name
# cmt_url_msec
name = c.name
url = c.conf['url']
pattern = c.conf.get('pattern',"")
my_redirects = c.conf.get("allow_redirects",False) == True
my_sslverify = c.conf.get("ssl_verify",False) == True
my_host = c.conf.get("host","")
ci = CheckItem('url_name',name,'')
c.add_item(ci)
ci = CheckItem('url',url,'')
c.add_item(ci)
start_time = time.time()
headers = {}
if my_host != "":
headers['Host'] = my_host
try:
resp = requests.get(url, headers = headers, timeout=5, verify=my_sslverify, allow_redirects = my_redirects)
except:
c.alert += 1
c.add_message("{} - {} [Host: {}] - no response to query".format(name,url, my_host))
return c
elapsed_time = int ( 1000 * (time.time() - start_time) )
ci = CheckItem('url_msec',elapsed_time,unit='ms')
c.add_item(ci)
if resp.status_code != 200:
c.alert += 1
c.add_message("{} - {} [Host: {}]- bad http code response ({})".format(name,url, my_host, resp.status_code))
return c
# check pattern
mysearch = re.search(pattern,resp.text)
if not mysearch:
c.alert += 1
c.add_message("expected pattern not found for {} ({} [Host: {}])".format(name, url, my_sslverify))
return c
c.add_message("{} - {} [Host: {}] - http={} - {} ms ; pattern OK".format(name, url, my_host, resp.status_code, elapsed_time))
return c
def check_mount(c):
'''Checks mount points
OUTPUT
- cmt_mount
'''
partitions = psutil.disk_partitions(all=False)
if cmt.ARGS["available"]:
print("-" * 25)
print("Available mountpoints :")
print("-" * 25)
for p in partitions:
print(p)
print("-" * 25)
return c
#sdiskpart(device='/dev/sda1', mountpoint='/', fstype='ext4',
# opts='rw,relatime,errors=remount-ro,data=ordered')
path = c.conf['path']
ci = CheckItem('mount', path)
c.add_item(ci)
for part in partitions:
if part.mountpoint == path:
c.add_message("path {} found".format(path))
return c
c.alert += 1
c.add_message("path {} not found".format(path))
return c
def check_ping(c):
'''Ping a remote host and return availability
Output:
- cmt_ping
'''
host = c.conf['host']
ci = CheckItem('ping', host)
c.add_item(ci)
#response = os.system("ping -c 1 -W 2 " + host + "> /dev/null 2>&1")
proc = subprocess.Popen(
["ping", "-c", "1", "-W"
"2", host],
stdout=subprocess.DEVNULL,
stderr=subprocess.DEVNULL,
) # starts the process
proc.wait() # waits for it to finish (there's a timeout option)
response = proc.returncode
if response == 0:
c.add_message("{} ok".format(host))
else:
c.alert += 1
c.add_message("{} not responding".format(host))
return c
def check_cpu(c):
'''Get CPU percentage. No alert. Send cpu float value.'''
cpu = psutil.cpu_percent(interval=2)
# c.persist['cpu'] = cpu
i = CheckItem('cpu', cpu, "CPU Percentage", unit='%')
c.add_item(i)
c.add_message("usage : {} %".format(cpu))
return c
def check_boottime(c):
#c = Check(module='boottime')
boottime = int(time.time() - psutil.boot_time())
days = int(boottime / 86400)
m1 = CheckItem('boottime_seconds',
boottime,
"Seconds since last reboot",
unit='seconds')
h_sec = m1.human()
c.add_item(m1)
m2 = CheckItem('boottime_days',
days,
'Days since last reboot',
unit='days')
c.add_item(m2)
c.add_message("days since last reboot : {} days - {} sec.".format(
days, h_sec))
return c
def check_load(c):
# (1.17, 0.86, 0.52)
load = os.getloadavg()
l1 = CheckItem('load1',load[0])
l1.description ='CPU Load Average, one minute'
c.add_item(l1)
l5 = CheckItem('load5',load[1])
l5.description='CPU Load Average, 5 minutes'
c.add_item(l5)
l15 = CheckItem('load15',load[2])
l15.description='CPU Load Average, 15 minutes'
c.add_item(l15)
c.add_message("1/5/15 min : {} {} {}".format(load[0], load[1], load[2]))
return c
def check_swap(c):
# sswap(total=2147479552, used=0, free=2147479552, percent=0.0, sin=0, sout=0)
swap = psutil.swap_memory()
m1 = CheckItem('swap_percent',swap.percent,"Swap used (percent)", unit = '%')
c.add_item(m1)
m2 = CheckItem('swap_used',swap.used,'Swap used (bytes)', unit = 'bytes')
h_used = m2.human()
c.add_item(m2)
m3 = CheckItem('swap_total',swap.total,'Swap total (bytes)', unit = 'bytes')
h_total = m3.human()
c.add_item(m3)
c.add_message("used: {} % / {} - total {}".format(swap.percent, h_used, h_total))
return c
def check_process(c):
# --available option ?
if cmt.ARGS["available"]:
print("-" * 25)
print("Process available :")
print("-" * 25)
for p in psutil.process_iter():
try:
# Get process name & pid from process object.
print(p.name())
#print(processName , ' ::: ', processID)
except (psutil.NoSuchProcess, psutil.AccessDenied,
psutil.ZombieProcess):
pass
print("-" * 25)
return c
# real check
name = c.name
psname = c.conf['psname']
#{'name': 'python3', 'cpu_times': pcputimes(user=0.39, system=0.3,
# children_user=0.0, children_system=0.0),
# 'memory_info': pmem(rss=27049984, vms=123904000, shared=13443072, text=3883008,
# lib=0, data=13901824, dirty=0), 'username': '******', 'pid': 3125}
#for proc in psutil.process_iter(['pid', 'name', 'username','cpu_times','memory_info']):
# #print(proc.info)
ci = CheckItem('process_name', name, "")
c.add_item(ci)
for proc in psutil.process_iter():
try:
# Get process name & pid from process object.
processName = proc.name()
processID = proc.pid
#print(processName , ' ::: ', processID)
except (psutil.NoSuchProcess, psutil.AccessDenied,
psutil.ZombieProcess):
pass
# pinfo = proc.as_dict(attrs=['name','pid','memory_info','cpu_times'])
if processName == psname:
mem = proc.memory_info().rss
ci = CheckItem('process_memory', mem, "rss", unit="bytes")
h_mem = ci.human()
c.add_item(ci)
cpu = proc.cpu_times().user
ci = CheckItem('process_cpu',
cpu,
"cpu time, user",
unit='seconds')
c.add_item(ci)
c.add_message("{} found ({}) - memory rss {} - cpu {} sec.".format(
name, psname, h_mem, cpu))
return c
c.alert += 1
c.add_message("{} missing ({})".format(name, psname))
return c
def check_folder(c):
'''Check for various folder attributes '''
global s_dirs
path = c.conf['path']
name = path
if 'name' in c.conf:
name = c.conf['name']
recursive = c.conf.get("recursive",False) == True
targets = []
if 'target' in c.conf:
targets = c.conf['target']
ci = CheckItem('folder_path',path)
c.add_item(ci)
ci = CheckItem('folder_name',name)
c.add_item(ci)
if not os.path.exists(path):
c.alert += 1
c.add_message("{} missing".format(path))
return c
# scan
# ----
s_count = 0
s_size = 0
s_minage = -1
s_maxage = 0
s_files = []
#for entry in os.scandir(path):
if recursive:
for entry in scanRecurse(path):
s_files.append(entry.name)
s_count += 1
statinfo = os.stat(entry.path)
s_size += statinfo.st_size
if statinfo.st_mtime > s_maxage:
s_maxage = statinfo.st_mtime
if statinfo.st_mtime < s_minage or s_minage == -1 :
s_minage = statinfo.st_mtime
else:
for entry in scanNoRecurse(path):
s_files.append(entry.name)
s_count += 1
statinfo = os.stat(entry.path)
s_size += statinfo.st_size
if statinfo.st_mtime > s_maxage:
s_maxage = statinfo.st_mtime
if statinfo.st_mtime < s_minage or s_minage == -1 :
s_minage = statinfo.st_mtime
# file count
ci = CheckItem('folder_files', s_count, "Number of files in folder " + name, unit="files")
c.add_item(ci)
# dirs count
ci = CheckItem('folder_dirs',s_dirs,"Number of dirs/subdirs in folder " + name, unit="dirs")
c.add_item(ci)
# size
ci = CheckItem('folder_size',s_size,"Total Size (bytes)", unit="bytes")
c.add_item(ci)
# age
now = time.time()
if s_maxage > 0:
ci = CheckItem('folder_age_min',"","Min age (seconds)", unit="sec")
ci.value = int(now - s_maxage)
c.add_item(ci)
if s_minage != -1:
ci = CheckItem('folder_age_max',"","Max age (seconds)", unit="sec")
ci.value = int(now - s_minage)
c.add_item(ci)
# Target checks
# --------------
tgcount = 0
# target : files_min: 4
if 'files_min' in targets:
tgcount += 1
if s_count < targets['files_min']:
c.alert += 1
c.add_message ("{} : too few files ({})".format(path,s_count))
return c
# target : files_max: 23
if 'files_max' in targets:
tgcount += 1
if s_count > targets['files_max']:
c.alert += 1
c.add_message ("{} : too many files ({})".format(path,s_count))
return c
# target : size_max (folder max bytes)
if 'size_max' in targets:
tgcount += 1
if s_size > targets['size_max']:
c.alert += 1
c.add_message("{} : too big ({})".format(path,s_size))
return c
# target : size_min (folder min bytes)
if 'size_min' in targets:
tgcount += 1
if s_size < targets['size_min']:
c.alert += 1
c.add_message("{} : too small ({})".format(path,s_size))
return c
# target : age_max:
if 'age_max' in targets:
tgcount += 1
if s_minage != -1:
if int(now - s_minage) > targets ['age_max']:
c.alert += 1
c.add_message("{} : some files too old ({} sec)".format(path,int(now - s_minage)))
return c
# target : age_min:
if 'age_min' in targets:
tgcount += 1
if s_maxage != 0:
if int(now - s_maxage) < targets ['age_min']:
c.alert += 1
c.add_message("{} : some files too young ({} sec)".format(path,int(now - s_maxage)))
return c
# target : has_file: filename
if 'has_files' in targets:
tgcount += 1
flist = targets['has_files']
for f in flist:
if f not in s_files:
c.alert += 1
c.add_message("{} : expected file not found ({})".format(path,f))
return c
c.add_message("{} ({}) ok - {} files, {} dirs, {} bytes - {} targets OK".format(name, path, s_count, s_dirs, s_size, tgcount))
return c
def check_disk(c):
#c = Check(module='disk')
path = c.conf['path']
alert_threshold = int(c.conf['alert'])
# sdiskusage(total=21378641920, used=4809781248, free=15482871808, percent=22.5)
disk=psutil.disk_usage(path)
ci = CheckItem('disk',path,"Path")
c.add_item(ci)
ci = CheckItem('disk_total',disk[0],"Total (Bytes)", unit='bytes')
h_total = ci.human()
c.add_item(ci)
ci = CheckItem('disk_used',disk[1],"Used (Bytes)", unit='bytes')
h_used = ci.human()
c.add_item(ci)
ci = CheckItem('disk_free',disk[2],"Free (Bytes)", unit='bytes')
h_free = ci.human()
c.add_item(ci)
ci = CheckItem('disk_percent',disk[3],"Used (percent)", unit='%')
if disk[3] > alert_threshold:
c.alert += 1
c.add_message("path : {} - critical capacity ({} %)".format(path,disk[3]))
else:
c.add_message("path : {} - used: {} % - used: {} - free: {} - total: {} ".format(path, disk[3],h_used, h_free,h_total))
c.add_item(ci)
return c