sensu-plugins-graphite-boutetnico 1.0.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/CHANGELOG.md +1 -0
- data/LICENSE +22 -0
- data/README.md +126 -0
- data/bin/check-graphite-data.rb +115 -0
- data/bin/check-graphite-hosts.rb +97 -0
- data/bin/check-graphite-replication.rb +234 -0
- data/bin/check-graphite-stats.rb +158 -0
- data/bin/check-graphite.rb +542 -0
- data/bin/handler-graphite-event.rb +86 -0
- data/bin/handler-graphite-notify.rb +41 -0
- data/bin/handler-graphite-occurrences.rb +39 -0
- data/bin/handler-graphite-status.rb +42 -0
- data/bin/mutator-graphite.rb +44 -0
- data/lib/sensu-plugins-graphite.rb +1 -0
- data/lib/sensu-plugins-graphite/graphite_proxy/options.rb +119 -0
- data/lib/sensu-plugins-graphite/graphite_proxy/proxy.rb +106 -0
- data/lib/sensu-plugins-graphite/version.rb +9 -0
- metadata +275 -0
@@ -0,0 +1,158 @@
|
|
1
|
+
#! /usr/bin/env ruby
|
2
|
+
#
|
3
|
+
# check-stats
|
4
|
+
#
|
5
|
+
# DESCRIPTION:
|
6
|
+
# Checks metrics in Graphite, averaged over a period of time.
|
7
|
+
#
|
8
|
+
# The fired Sensu event will only be critical if a stat is
|
9
|
+
# above the critical threshold. Otherwise, the event will be warning,
|
10
|
+
# if a stat is above the warning threshold.
|
11
|
+
#
|
12
|
+
# Multiple stats will be checked if * are used
|
13
|
+
# in the "target" query.
|
14
|
+
#
|
15
|
+
# OUTPUT:
|
16
|
+
# plain text
|
17
|
+
#
|
18
|
+
# PLATFORMS:
|
19
|
+
# Linux
|
20
|
+
#
|
21
|
+
# DEPENDENCIES:
|
22
|
+
# gem: sensu-plugin
|
23
|
+
#
|
24
|
+
# USAGE:
|
25
|
+
# example commands
|
26
|
+
#
|
27
|
+
# NOTES:
|
28
|
+
#
|
29
|
+
# LICENSE:
|
30
|
+
# Alan Smith (alan@asmith.me)
|
31
|
+
# Released under the same terms as Sensu (the MIT license); see LICENSE
|
32
|
+
# for details.
|
33
|
+
#
|
34
|
+
|
35
|
+
require 'json'
|
36
|
+
require 'net/http'
|
37
|
+
require 'sensu-plugin/check/cli'
|
38
|
+
|
39
|
+
class CheckGraphiteStat < Sensu::Plugin::Check::CLI
|
40
|
+
option :host,
|
41
|
+
short: '-h HOST',
|
42
|
+
long: '--host HOST',
|
43
|
+
description: 'Graphite hostname',
|
44
|
+
proc: proc(&:to_s),
|
45
|
+
default: 'graphite'
|
46
|
+
|
47
|
+
option :period,
|
48
|
+
short: '-p PERIOD',
|
49
|
+
long: '--period PERIOD',
|
50
|
+
description: 'The period back in time to extract from Graphite. Use -24hours, -2days, -15mins, etc, same format as in Graphite',
|
51
|
+
proc: proc(&:to_s),
|
52
|
+
required: true
|
53
|
+
|
54
|
+
option :target,
|
55
|
+
short: '-t TARGET',
|
56
|
+
long: '--target TARGET',
|
57
|
+
description: 'The Graphite metric name. Can include * to query multiple metrics',
|
58
|
+
proc: proc(&:to_s),
|
59
|
+
required: true
|
60
|
+
|
61
|
+
option :warn,
|
62
|
+
short: '-w WARN',
|
63
|
+
long: '--warn WARN',
|
64
|
+
description: 'Warning level',
|
65
|
+
proc: proc(&:to_f),
|
66
|
+
required: false
|
67
|
+
|
68
|
+
option :crit,
|
69
|
+
short: '-c CRIT',
|
70
|
+
long: '--crit CRIT',
|
71
|
+
description: 'Critical level',
|
72
|
+
proc: proc(&:to_f),
|
73
|
+
required: false
|
74
|
+
|
75
|
+
option :unknown_ignore,
|
76
|
+
short: '-u',
|
77
|
+
long: '--unknown-ignore',
|
78
|
+
description: "Do nothing for UNKNOWN status (when you wildcard-match a ton of metrics at once and you don't care about a few missing data)",
|
79
|
+
boolean: true,
|
80
|
+
default: false
|
81
|
+
|
82
|
+
option :reverse_scale,
|
83
|
+
short: '-r',
|
84
|
+
long: '--reverse-scale',
|
85
|
+
description: 'Reverse the warn/crit scale (if value is less than instead of greater than)',
|
86
|
+
boolean: true,
|
87
|
+
default: false
|
88
|
+
|
89
|
+
def average(datapoints)
|
90
|
+
total = 0
|
91
|
+
datapoints.to_a.each { |i| total += i.to_f }
|
92
|
+
|
93
|
+
total / datapoints.length
|
94
|
+
end
|
95
|
+
|
96
|
+
def danger(metric)
|
97
|
+
datapoints = metric['datapoints'].map(&:first).compact
|
98
|
+
|
99
|
+
# #YELLOW
|
100
|
+
unless datapoints.empty? # rubocop:disable Style/UnlessElse
|
101
|
+
avg = average(datapoints)
|
102
|
+
if config[:reverse_scale] == false
|
103
|
+
if !config[:crit].nil? && avg > config[:crit]
|
104
|
+
return [2, "#{metric['target']} is #{avg}"]
|
105
|
+
elsif !config[:warn].nil? && avg > config[:warn]
|
106
|
+
return [1, "#{metric['target']} is #{avg}"]
|
107
|
+
end
|
108
|
+
else
|
109
|
+
if !config[:crit].nil? && avg < config[:crit] # rubocop:disable Style/IfInsideElse
|
110
|
+
return [2, "#{metric['target']} is #{avg}"]
|
111
|
+
elsif !config[:warn].nil? && avg < config[:warn]
|
112
|
+
return [1, "#{metric['target']} is #{avg}"]
|
113
|
+
end
|
114
|
+
end
|
115
|
+
else
|
116
|
+
return [3, "#{metric['target']} has no datapoints"] unless config[:unknown_ignore]
|
117
|
+
end
|
118
|
+
[0, nil]
|
119
|
+
end
|
120
|
+
|
121
|
+
def run
|
122
|
+
body =
|
123
|
+
begin
|
124
|
+
uri = URI.parse(CGI.escape("http://#{config[:host]}/render?format=json&target=#{config[:target]}&from=#{config[:period]}"))
|
125
|
+
res = Net::HTTP.get_response(uri)
|
126
|
+
res.body
|
127
|
+
rescue => e
|
128
|
+
warning "Failed to query Graphite: #{e.inspect}"
|
129
|
+
end
|
130
|
+
|
131
|
+
status = 0
|
132
|
+
message = ''
|
133
|
+
data =
|
134
|
+
begin
|
135
|
+
JSON.parse(body)
|
136
|
+
rescue
|
137
|
+
[]
|
138
|
+
end
|
139
|
+
|
140
|
+
unknown 'No data from Graphite' if data.empty?
|
141
|
+
|
142
|
+
data.each do |metric|
|
143
|
+
s, msg = danger(metric)
|
144
|
+
|
145
|
+
message += "#{msg} " unless s.zero?
|
146
|
+
status = s unless s < status
|
147
|
+
end
|
148
|
+
|
149
|
+
if status == 2
|
150
|
+
critical message
|
151
|
+
elsif status == 1
|
152
|
+
warning message
|
153
|
+
elsif status == 3
|
154
|
+
unknown message
|
155
|
+
end
|
156
|
+
ok
|
157
|
+
end
|
158
|
+
end
|
@@ -0,0 +1,542 @@
|
|
1
|
+
#! /usr/bin/env ruby
|
2
|
+
#
|
3
|
+
# <script name>
|
4
|
+
#
|
5
|
+
# DESCRIPTION:
|
6
|
+
# Get time series values from Graphite and create events based on values
|
7
|
+
#
|
8
|
+
# OUTPUT:
|
9
|
+
# plain text
|
10
|
+
#
|
11
|
+
# PLATFORMS:
|
12
|
+
# Linux
|
13
|
+
#
|
14
|
+
# DEPENDENCIES:
|
15
|
+
# gem: sensu-plugin
|
16
|
+
# gem: array_stats
|
17
|
+
#
|
18
|
+
# USAGE:
|
19
|
+
# #YELLOW
|
20
|
+
#
|
21
|
+
# NOTES:
|
22
|
+
#
|
23
|
+
# LICENSE:
|
24
|
+
# Copyright 2012 Ulf Mansson @ Recorded Future
|
25
|
+
# Modifications by Chris Jansen to support wildcard targets
|
26
|
+
# Released under the same terms as Sensu (the MIT license); see LICENSE
|
27
|
+
# for details.
|
28
|
+
#
|
29
|
+
|
30
|
+
require 'sensu-plugin/check/cli'
|
31
|
+
require 'json'
|
32
|
+
require 'net/http'
|
33
|
+
require 'net/https'
|
34
|
+
require 'socket'
|
35
|
+
require 'array_stats'
|
36
|
+
|
37
|
+
class Graphite < Sensu::Plugin::Check::CLI
|
38
|
+
option :host,
|
39
|
+
short: '-h HOST',
|
40
|
+
long: '--host HOST',
|
41
|
+
description: 'Graphite host to connect to, include port',
|
42
|
+
required: true
|
43
|
+
|
44
|
+
option :target,
|
45
|
+
description: 'The graphite metric name. Could be a comma separated list of metric names.',
|
46
|
+
short: '-t TARGET',
|
47
|
+
long: '--target TARGET',
|
48
|
+
required: true
|
49
|
+
|
50
|
+
option :complex_target,
|
51
|
+
description: 'Allows complex targets which contain functions. Disables splitting on comma.',
|
52
|
+
short: '-x',
|
53
|
+
long: '--complex_target',
|
54
|
+
default: false
|
55
|
+
|
56
|
+
option :period,
|
57
|
+
description: 'The period back in time to extract from Graphite and compare with. Use 24hours,2days etc, same format as in Graphite',
|
58
|
+
short: '-p PERIOD',
|
59
|
+
long: '--period PERIOD',
|
60
|
+
default: '2hours'
|
61
|
+
|
62
|
+
option :updated_since,
|
63
|
+
description: 'The graphite value should have been updated within UPDATED_SINCE seconds, default to 600 seconds',
|
64
|
+
short: '-u UPDATED_SINCE',
|
65
|
+
long: '--updated_since UPDATED_SINCE',
|
66
|
+
default: 600
|
67
|
+
|
68
|
+
option :acceptable_diff_percentage,
|
69
|
+
description: 'The acceptable diff from max values in percentage, used in check_function_increasing',
|
70
|
+
short: '-D ACCEPTABLE_DIFF_PERCENTAGE',
|
71
|
+
long: '--acceptable_diff_percentage ACCEPTABLE_DIFF_PERCENTAGE',
|
72
|
+
default: 0
|
73
|
+
|
74
|
+
option :check_function_increasing,
|
75
|
+
description: 'Check that value is increasing or equal over time (use acceptable_diff_percentage if it should allow to be lower)',
|
76
|
+
short: '-i',
|
77
|
+
long: '--check_function_increasing',
|
78
|
+
default: false,
|
79
|
+
boolean: true
|
80
|
+
|
81
|
+
option :greater_than,
|
82
|
+
description: 'Change whether value is greater than or less than check',
|
83
|
+
short: '-g',
|
84
|
+
long: '--greater_than',
|
85
|
+
default: false
|
86
|
+
|
87
|
+
option :check_last,
|
88
|
+
description: 'Check that the last value in GRAPHITE is greater/less than VALUE',
|
89
|
+
short: '-l VALUE',
|
90
|
+
long: '--last VALUE',
|
91
|
+
default: nil
|
92
|
+
|
93
|
+
option :ignore_nulls,
|
94
|
+
description: 'Do not error on null values, used in check_function_increasing',
|
95
|
+
short: '-n',
|
96
|
+
long: '--ignore_nulls',
|
97
|
+
default: false,
|
98
|
+
boolean: true
|
99
|
+
|
100
|
+
option :concat_output,
|
101
|
+
description: 'Include warning messages in output even if overall status is critical',
|
102
|
+
short: '-c',
|
103
|
+
long: '--concat_output',
|
104
|
+
default: false,
|
105
|
+
boolean: true
|
106
|
+
|
107
|
+
option :short_output,
|
108
|
+
description: 'Report only the highest status per series in output',
|
109
|
+
short: '-s',
|
110
|
+
long: '--short_output',
|
111
|
+
default: false,
|
112
|
+
boolean: true
|
113
|
+
|
114
|
+
option :check_average,
|
115
|
+
description: 'MAX_VALUE should be greater than the average of Graphite values from PERIOD',
|
116
|
+
short: '-a MAX_VALUE',
|
117
|
+
long: '--average_value MAX_VALUE'
|
118
|
+
|
119
|
+
option :data_points,
|
120
|
+
description: 'Number of data points to include in average check (smooths out spikes)',
|
121
|
+
short: '-d VALUE',
|
122
|
+
long: '--data_points VALUE',
|
123
|
+
default: 1
|
124
|
+
|
125
|
+
option :check_average_percent,
|
126
|
+
description: 'MAX_VALUE% should be greater than the average of Graphite values from PERIOD',
|
127
|
+
short: '-b MAX_VALUE',
|
128
|
+
long: '--average_percent_value MAX_VALUE'
|
129
|
+
|
130
|
+
option :percentile,
|
131
|
+
description: 'Percentile value, should be used in conjunction with percentile_value, defaults to 90',
|
132
|
+
long: '--percentile PERCENTILE',
|
133
|
+
default: 90
|
134
|
+
|
135
|
+
option :check_percentile,
|
136
|
+
description: 'Values should not be greater than the VALUE of Graphite values from PERIOD',
|
137
|
+
long: '--percentile_value VALUE'
|
138
|
+
|
139
|
+
option :http_user,
|
140
|
+
description: 'Basic HTTP authentication user',
|
141
|
+
short: '-U USER',
|
142
|
+
long: '--http-user USER',
|
143
|
+
default: nil
|
144
|
+
|
145
|
+
option :http_password,
|
146
|
+
description: 'Basic HTTP authentication password',
|
147
|
+
short: '-P PASSWORD',
|
148
|
+
long: '--http-password USER',
|
149
|
+
default: nil
|
150
|
+
|
151
|
+
def initialize
|
152
|
+
super
|
153
|
+
@graphite_cache = {}
|
154
|
+
end
|
155
|
+
|
156
|
+
def graphite_cache(target = nil)
|
157
|
+
# #YELLOW
|
158
|
+
if @graphite_cache.key?(target)
|
159
|
+
graphite_value = @graphite_cache[target].select { |value| value[:period] == @period }
|
160
|
+
graphite_value if graphite_value.size > 0
|
161
|
+
end
|
162
|
+
end
|
163
|
+
|
164
|
+
# Create a graphite url from params
|
165
|
+
#
|
166
|
+
#
|
167
|
+
def graphite_url(target = nil)
|
168
|
+
url = "#{config[:host]}/render/"
|
169
|
+
url = 'http://' + url unless url[0..3] == 'http'
|
170
|
+
# #YELLOW
|
171
|
+
url = url + "?target=#{target}" if target # rubocop:disable Style/SelfAssignment
|
172
|
+
URI.parse(url)
|
173
|
+
end
|
174
|
+
|
175
|
+
def get_levels(config_param)
|
176
|
+
values = config_param.split(',')
|
177
|
+
i = 0
|
178
|
+
levels = {}
|
179
|
+
%w[warning error fatal].each do |type|
|
180
|
+
levels[type] = values[i] if values[i]
|
181
|
+
i += 1
|
182
|
+
end
|
183
|
+
levels
|
184
|
+
end
|
185
|
+
|
186
|
+
def get_graphite_values(target)
|
187
|
+
cache_value = graphite_cache target
|
188
|
+
return cache_value if cache_value
|
189
|
+
|
190
|
+
params = {
|
191
|
+
target: target,
|
192
|
+
from: "-#{@period}",
|
193
|
+
format: 'json'
|
194
|
+
}
|
195
|
+
|
196
|
+
req = Net::HTTP::Post.new(graphite_url.path)
|
197
|
+
|
198
|
+
# If the basic http authentication credentials have been provided, then use them
|
199
|
+
if !config[:http_user].nil? && !config[:http_password].nil?
|
200
|
+
req.basic_auth(config[:http_user], config[:http_password])
|
201
|
+
end
|
202
|
+
|
203
|
+
req.set_form_data(params)
|
204
|
+
nethttp = Net::HTTP.new(graphite_url.host, graphite_url.port)
|
205
|
+
if graphite_url.scheme == 'https'
|
206
|
+
nethttp.use_ssl = true
|
207
|
+
end
|
208
|
+
resp = nethttp.start { |http| http.request(req) }
|
209
|
+
|
210
|
+
data = JSON.parse(resp.body)
|
211
|
+
@graphite_cache[target] = []
|
212
|
+
if data.size > 0
|
213
|
+
data.each { |d| @graphite_cache[target] << { target: d['target'], period: @period, datapoints: d['datapoints'] } }
|
214
|
+
graphite_cache target
|
215
|
+
end
|
216
|
+
end
|
217
|
+
|
218
|
+
# Will give max values for [0..-2]
|
219
|
+
def max_graphite_value(target)
|
220
|
+
max_values = {}
|
221
|
+
values = get_graphite_values target
|
222
|
+
if values
|
223
|
+
values.each do |val|
|
224
|
+
max = get_max_value(val[:datapoints])
|
225
|
+
max_values[val[:target]] = max
|
226
|
+
end
|
227
|
+
end
|
228
|
+
max_values
|
229
|
+
end
|
230
|
+
|
231
|
+
def get_max_value(values)
|
232
|
+
if values
|
233
|
+
values.map { |i| i[0] || 0 }[0..-2].max
|
234
|
+
end
|
235
|
+
end
|
236
|
+
|
237
|
+
def last_graphite_metric(target, count = 1)
|
238
|
+
last_values = {}
|
239
|
+
values = get_graphite_values target
|
240
|
+
if values
|
241
|
+
values.each do |val|
|
242
|
+
last = get_last_metric(val[:datapoints], count)
|
243
|
+
last_values[val[:target]] = last
|
244
|
+
end
|
245
|
+
end
|
246
|
+
last_values
|
247
|
+
end
|
248
|
+
|
249
|
+
def get_last_metric(values, count = 1)
|
250
|
+
if values
|
251
|
+
ret = []
|
252
|
+
values_size = values.size
|
253
|
+
count = values_size if count > values_size
|
254
|
+
while count > 0
|
255
|
+
values_size -= 1
|
256
|
+
break if values[values_size].nil?
|
257
|
+
|
258
|
+
count -= 1 if values[values_size][0]
|
259
|
+
ret.push(values[values_size]) if values[values_size][0]
|
260
|
+
end
|
261
|
+
ret
|
262
|
+
end
|
263
|
+
end
|
264
|
+
|
265
|
+
def last_graphite_value(target, count = 1)
|
266
|
+
last_metrics = last_graphite_metric(target, count)
|
267
|
+
last_values = {}
|
268
|
+
if last_metrics
|
269
|
+
last_metrics.each do |target_name, metrics|
|
270
|
+
last_values[target_name] = metrics.map { |metric| metric[0] }.mean
|
271
|
+
end
|
272
|
+
end
|
273
|
+
last_values
|
274
|
+
end
|
275
|
+
|
276
|
+
def been_updated_since(target, time, updated_since)
|
277
|
+
last_time_stamp = last_graphite_metric target
|
278
|
+
warnings = []
|
279
|
+
if last_time_stamp
|
280
|
+
last_time_stamp.each do |target_name, value|
|
281
|
+
last_time_stamp_bool = value[1] > time.to_i
|
282
|
+
warnings << "The metric #{target_name} has not been updated in #{updated_since} seconds" unless last_time_stamp_bool
|
283
|
+
end
|
284
|
+
end
|
285
|
+
warnings
|
286
|
+
end
|
287
|
+
|
288
|
+
def greater_less
|
289
|
+
return 'greater' if config[:greater_than]
|
290
|
+
return 'less' unless config[:greater_than]
|
291
|
+
end
|
292
|
+
|
293
|
+
def check_increasing(target)
|
294
|
+
updated_since = config[:updated_since].to_i
|
295
|
+
time_to_be_updated_since = Time.now - updated_since
|
296
|
+
critical_errors = []
|
297
|
+
warnings = []
|
298
|
+
max_gv = max_graphite_value target
|
299
|
+
last_gv = last_graphite_value target
|
300
|
+
if last_gv.is_a?(Hash) && max_gv.is_a?(Hash)
|
301
|
+
# #YELLOW
|
302
|
+
last_gv.each do |target_name, value|
|
303
|
+
if value && max_gv[target_name]
|
304
|
+
last = value
|
305
|
+
max = max_gv[target_name]
|
306
|
+
if max > last * (1 + config[:acceptable_diff_percentage].to_f / 100)
|
307
|
+
msg = "The metric #{target} with last value #{last} is less than max value #{max} during #{config[:period]} period"
|
308
|
+
critical_errors << msg
|
309
|
+
end
|
310
|
+
end
|
311
|
+
end
|
312
|
+
else
|
313
|
+
warnings << "Could not found any value in Graphite for metric #{target}, see #{graphite_url(target)}"
|
314
|
+
end
|
315
|
+
unless config[:ignore_nulls]
|
316
|
+
warnings.concat(been_updated_since(target, time_to_be_updated_since, updated_since))
|
317
|
+
end
|
318
|
+
[warnings, critical_errors, []]
|
319
|
+
end
|
320
|
+
|
321
|
+
def check_average_percent(target, max_values, data_points = 1)
|
322
|
+
values = get_graphite_values target
|
323
|
+
last_values = last_graphite_value(target, data_points)
|
324
|
+
return [[], [], []] unless values
|
325
|
+
|
326
|
+
warnings = []
|
327
|
+
criticals = []
|
328
|
+
fatal = []
|
329
|
+
values.each do |data| # rubocop:disable Metrics/BlockLength
|
330
|
+
target = data[:target]
|
331
|
+
values_pair = data[:datapoints]
|
332
|
+
values_array = values_pair.select(&:first).map { |v| v.first unless v.first.nil? }
|
333
|
+
# #YELLOW
|
334
|
+
avg_value = values_array.reduce { |sum, el| sum + el if el }.to_f / values_array.size
|
335
|
+
last_value = last_values[target]
|
336
|
+
percent = last_value / avg_value unless last_value.nil? || avg_value.nil?
|
337
|
+
# #YELLOW
|
338
|
+
%w[fatal error warning].each do |type|
|
339
|
+
next unless max_values.key?(type)
|
340
|
+
|
341
|
+
max_value = max_values[type]
|
342
|
+
var1 = config[:greater_than] ? percent : max_value.to_f
|
343
|
+
var2 = config[:greater_than] ? max_value.to_f : percent
|
344
|
+
if !percent.nil? && var1 > var2 && (values_array.size > 0 || !config[:ignore_nulls])
|
345
|
+
text = "The last value of metric #{target} is #{percent}% #{greater_less} than allowed #{max_value}% of the average value #{avg_value}"
|
346
|
+
case type
|
347
|
+
when 'warning'
|
348
|
+
warnings << text
|
349
|
+
when 'error'
|
350
|
+
criticals << text
|
351
|
+
when 'fatal'
|
352
|
+
fatal << text
|
353
|
+
else
|
354
|
+
raise "Unknown type #{type}"
|
355
|
+
end
|
356
|
+
break if config[:short_output]
|
357
|
+
end
|
358
|
+
end
|
359
|
+
end
|
360
|
+
[warnings, criticals, fatal]
|
361
|
+
end
|
362
|
+
|
363
|
+
def check_average(target, max_values)
|
364
|
+
values = get_graphite_values target
|
365
|
+
return [[], [], []] unless values
|
366
|
+
|
367
|
+
warnings = []
|
368
|
+
criticals = []
|
369
|
+
fatal = []
|
370
|
+
values.each do |data|
|
371
|
+
target = data[:target]
|
372
|
+
values_pair = data[:datapoints]
|
373
|
+
values_array = values_pair.select(&:first).map { |v| v.first unless v.first.nil? }
|
374
|
+
# #YELLOW
|
375
|
+
avg_value = values_array.reduce { |sum, el| sum + el if el }.to_f / values_array.size
|
376
|
+
# YELLOW
|
377
|
+
%w[fatal error warning].each do |type|
|
378
|
+
next unless max_values.key?(type)
|
379
|
+
|
380
|
+
max_value = max_values[type]
|
381
|
+
var1 = config[:greater_than] ? avg_value : max_value.to_f
|
382
|
+
var2 = config[:greater_than] ? max_value.to_f : avg_value
|
383
|
+
if var1 > var2 && (values_array.size > 0 || !config[:ignore_nulls])
|
384
|
+
text = "The average value of metric #{target} is #{avg_value} that is #{greater_less} than allowed average of #{max_value}"
|
385
|
+
case type
|
386
|
+
when 'warning'
|
387
|
+
warnings << text
|
388
|
+
when 'error'
|
389
|
+
criticals << text
|
390
|
+
when 'fatal'
|
391
|
+
fatal << text
|
392
|
+
else
|
393
|
+
raise "Unknown type #{type}"
|
394
|
+
end
|
395
|
+
break if config[:short_output]
|
396
|
+
end
|
397
|
+
end
|
398
|
+
end
|
399
|
+
[warnings, criticals, fatal]
|
400
|
+
end
|
401
|
+
|
402
|
+
def check_percentile(target, max_values, percentile, data_points = 1)
|
403
|
+
values = get_graphite_values target
|
404
|
+
last_values = last_graphite_value(target, data_points)
|
405
|
+
return [[], [], []] unless values
|
406
|
+
|
407
|
+
warnings = []
|
408
|
+
criticals = []
|
409
|
+
fatal = []
|
410
|
+
values.each do |data| # rubocop:disable Metrics/BlockLength
|
411
|
+
target = data[:target]
|
412
|
+
values_pair = data[:datapoints]
|
413
|
+
values_array = values_pair.select(&:first).map { |v| v.first unless v.first.nil? }
|
414
|
+
percentile_value = values_array.percentile(percentile)
|
415
|
+
last_value = last_values[target]
|
416
|
+
percent = last_value / percentile_value unless last_value.nil? || percentile_value.nil?
|
417
|
+
# #YELLOW
|
418
|
+
%w[fatal error warning].each do |type|
|
419
|
+
next unless max_values.key?(type)
|
420
|
+
|
421
|
+
max_value = max_values[type]
|
422
|
+
var1 = config[:greater_than] ? percent : max_value.to_f
|
423
|
+
var2 = config[:greater_than] ? max_value.to_f : percent
|
424
|
+
if !percentile_value.nil? && var1 > var2
|
425
|
+
text = "The percentile value of metric #{target} (#{last_value}) is #{greater_less} than the
|
426
|
+
#{percentile}th percentile (#{percentile_value}) by more than #{max_value}%"
|
427
|
+
case type
|
428
|
+
when 'warning'
|
429
|
+
warnings << text
|
430
|
+
when 'error'
|
431
|
+
criticals << text
|
432
|
+
when 'fatal'
|
433
|
+
fatal << text
|
434
|
+
else
|
435
|
+
raise "Unknown type #{type}"
|
436
|
+
end
|
437
|
+
break if config[:short_output]
|
438
|
+
end
|
439
|
+
end
|
440
|
+
end
|
441
|
+
[warnings, criticals, fatal]
|
442
|
+
end
|
443
|
+
|
444
|
+
def check_last(target, max_values)
|
445
|
+
last_targets = last_graphite_value target
|
446
|
+
return [[], [], []] unless last_targets
|
447
|
+
|
448
|
+
warnings = []
|
449
|
+
criticals = []
|
450
|
+
fatal = []
|
451
|
+
# #YELLOW
|
452
|
+
last_targets.each do |target_name, last_value|
|
453
|
+
unless last_value.nil?
|
454
|
+
# #YELLOW
|
455
|
+
%w[fatal error warning].each do |type|
|
456
|
+
next unless max_values.key?(type)
|
457
|
+
|
458
|
+
max_value = max_values[type]
|
459
|
+
var1 = config[:greater_than] ? last_value : max_value.to_f
|
460
|
+
var2 = config[:greater_than] ? max_value.to_f : last_value
|
461
|
+
if var1 > var2
|
462
|
+
text = "The metric #{target_name} is #{last_value} that is #{greater_less} than last allowed #{max_value}"
|
463
|
+
case type
|
464
|
+
when 'warning'
|
465
|
+
warnings << text
|
466
|
+
when 'error'
|
467
|
+
criticals << text
|
468
|
+
when 'fatal'
|
469
|
+
fatal << text
|
470
|
+
else
|
471
|
+
raise "Unknown type #{type}"
|
472
|
+
end
|
473
|
+
break if config[:short_output]
|
474
|
+
end
|
475
|
+
end
|
476
|
+
end
|
477
|
+
end
|
478
|
+
[warnings, criticals, fatal]
|
479
|
+
end
|
480
|
+
|
481
|
+
def run
|
482
|
+
targets = config[:complex_target] ? [config[:target]] : config[:target].split(',')
|
483
|
+
@period = config[:period]
|
484
|
+
critical_errors = []
|
485
|
+
warnings = []
|
486
|
+
fatals = []
|
487
|
+
# #YELLOW
|
488
|
+
targets.each do |target| # rubocop:disable Metrics/BlockLength
|
489
|
+
if config[:check_function_increasing]
|
490
|
+
inc_warnings, inc_critical, inc_fatal = check_increasing target
|
491
|
+
warnings += inc_warnings
|
492
|
+
critical_errors += inc_critical
|
493
|
+
fatals += inc_fatal
|
494
|
+
end
|
495
|
+
if config[:check_last]
|
496
|
+
max_values = get_levels config[:check_last]
|
497
|
+
lt_warnings, lt_critical, lt_fatal = check_last(target, max_values)
|
498
|
+
warnings += lt_warnings
|
499
|
+
critical_errors += lt_critical
|
500
|
+
fatals += lt_fatal
|
501
|
+
end
|
502
|
+
if config[:check_average]
|
503
|
+
max_values = get_levels config[:check_average]
|
504
|
+
avg_warnings, avg_critical, avg_fatal = check_average(target, max_values)
|
505
|
+
warnings += avg_warnings
|
506
|
+
critical_errors += avg_critical
|
507
|
+
fatals += avg_fatal
|
508
|
+
end
|
509
|
+
if config[:check_average_percent]
|
510
|
+
max_values = get_levels config[:check_average_percent]
|
511
|
+
avg_warnings, avg_critical, avg_fatal = check_average_percent(target, max_values, config[:data_points].to_i)
|
512
|
+
warnings += avg_warnings
|
513
|
+
critical_errors += avg_critical
|
514
|
+
fatals += avg_fatal
|
515
|
+
end
|
516
|
+
if config[:check_percentile]
|
517
|
+
max_values = get_levels config[:check_percentile]
|
518
|
+
pct_warnings, pct_critical, pct_fatal = check_percentile(target, max_values, config[:percentile].to_i, config[:data_points].to_i)
|
519
|
+
warnings += pct_warnings
|
520
|
+
critical_errors += pct_critical
|
521
|
+
fatals += pct_fatal
|
522
|
+
end
|
523
|
+
end
|
524
|
+
fatals_string = fatals.size > 0 ? fatals.join("\n") : ''
|
525
|
+
criticals_string = critical_errors.size > 0 ? critical_errors.join("\n") : ''
|
526
|
+
warnings_string = warnings.size > 0 ? warnings.join("\n") : ''
|
527
|
+
|
528
|
+
if config[:concat_output]
|
529
|
+
fatals_string = fatals_string + "\n" + criticals_string if critical_errors.size > 0
|
530
|
+
fatals_string = fatals_string + "\nGraphite WARNING: " + warnings_string if warnings.size > 0
|
531
|
+
criticals_string = criticals_string + "\nGraphite WARNING: " + warnings_string if warnings.size > 0
|
532
|
+
critical fatals_string if fatals.size > 0
|
533
|
+
critical criticals_string if critical_errors.size > 0
|
534
|
+
warning warnings_string if warnings.size > 0 # rubocop:disable Style/IdenticalConditionalBranches
|
535
|
+
else
|
536
|
+
critical fatals_string if fatals.size > 0
|
537
|
+
critical criticals_string if critical_errors.size > 0
|
538
|
+
warning warnings_string if warnings.size > 0 # rubocop:disable Style/IdenticalConditionalBranches
|
539
|
+
end
|
540
|
+
ok
|
541
|
+
end
|
542
|
+
end
|