cpuidle.chart.py: add plugin for monitoring cpuidle states
Signed-off-by: Steven Noonan <steven@uplinklabs.net>
Steven Noonan committed
Jan 9, 2017 at 14:20 UTC
70daed00e400613b10a059fe176463c43a16ade0
3 files changed
+156
python.d/Makefile.am
+1
@@ -12,6 +12,7 @@ dist_python_SCRIPTS = \
12
apache_cache.chart.py \
13
bind_rndc.chart.py \
14
cpufreq.chart.py \
15
+ cpuidle.chart.py \
16
dovecot.chart.py \
17
example.chart.py \
18
exim.chart.py \
python.d/README.md
+12
@@ -209,6 +209,18 @@ Directory is also prefixed with `NETDATA_HOST_PREFIX` if specified.
209
210
---
211
212
+# cpuidle
213
+
214
+This module monitors the usage of CPU idle states.
215
+
216
+**Requirement:**
217
+Your kernel needs to have `CONFIG_CPU_IDLE` enabled.
218
+
219
+It produces one stacked chart per CPU, showing the percentage of time spent in
220
+each state.
221
+
222
+---
223
+
224
# dovecot
225
226
This module provides statistics information from dovecot server.
python.d/cpuidle.chart.py
new
+143
@@ -0,0 +1,143 @@
1
+# -*- coding: utf-8 -*-
2
+# Description: cpuidle netdata python.d module
3
+# Author: Steven Noonan (tycho)
4
+
5
+import glob
6
+import os
7
+import platform
8
+import time
9
+from base import SimpleService
10
+
11
+import ctypes
12
+syscall = ctypes.CDLL('libc.so.6').syscall
13
+
14
+# default module values (can be overridden per job in `config`)
15
+# update_every = 2
16
+
17
+class Service(SimpleService):
18
+ def __init__(self, configuration=None, name=None):
19
+ prefix = os.getenv('NETDATA_HOST_PREFIX', "")
20
+ if prefix.endswith('/'):
21
+ prefix = prefix[:-1]
22
+ self.sys_dir = prefix + "/sys/devices/system/cpu"
23
+ self.schedstat_path = prefix + "/proc/schedstat"
24
+ SimpleService.__init__(self, configuration=configuration, name=name)
25
+ self.order = []
26
+ self.definitions = {}
27
+ self._orig_name = ""
28
+ self.assignment = {}
29
+
30
+ def __gettid(self):
31
+ # This is horrendous. We need the *thread id* (not the *process id*),
32
+ # but there's no Python standard library way of doing that. If you need
33
+ # to enable this module on a non-x86 machine type, you'll have to find
34
+ # the Linux syscall number for gettid() and add it to the dictionary
35
+ # below.
36
+ syscalls = {
37
+ 'i386': 224,
38
+ 'x86_64': 186,
39
+ }
40
+ if platform.machine() not in syscalls:
41
+ return None
42
+ tid = syscall(syscalls[platform.machine()])
43
+ return tid
44
+
45
+ def __wake_cpus(self):
46
+ # Requires Python 3.3+. This will "tickle" each CPU to force it to
47
+ # update its idle counters.
48
+ if hasattr(os, 'sched_setaffinity'):
49
+ pid = self.__gettid()
50
+ save_affinity = os.sched_getaffinity(pid)
51
+ for idx in range(0, len(self.assignment)):
52
+ os.sched_setaffinity(pid, [idx])
53
+ os.sched_getaffinity(pid)
54
+ os.sched_setaffinity(pid, save_affinity)
55
+
56
+ def __read_schedstat(self):
57
+ cpus = {}
58
+ for line in open(self.schedstat_path, 'r'):
59
+ if not line.startswith('cpu'):
60
+ continue
61
+ line = line.rstrip().split()
62
+ cpu = line[0]
63
+ active_time = line[7]
64
+ cpus[cpu] = int(active_time) // 1000
65
+ return cpus
66
+
67
+ def _get_data(self):
68
+ results = {}
69
+
70
+ # This line is critical for the stats to update. If we don't "tickle"
71
+ # all the CPUs, then all the counters stop counting.
72
+ self.__wake_cpus()
73
+
74
+ # Use the kernel scheduler stats to determine how much time was spent
75
+ # in C0 (active).
76
+ schedstat = self.__read_schedstat()
77
+
78
+ for cpu, metrics in self.assignment.items():
79
+ update_time = schedstat[cpu]
80
+ results[cpu + '_active_time'] = update_time
81
+
82
+ for metric, path in metrics.items():
83
+ residency = int(open(path, 'r').read())
84
+ results[metric] = residency
85
+
86
+ return results
87
+
88
+ def check(self):
89
+ if self.__gettid() is None:
90
+ self.error("Cannot get thread ID. Stats would be completely broken.")
91
+ return False
92
+
93
+ self._orig_name = self.chart_name
94
+
95
+ for path in sorted(glob.glob(self.sys_dir + '/cpu*/cpuidle/state*/name')):
96
+ # ['', 'sys', 'devices', 'system', 'cpu', 'cpu0', 'cpuidle', 'state3', 'name']
97
+ path_elem = path.split('/')
98
+ cpu = path_elem[-4]
99
+ state = path_elem[-2]
100
+ statename = open(path, 'rt').read().rstrip()
101
+
102
+ orderid = '%s_cpuidle' % (cpu,)
103
+ if orderid not in self.definitions:
104
+ self.order.append(orderid)
105
+ active_name = '%s_active_time' % (cpu,)
106
+ self.definitions[orderid] = {
107
+ 'options': [None, 'C-state residency', 'time%', 'cpuidle', None, 'stacked'],
108
+ 'lines': [
109
+ [active_name, 'C0 (active)', 'percentage-of-incremental-row', 1, 1],
110
+ ],
111
+ }
112
+ self.assignment[cpu] = {}
113
+
114
+ defid = '%s_%s_time' % (orderid, state)
115
+
116
+ self.definitions[orderid]['lines'].append(
117
+ [defid, statename, 'percentage-of-incremental-row', 1, 1]
118
+ )
119
+
120
+ self.assignment[cpu][defid] = '/'.join(path_elem[:-1] + ['time'])
121
+
122
+ # Sort order by kernel-specified CPU index
123
+ self.order.sort(key=lambda x: int(x.split('_')[0][3:]))
124
+
125
+ if len(self.definitions) == 0:
126
+ self.error("couldn't find cstate stats")
127
+ return False
128
+
129
+ return True
130
+
131
+ def create(self):
132
+ self.chart_name = "cpu"
133
+ status = SimpleService.create(self)
134
+ self.chart_name = self._orig_name
135
+ return status
136
+
137
+ def update(self, interval):
138
+ self.chart_name = "cpu"
139
+ status = SimpleService.update(self, interval=interval)
140
+ self.chart_name = self._orig_name
141
+ return status
142
+
143
+# vim: set ts=4 sts=4 sw=4 et: