sensu-plugins-elasticsearch-boutetnico 1.0.4
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 +60 -0
- data/bin/check-es-circuit-breakers.rb +151 -0
- data/bin/check-es-cluster-health.rb +132 -0
- data/bin/check-es-cluster-status.rb +177 -0
- data/bin/check-es-file-descriptors.rb +173 -0
- data/bin/check-es-heap.rb +205 -0
- data/bin/check-es-indexes.rb +89 -0
- data/bin/check-es-indices-field-count.rb +194 -0
- data/bin/check-es-indices-sizes.rb +199 -0
- data/bin/check-es-node-status.rb +137 -0
- data/bin/check-es-query-average.rb +287 -0
- data/bin/check-es-query-count.rb +275 -0
- data/bin/check-es-query-exists.rb +205 -0
- data/bin/check-es-query-ratio.rb +306 -0
- data/bin/check-es-shard-allocation-status.rb +143 -0
- data/bin/handler-es-delete-indices.rb +99 -0
- data/bin/metrics-es-cluster.rb +215 -0
- data/bin/metrics-es-node-graphite.rb +354 -0
- data/bin/metrics-es-node.rb +143 -0
- data/lib/sensu-plugins-elasticsearch.rb +3 -0
- data/lib/sensu-plugins-elasticsearch/elasticsearch-common.rb +63 -0
- data/lib/sensu-plugins-elasticsearch/elasticsearch-query.rb +153 -0
- data/lib/sensu-plugins-elasticsearch/version.rb +9 -0
- metadata +351 -0
@@ -0,0 +1,143 @@
|
|
1
|
+
#! /usr/bin/env ruby
|
2
|
+
#
|
3
|
+
# Checks ElasticSearch shard allocation setting status
|
4
|
+
# ===
|
5
|
+
#
|
6
|
+
# DESCRIPTION:
|
7
|
+
# Checks the ElasticSearch shard allocation persistent and transient settings
|
8
|
+
# and will return status based on a difference in those settings.
|
9
|
+
#
|
10
|
+
# OUTPUT:
|
11
|
+
# plain-text
|
12
|
+
#
|
13
|
+
# PLATFORMS:
|
14
|
+
# Linux
|
15
|
+
#
|
16
|
+
# DEPENDENCIES:
|
17
|
+
# gem: sensu-plugin
|
18
|
+
# gem: rest-client
|
19
|
+
#
|
20
|
+
# USAGE:
|
21
|
+
#
|
22
|
+
# NOTES:
|
23
|
+
#
|
24
|
+
# LICENSE:
|
25
|
+
# Copyright 2014 Yieldbot, Inc <devops@yieldbot.com>
|
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 'rest-client'
|
32
|
+
require 'json'
|
33
|
+
require 'base64'
|
34
|
+
|
35
|
+
#
|
36
|
+
# == Elastic Search Shard Allocation Status
|
37
|
+
#
|
38
|
+
class ESShardAllocationStatus < Sensu::Plugin::Check::CLI
|
39
|
+
option :scheme,
|
40
|
+
description: 'URI scheme',
|
41
|
+
long: '--scheme SCHEME',
|
42
|
+
default: 'http'
|
43
|
+
|
44
|
+
option :server,
|
45
|
+
description: 'Elasticsearch server',
|
46
|
+
short: '-s SERVER',
|
47
|
+
long: '--server SERVER',
|
48
|
+
default: 'localhost'
|
49
|
+
|
50
|
+
option :port,
|
51
|
+
description: 'Port',
|
52
|
+
short: '-p PORT',
|
53
|
+
long: '--port PORT',
|
54
|
+
default: '9200'
|
55
|
+
|
56
|
+
option :allow_non_master,
|
57
|
+
description: 'Allow check to run on non-master nodes',
|
58
|
+
short: '-a',
|
59
|
+
long: '--allow-non-master',
|
60
|
+
default: false
|
61
|
+
|
62
|
+
option :timeout,
|
63
|
+
description: 'Sets the connection timeout for REST client',
|
64
|
+
short: '-t SECS',
|
65
|
+
long: '--timeout SECS',
|
66
|
+
proc: proc(&:to_i),
|
67
|
+
default: 45
|
68
|
+
|
69
|
+
option :user,
|
70
|
+
description: 'Elasticsearch User',
|
71
|
+
short: '-u USER',
|
72
|
+
long: '--user USER'
|
73
|
+
|
74
|
+
option :password,
|
75
|
+
description: 'Elasticsearch Password',
|
76
|
+
short: '-P PASS',
|
77
|
+
long: '--password PASS'
|
78
|
+
|
79
|
+
option :cert_file,
|
80
|
+
description: 'Cert file to use',
|
81
|
+
long: '--cert-file CERT'
|
82
|
+
|
83
|
+
def get_es_resource(resource)
|
84
|
+
headers = {}
|
85
|
+
if config[:user] && config[:password]
|
86
|
+
auth = 'Basic ' + Base64.strict_encode64("#{config[:user]}:#{config[:password]}").chomp
|
87
|
+
headers = { 'Authorization' => auth }
|
88
|
+
end
|
89
|
+
|
90
|
+
r = if config[:cert_file]
|
91
|
+
RestClient::Resource.new("#{config[:scheme]}://#{config[:server]}:#{config[:port]}#{resource}",
|
92
|
+
ssl_ca_file: config[:cert_file].to_s,
|
93
|
+
timeout: config[:timeout],
|
94
|
+
headers: headers)
|
95
|
+
else
|
96
|
+
RestClient::Resource.new("#{config[:scheme]}://#{config[:server]}:#{config[:port]}#{resource}",
|
97
|
+
timeout: config[:timeout],
|
98
|
+
headers: headers)
|
99
|
+
end
|
100
|
+
JSON.parse(r.get)
|
101
|
+
rescue Errno::ECONNREFUSED
|
102
|
+
warning 'Connection refused'
|
103
|
+
rescue RestClient::RequestTimeout
|
104
|
+
critical 'Connection timed out'
|
105
|
+
rescue RestClient::ServiceUnavailable
|
106
|
+
critical 'Service is unavailable'
|
107
|
+
rescue Errno::ECONNRESET
|
108
|
+
critical 'Connection reset by peer'
|
109
|
+
end
|
110
|
+
|
111
|
+
def master?
|
112
|
+
state = get_es_resource('/_cluster/state/master_node')
|
113
|
+
local = get_es_resource('/_nodes/_local')
|
114
|
+
local['nodes'].keys.first == state['master_node']
|
115
|
+
end
|
116
|
+
|
117
|
+
def get_status(type)
|
118
|
+
settings = get_es_resource('/_cluster/settings')
|
119
|
+
# Get the status for the given type, or default to 'all'
|
120
|
+
# which is the ES default
|
121
|
+
begin
|
122
|
+
settings[type]['cluster']['routing']['allocation']['enable'].downcase
|
123
|
+
rescue StandardError
|
124
|
+
'all'
|
125
|
+
end
|
126
|
+
end
|
127
|
+
|
128
|
+
def run
|
129
|
+
if config[:allow_non_master] || master?
|
130
|
+
transient = get_status('transient')
|
131
|
+
persistent = get_status('persistent')
|
132
|
+
|
133
|
+
if transient == persistent
|
134
|
+
ok "Persistent and transient allocation match: #{persistent}"
|
135
|
+
else
|
136
|
+
critical "Persistent(#{persistent}) and transient(#{transient}) \
|
137
|
+
shard allocation do not match."
|
138
|
+
end
|
139
|
+
else
|
140
|
+
ok 'Not the master'
|
141
|
+
end
|
142
|
+
end
|
143
|
+
end
|
@@ -0,0 +1,99 @@
|
|
1
|
+
#! /usr/bin/env ruby
|
2
|
+
#
|
3
|
+
# handler-es-delete-indices.rb
|
4
|
+
#
|
5
|
+
# DESCRIPTION:
|
6
|
+
# This handler deletes indices.
|
7
|
+
#
|
8
|
+
# OUTPUT:
|
9
|
+
# plain text
|
10
|
+
#
|
11
|
+
# PLATFORMS:
|
12
|
+
# Linux
|
13
|
+
#
|
14
|
+
# DEPENDENCIES:
|
15
|
+
# gem: sensu-plugin
|
16
|
+
# gem: elasticsearch
|
17
|
+
# gem: aws_es_transport
|
18
|
+
#
|
19
|
+
# USAGE:
|
20
|
+
# Deletes the indices given to it from a check output and a configured
|
21
|
+
# regex, and then deletes the indices matched.
|
22
|
+
#
|
23
|
+
# NOTES:
|
24
|
+
#
|
25
|
+
# LICENSE:
|
26
|
+
# Brendan Leon Gibat <brendan.gibat@gmail.com>
|
27
|
+
# Released under the same terms as Sensu (the MIT license); see LICENSE
|
28
|
+
# for details.
|
29
|
+
#
|
30
|
+
|
31
|
+
require 'sensu-handler'
|
32
|
+
require 'elasticsearch'
|
33
|
+
require 'aws_es_transport'
|
34
|
+
require 'sensu-plugins-elasticsearch'
|
35
|
+
|
36
|
+
class ESIndexCleanup < Sensu::Handler
|
37
|
+
include ElasticsearchCommon
|
38
|
+
|
39
|
+
option :transport,
|
40
|
+
long: '--transport TRANSPORT',
|
41
|
+
description: 'Transport to use to communicate with ES. Use "AWS" for signed AWS transports.'
|
42
|
+
|
43
|
+
option :region,
|
44
|
+
long: '--region REGION',
|
45
|
+
description: 'Region (necessary for AWS Transport)'
|
46
|
+
|
47
|
+
option :host,
|
48
|
+
description: 'Elasticsearch host',
|
49
|
+
short: '-h HOST',
|
50
|
+
long: '--host HOST',
|
51
|
+
default: 'localhost'
|
52
|
+
|
53
|
+
option :port,
|
54
|
+
description: 'Elasticsearch port',
|
55
|
+
short: '-p PORT',
|
56
|
+
long: '--port PORT',
|
57
|
+
proc: proc(&:to_i),
|
58
|
+
default: 9200
|
59
|
+
|
60
|
+
option :scheme,
|
61
|
+
description: 'Elasticsearch connection scheme, defaults to https for authenticated connections',
|
62
|
+
short: '-s SCHEME',
|
63
|
+
long: '--scheme SCHEME'
|
64
|
+
|
65
|
+
option :password,
|
66
|
+
description: 'Elasticsearch connection password',
|
67
|
+
short: '-P PASSWORD',
|
68
|
+
long: '--password PASSWORD'
|
69
|
+
|
70
|
+
option :user,
|
71
|
+
description: 'Elasticsearch connection user',
|
72
|
+
short: '-u USER',
|
73
|
+
long: '--user USER'
|
74
|
+
|
75
|
+
option :timeout,
|
76
|
+
description: 'Elasticsearch query timeout in seconds',
|
77
|
+
short: '-t TIMEOUT',
|
78
|
+
long: '--timeout TIMEOUT',
|
79
|
+
proc: proc(&:to_i),
|
80
|
+
default: 60
|
81
|
+
|
82
|
+
option :event_regex,
|
83
|
+
description: 'Elasticsearch connection user',
|
84
|
+
short: '-e EVENT_REGEX',
|
85
|
+
long: '--event-regex EVENT_REGEX',
|
86
|
+
default: 'INDEX\[([^\]]+)\]'
|
87
|
+
|
88
|
+
def handle
|
89
|
+
event_regex = Regexp.new(config[:event_regex])
|
90
|
+
indices_to_delete = @event['check']['output'].scan(event_regex).flatten
|
91
|
+
if !indices_to_delete.nil? && !indices_to_delete.empty?
|
92
|
+
|
93
|
+
puts("Deleting indices: [ #{indices_to_delete.sort.join(', ')} ]")
|
94
|
+
client.indices.delete index: indices_to_delete
|
95
|
+
else
|
96
|
+
puts('No indices matched pattern to delete.')
|
97
|
+
end
|
98
|
+
end
|
99
|
+
end
|
@@ -0,0 +1,215 @@
|
|
1
|
+
#! /usr/bin/env ruby
|
2
|
+
#
|
3
|
+
# es-cluster-metrics
|
4
|
+
#
|
5
|
+
# DESCRIPTION:
|
6
|
+
# This plugin uses the ES API to collect metrics, producing a JSON
|
7
|
+
# document which is outputted to STDOUT. An exit status of 0 indicates
|
8
|
+
# the plugin has successfully collected and produced metrics.
|
9
|
+
#
|
10
|
+
# OUTPUT:
|
11
|
+
# metric data
|
12
|
+
#
|
13
|
+
# PLATFORMS:
|
14
|
+
# Linux
|
15
|
+
#
|
16
|
+
# DEPENDENCIES:
|
17
|
+
# gem: sensu-plugin
|
18
|
+
# gem: rest-client
|
19
|
+
#
|
20
|
+
# USAGE:
|
21
|
+
# #YELLOW
|
22
|
+
#
|
23
|
+
# NOTES:
|
24
|
+
#
|
25
|
+
# LICENSE:
|
26
|
+
# Copyright 2011 Sonian, Inc <chefs@sonian.net>
|
27
|
+
# Released under the same terms as Sensu (the MIT license); see LICENSE
|
28
|
+
# for details.
|
29
|
+
#
|
30
|
+
|
31
|
+
require 'sensu-plugin/metric/cli'
|
32
|
+
require 'rest-client'
|
33
|
+
require 'json'
|
34
|
+
require 'base64'
|
35
|
+
|
36
|
+
#
|
37
|
+
# ES Cluster Metrics
|
38
|
+
#
|
39
|
+
class ESClusterMetrics < Sensu::Plugin::Metric::CLI::Graphite
|
40
|
+
option :scheme,
|
41
|
+
description: 'Metric naming scheme, text to prepend to metric',
|
42
|
+
short: '-s SCHEME',
|
43
|
+
long: '--scheme SCHEME',
|
44
|
+
default: "#{Socket.gethostname}.elasticsearch.cluster"
|
45
|
+
|
46
|
+
option :host,
|
47
|
+
description: 'Elasticsearch host',
|
48
|
+
short: '-h HOST',
|
49
|
+
long: '--host HOST',
|
50
|
+
default: 'localhost'
|
51
|
+
|
52
|
+
option :port,
|
53
|
+
description: 'Elasticsearch port',
|
54
|
+
short: '-p PORT',
|
55
|
+
long: '--port PORT',
|
56
|
+
proc: proc(&:to_i),
|
57
|
+
default: 9200
|
58
|
+
|
59
|
+
option :timeout,
|
60
|
+
description: 'Sets the connection timeout for REST client',
|
61
|
+
short: '-t SECS',
|
62
|
+
long: '--timeout SECS',
|
63
|
+
proc: proc(&:to_i),
|
64
|
+
default: 30
|
65
|
+
|
66
|
+
option :allow_non_master,
|
67
|
+
description: 'Allow check to run on non-master nodes',
|
68
|
+
short: '-a',
|
69
|
+
long: '--allow-non-master',
|
70
|
+
default: false
|
71
|
+
|
72
|
+
option :enable_percolate,
|
73
|
+
description: 'Enables percolator stats (ES 2 and older only)',
|
74
|
+
short: '-o',
|
75
|
+
long: '--enable-percolate',
|
76
|
+
default: false
|
77
|
+
|
78
|
+
option :user,
|
79
|
+
description: 'Elasticsearch User',
|
80
|
+
short: '-u USER',
|
81
|
+
long: '--user USER'
|
82
|
+
|
83
|
+
option :password,
|
84
|
+
description: 'Elasticsearch Password',
|
85
|
+
short: '-P PASS',
|
86
|
+
long: '--password PASS'
|
87
|
+
|
88
|
+
option :https,
|
89
|
+
description: 'Enables HTTPS',
|
90
|
+
short: '-e',
|
91
|
+
long: '--https'
|
92
|
+
|
93
|
+
option :cert_file,
|
94
|
+
description: 'Cert file to use',
|
95
|
+
long: '--cert-file CERT_FILE'
|
96
|
+
|
97
|
+
def acquire_es_version
|
98
|
+
info = get_es_resource('/')
|
99
|
+
info['version']['number']
|
100
|
+
end
|
101
|
+
|
102
|
+
def get_es_resource(resource)
|
103
|
+
headers = {}
|
104
|
+
if config[:user] && config[:password]
|
105
|
+
auth = 'Basic ' + Base64.strict_encode64("#{config[:user]}:#{config[:password]}").chomp
|
106
|
+
headers = { 'Authorization' => auth }
|
107
|
+
end
|
108
|
+
|
109
|
+
protocol = if config[:https]
|
110
|
+
'https'
|
111
|
+
else
|
112
|
+
'http'
|
113
|
+
end
|
114
|
+
|
115
|
+
r = if config[:cert_file]
|
116
|
+
RestClient::Resource.new("#{protocol}://#{config[:host]}:#{config[:port]}#{resource}",
|
117
|
+
ssl_ca_file: config[:cert_file].to_s,
|
118
|
+
timeout: config[:timeout],
|
119
|
+
headers: headers)
|
120
|
+
else
|
121
|
+
RestClient::Resource.new("#{protocol}://#{config[:host]}:#{config[:port]}#{resource}",
|
122
|
+
timeout: config[:timeout],
|
123
|
+
headers: headers)
|
124
|
+
end
|
125
|
+
::JSON.parse(r.get)
|
126
|
+
rescue Errno::ECONNREFUSED
|
127
|
+
warning 'Connection refused'
|
128
|
+
rescue RestClient::RequestTimeout
|
129
|
+
warning 'Connection timed out'
|
130
|
+
end
|
131
|
+
|
132
|
+
def master?
|
133
|
+
state = if Gem::Version.new(acquire_es_version) >= Gem::Version.new('3.0.0')
|
134
|
+
get_es_resource('/_cluster/state/master_node')
|
135
|
+
else
|
136
|
+
get_es_resource('/_cluster/state?filter_routing_table=true&filter_metadata=true&filter_indices=true')
|
137
|
+
end
|
138
|
+
local = if Gem::Version.new(acquire_es_version) >= Gem::Version.new('1.0.0')
|
139
|
+
get_es_resource('/_nodes/_local')
|
140
|
+
else
|
141
|
+
get_es_resource('/_cluster/nodes/_local')
|
142
|
+
end
|
143
|
+
local['nodes'].keys.first == state['master_node']
|
144
|
+
end
|
145
|
+
|
146
|
+
def acquire_health
|
147
|
+
health = get_es_resource('/_cluster/health').reject { |k, _v| %w[cluster_name timed_out].include?(k) }
|
148
|
+
health['status'] = %w[red yellow green].index(health['status'])
|
149
|
+
health
|
150
|
+
end
|
151
|
+
|
152
|
+
def acquire_document_count
|
153
|
+
document_count = get_es_resource('/_stats/docs')
|
154
|
+
count = document_count['_all']['total']
|
155
|
+
if count.empty?
|
156
|
+
return 0
|
157
|
+
else
|
158
|
+
return count['docs']['count']
|
159
|
+
end
|
160
|
+
end
|
161
|
+
|
162
|
+
def acquire_cluster_metrics
|
163
|
+
cluster_stats = get_es_resource('/_cluster/stats')
|
164
|
+
cluster_metrics = Hash.new { |h, k| h[k] = {} }
|
165
|
+
cluster_metrics['fs']['total_in_bytes'] = cluster_stats['nodes']['fs']['total_in_bytes']
|
166
|
+
cluster_metrics['fs']['free_in_bytes'] = cluster_stats['nodes']['fs']['free_in_bytes']
|
167
|
+
cluster_metrics['fs']['store_in_bytes'] = cluster_stats['indices']['store']['size_in_bytes']
|
168
|
+
cluster_metrics['fs']['disk_reads'] = cluster_stats['nodes']['fs']['disk_reads']
|
169
|
+
cluster_metrics['fs']['disk_writes'] = cluster_stats['nodes']['fs']['disk_writes']
|
170
|
+
cluster_metrics['fs']['disk_read_size_in_bytes'] = cluster_stats['nodes']['fs']['disk_read_size_in_bytes']
|
171
|
+
cluster_metrics['fs']['disk_write_size_in_bytes'] = cluster_stats['nodes']['fs']['disk_write_size_in_bytes']
|
172
|
+
cluster_metrics['fielddata']['memory_size_in_bytes'] = cluster_stats['indices']['fielddata']['memory_size_in_bytes']
|
173
|
+
cluster_metrics['fielddata']['evictions'] = cluster_stats['indices']['fielddata']['evictions']
|
174
|
+
|
175
|
+
# Elasticsearch changed the name filter_cache to query_cache in 2.0+
|
176
|
+
cache_name = Gem::Version.new(acquire_es_version) < Gem::Version.new('2.0.0') ? 'filter_cache' : 'query_cache'
|
177
|
+
|
178
|
+
cluster_metrics[cache_name]['memory_size_in_bytes'] = cluster_stats['indices'][cache_name]['memory_size_in_bytes']
|
179
|
+
cluster_metrics[cache_name]['evictions'] = cluster_stats['indices'][cache_name]['evictions']
|
180
|
+
cluster_metrics['mem'] = cluster_stats['nodes']['jvm']['mem']
|
181
|
+
|
182
|
+
if config[:enable_percolate] && Gem::Version.new(acquire_es_version) < Gem::Version.new('5.0.0')
|
183
|
+
cluster_metrics['percolate']['total'] = cluster_stats['indices']['percolate']['total']
|
184
|
+
cluster_metrics['percolate']['time_in_millis'] = cluster_stats['indices']['percolate']['time_in_millis']
|
185
|
+
cluster_metrics['percolate']['queries'] = cluster_stats['indices']['percolate']['queries']
|
186
|
+
end
|
187
|
+
cluster_metrics
|
188
|
+
end
|
189
|
+
|
190
|
+
def acquire_allocation_status
|
191
|
+
cluster_config = get_es_resource('/_cluster/settings')
|
192
|
+
transient_settings = cluster_config['transient']
|
193
|
+
if transient_settings.key?('cluster')
|
194
|
+
return %w[none new_primaries primaries all].index(transient_settings['cluster']['routing']['allocation']['enable'])
|
195
|
+
else
|
196
|
+
return nil
|
197
|
+
end
|
198
|
+
end
|
199
|
+
|
200
|
+
def run
|
201
|
+
if config[:allow_non_master] || master?
|
202
|
+
acquire_health.each do |k, v|
|
203
|
+
output(config[:scheme] + '.' + k, v)
|
204
|
+
end
|
205
|
+
acquire_cluster_metrics.each do |cluster_metric|
|
206
|
+
cluster_metric[1].each do |k, v|
|
207
|
+
output(config[:scheme] + '.' + cluster_metric[0] + '.' + k, v || 0)
|
208
|
+
end
|
209
|
+
end
|
210
|
+
output(config[:scheme] + '.document_count', acquire_document_count)
|
211
|
+
output(config[:scheme] + '.allocation_status', acquire_allocation_status) unless acquire_allocation_status.nil?
|
212
|
+
end
|
213
|
+
ok
|
214
|
+
end
|
215
|
+
end
|