1
|
|
|
# -*- coding: utf-8 -*- |
2
|
|
|
|
3
|
|
|
__author__ = 'Kenny Freeman' |
4
|
|
|
__email__ = '[email protected]' |
5
|
|
|
__license__ = "ISCL" |
6
|
|
|
__docformat__ = 'reStructuredText' |
7
|
|
|
|
8
|
|
|
import time |
9
|
|
|
from collections import deque |
10
|
|
|
|
11
|
|
|
import plumd |
12
|
|
|
import plumd.plugins |
13
|
|
|
from plumd.calc import Differential |
14
|
|
|
from plumd.util import get_file_map |
15
|
|
|
|
16
|
|
|
|
17
|
|
|
## todo: switch from list with pop(0) to deque |
18
|
|
|
class Stat(plumd.plugins.Reader): |
19
|
|
|
"""Class to read metrics from /proc/stat""" |
20
|
|
|
defaults = { |
21
|
|
|
'proc_stat_gauges': ['procs_running', 'procs_blocked'], |
22
|
|
|
'proc_stat_rates': ['intr', 'ctxt', 'softirq'], |
23
|
|
|
'cpu_metrics': ["user", "nice", "system", "idle", "iowait", "irq", |
24
|
|
|
"softirq","steal", "guest", "guest_nice"], |
25
|
|
|
'per_cpu': False, |
26
|
|
|
} |
27
|
|
|
|
28
|
|
View Code Duplication |
def __init__(self, log, config): |
|
|
|
|
29
|
|
|
"""Plugin to measure various kernel metrics from /proc/stat |
30
|
|
|
|
31
|
|
|
:param log: A logger |
32
|
|
|
:type log: logging.RootLogger |
33
|
|
|
:param config: a plumd.config.Conf configuration helper instance. |
34
|
|
|
:type config: plumd.config.Conf |
35
|
|
|
""" |
36
|
|
|
super(Stat, self).__init__(log, config) |
37
|
|
|
self.config.defaults(Stat.defaults) |
38
|
|
|
self.calc = Differential() |
39
|
|
|
self.proc_file = "{0}/stat".format(config.get('proc_path')) |
40
|
|
|
self.per_cpu = self.config.get('per_cpu') |
41
|
|
|
self.cpu_metrics = self.config.get('cpu_metrics') |
42
|
|
|
self.gauges = self.config.get('proc_stat_gauges') |
43
|
|
|
self.rates = self.config.get('proc_stat_rates') |
44
|
|
|
|
45
|
|
|
|
46
|
|
|
def poll(self): |
47
|
|
|
"""Return cpu utilization and process metrics from proc file stat. |
48
|
|
|
|
49
|
|
|
:rtype: plumd.ResultSet |
50
|
|
|
""" |
51
|
|
|
return plumd.ResultSet(self.check()) |
52
|
|
|
|
53
|
|
|
|
54
|
|
|
def check(self): |
55
|
|
|
"""Return cpu utilization and process metrics from proc file stat. |
56
|
|
|
|
57
|
|
|
:rtype: collections.deque |
58
|
|
|
""" |
59
|
|
|
results = deque() |
60
|
|
|
result = plumd.Result("stat") |
61
|
|
|
|
62
|
|
|
dat = get_file_map(self.proc_file, 0, 0) |
63
|
|
|
ts = time.time() |
64
|
|
|
|
65
|
|
|
# record gauges |
66
|
|
|
for i, metric in enumerate(self.gauges): |
67
|
|
|
if metric not in dat: |
68
|
|
|
self.log.warn("stat: unknown metric {0}".format(metric)) |
69
|
|
|
del(self.gauges[i]) |
70
|
|
|
continue |
71
|
|
|
result.add(plumd.Int(metric, dat[metric][0])) |
72
|
|
|
|
73
|
|
|
# record rates |
74
|
|
|
for i, metric in enumerate(self.rates): |
75
|
|
|
if metric not in dat: |
76
|
|
|
self.log.warn("stat: unknown metric {0}".format(metric)) |
77
|
|
View Code Duplication |
del(self.rates[i]) |
|
|
|
|
78
|
|
|
continue |
79
|
|
|
mval = self.calc.per_second(metric, float(dat[metric][0]), ts) |
80
|
|
|
result.add(plumd.Int(metric, mval)) |
81
|
|
|
|
82
|
|
|
# record cpu |
83
|
|
|
if "cpu" in dat: |
84
|
|
|
results.append(self.proc_stat_cpu_percent("cpu", dat["cpu"])) |
85
|
|
|
|
86
|
|
|
# record each cpu if configured |
87
|
|
|
if self.per_cpu: |
88
|
|
|
for i in xrange(0, len(dat)): |
89
|
|
|
mstr = "cpu{0}".format(i) |
90
|
|
|
if mstr not in dat: |
91
|
|
|
break |
92
|
|
|
results.append(self.proc_stat_cpu(mstr, dat[mstr], ts)) |
93
|
|
|
|
94
|
|
|
results.append(result) |
95
|
|
|
return results |
96
|
|
|
|
97
|
|
|
|
98
|
|
|
def proc_stat_cpu_percent(self, key, val): |
99
|
|
View Code Duplication |
"""Return cpu utilization metrics in percentage. |
|
|
|
|
100
|
|
|
|
101
|
|
|
:param key: The metric name (eg. cpu, cpu0, cpu1, etc) |
102
|
|
|
:type key: str |
103
|
|
|
:param val: A deque populated with the metric values from stat |
104
|
|
|
:type val: deque |
105
|
|
|
:rtype: list |
106
|
|
|
""" |
107
|
|
|
result = plumd.Result("cpu") |
108
|
|
|
total = sum([ float(i) for i in val]) |
109
|
|
|
cpu = self.config.get('cpu_metrics') |
110
|
|
|
for map_val in cpu: |
111
|
|
|
if len(val) < 1: |
112
|
|
|
break |
113
|
|
|
metric_val = float(val.popleft()) |
114
|
|
|
mstr = "{0}_{1}".format(key, map_val) |
115
|
|
|
percent_val = metric_val / total * 100.00 |
116
|
|
|
result.add(plumd.Float(mstr, percent_val)) |
117
|
|
|
return result |
118
|
|
|
|
119
|
|
|
|
120
|
|
|
def proc_stat_cpu(self, key, val, ts): |
121
|
|
|
"""Return cpu utilization metrics in USER_HZ or Jiffies |
122
|
|
|
(most likely units of 100Hz intervals ie. 100ms intervals). |
123
|
|
|
|
124
|
|
|
:param key: The metric name (eg. cpu, cpu0, cpu1, etc) |
125
|
|
|
:type key: str |
126
|
|
|
:param val: A deque populated with the metric values from stat |
127
|
|
|
:type val: deque |
128
|
|
|
:rtype: list |
129
|
|
|
""" |
130
|
|
|
result = plumd.Result("cpus") |
131
|
|
|
total = sum([ float(i) for i in val]) |
132
|
|
|
cpu = self.config.get('cpu_metrics') |
133
|
|
|
for map_val in cpu: |
134
|
|
|
if len(val) < 1: |
135
|
|
|
break |
136
|
|
|
metric_val = float(val.popleft()) |
137
|
|
|
mstr = "{0}_{1}".format(key, map_val) |
138
|
|
|
percent_val = float(metric_val / total) * 100.00 |
139
|
|
|
mval = self.calc.per_second(key, percent_val, ts) |
140
|
|
|
result.add(plumd.Float(mstr, mval)) |
141
|
|
|
return result |
142
|
|
|
|