def tolerance_check(
*,
set_sync_time: Optional[float],
levels: Optional[Tuple[float, float]],
notice_only: bool = False,
) -> Generator[Tuple[int, str], None, None]:
if set_sync_time is not None:
set_item_state("time_server", set_sync_time)
return
last_sync = get_item_state("time_server")
now = time.time()
pot_newline = "\n" if notice_only else ""
label = "Time since last sync"
if last_sync is None:
set_item_state("time_server", now)
yield 0, f"{pot_newline}{label}: N/A (started monitoring)"
return
state, text, _metric = check_levels(
now - last_sync,
None,
levels,
human_readable_func=render.timespan,
infoname=label,
)
yield state, text if state else f"{pot_newline}{text}"
def util_counter(stats: CPUInfo, this_time: float) -> CPUInfo:
# Compute jiffi-differences of all relevant counters
diff_values = []
for n, v in enumerate(stats[1:], start=1):
countername = "cpu.util.%d" % n
last_val = get_item_state(countername, (0, 0))[1]
diff_values.append(v - last_val)
set_item_state(countername, (this_time, v))
return CPUInfo(stats.name, *diff_values)
def cpu_util_time(this_time, core, perc, threshold, warn_core, crit_core):
core_state_name = "cpu.util.core.high.%s" % core
if perc > threshold:
timestamp = get_item_state(core_state_name, 0)
high_load_duration = (this_time - timestamp)
state, infotext, _ = check_levels(
high_load_duration,
"%s_is_under_high_load_for" % core, # Not used
(warn_core, crit_core),
human_readable_func=get_age_human_readable,
infoname="%s is under high load for" % core)
if timestamp == 0:
set_item_state(core_state_name, this_time)
elif state:
return state, infotext, []
return 0, "", []
clear_item_state(core_state_name)
return 0, "", []
def _get_value_diff(diff_name, svc_value, timespan):
this_time = time.time()
old_state = get_item_state(diff_name, None)
# first call: take current value as diff or assume 0.0
if old_state is None:
diff_val = 0
set_item_state(diff_name, (this_time, svc_value))
return diff_val
# Get previous value and time difference
last_time, last_val = old_state
timedif = max(this_time - last_time, 0)
if timedif < float(timespan):
diff_val = svc_value - last_val
else:
diff_val = 0
set_item_state(diff_name, (this_time, svc_value))
return diff_val
def check_cpu_util_unix(values: CPUInfo, params, cores=None, values_counter=True):
this_time = time.time()
if values_counter:
diff_values = util_counter(values, this_time)
sum_jiffies = diff_values.total_sum
if sum_jiffies == 0:
raise MKCounterWrapped("Too short time difference since last check")
user_perc, system_perc, wait_perc, steal_perc, guest_perc, util_total_perc = diff_values.utils_perc
else:
user_perc = values.user
system_perc = values.system
wait_perc = values.iowait
steal_perc = values.steal
guest_perc = values.guest
util_total_perc = values.util_total
yield check_levels(user_perc,
'user',
None,
human_readable_func=get_percent_human_readable,
infoname="User")
yield check_levels(system_perc,
'system',
None,
human_readable_func=get_percent_human_readable,
infoname="System")
yield check_levels(wait_perc,
'wait',
params.get('iowait'),
human_readable_func=get_percent_human_readable,
infoname="Wait")
# Compute values used in virtualized environments (Xen, etc.)
# Only do this for counters that have counted at least one tick
# since the system boot. This avoids silly output in systems
# where these counters are not being used
if values.steal:
yield check_levels(steal_perc,
"steal",
params.get('steal'),
human_readable_func=get_percent_human_readable,
infoname="Steal")
if values.guest:
yield check_levels(guest_perc,
'guest',
None,
human_readable_func=get_percent_human_readable,
infoname="Guest")
summary_cores = []
if cores:
for core in cores:
prev_total = get_item_state("cpu.util.%s.total" % core.name, 0)
util_total = core.util_total
total_diff = util_total - prev_total
set_item_state("cpu.util.%s.total" % core.name, util_total)
total_perc = (100.0 * total_diff / sum_jiffies) * len(cores)
summary_cores.append((core.name, total_perc))
for check_result in check_cpu_util(util_total_perc,
params,
this_time,
summary_cores,
perf_max=None):
yield check_result