gitlab-exporter 13.1.0 → 13.2.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/Gemfile.lock +1 -1
- data/config/gitlab-exporter.yml.example +2 -0
- data/lib/gitlab_exporter/sidekiq.rb +122 -84
- data/lib/gitlab_exporter/version.rb +1 -1
- data/spec/sidekiq_spec.rb +64 -0
- metadata +3 -1
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA256:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 1faa75a67146917f990466c949a63cc5b53c4c4280bcaae76601071a88c5fb0c
|
4
|
+
data.tar.gz: acd50e384eb4e03fdbf2ba20c8f030a47b886822a3e87cdddded7326790e5f25
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 6dbab1d6bce735e67ced073506ca2e9ddf1111ccd1b4a6e45a2289ce5a0ed069f5f004c2cda272a298d6328f7fe0fbd96cff30d434c722f01c1f1448b49199ce
|
7
|
+
data.tar.gz: c625442a214949794bad144d38a511c96161af0261380cac50f509f421c7da03c6549b94059451322c9f23498ca33ccc9ce528c02f4874357c79b21bf9486e04
|
data/Gemfile.lock
CHANGED
@@ -7,6 +7,7 @@ module GitLab
|
|
7
7
|
# A prober for Sidekiq queues
|
8
8
|
#
|
9
9
|
# It takes the Redis URL Sidekiq is connected to
|
10
|
+
# rubocop:disable Metrics/ClassLength
|
10
11
|
class SidekiqProber
|
11
12
|
# The maximum depth (from the head) of each queue to probe. Probing the
|
12
13
|
# entirety of a very large queue will take longer and run the risk of
|
@@ -37,35 +38,24 @@ module GitLab
|
|
37
38
|
@opts = opts
|
38
39
|
@metrics = metrics
|
39
40
|
@logger = logger
|
41
|
+
|
42
|
+
@probe_namespaced = !!opts[:probe_namespaced] # rubocop:disable Style/DoubleNegation
|
43
|
+
@probe_non_namespaced = !!opts[:probe_non_namespaced] # rubocop:disable Style/DoubleNegation
|
44
|
+
|
45
|
+
# to maintain backward compatibility if the config is missing values
|
46
|
+
@probe_namespaced = true if opts[:probe_namespaced].nil? && opts[:probe_non_namespaced].nil?
|
40
47
|
end
|
41
48
|
|
42
49
|
def probe_stats
|
43
|
-
with_sidekiq
|
44
|
-
|
45
|
-
|
46
|
-
@metrics.add("sidekiq_jobs_processed_total", stats.processed)
|
47
|
-
@metrics.add("sidekiq_jobs_failed_total", stats.failed)
|
48
|
-
@metrics.add("sidekiq_jobs_enqueued_size", stats.enqueued)
|
49
|
-
@metrics.add("sidekiq_jobs_scheduled_size", stats.scheduled_size)
|
50
|
-
@metrics.add("sidekiq_jobs_retry_size", stats.retry_size)
|
51
|
-
@metrics.add("sidekiq_jobs_dead_size", stats.dead_size)
|
52
|
-
|
53
|
-
@metrics.add("sidekiq_default_queue_latency_seconds", stats.default_queue_latency)
|
54
|
-
@metrics.add("sidekiq_processes_size", stats.processes_size)
|
55
|
-
@metrics.add("sidekiq_workers_size", stats.workers_size)
|
56
|
-
end
|
50
|
+
with_sidekiq { _probe_stats } if @probe_namespaced
|
51
|
+
with_sidekiq(false) { _probe_stats(false) } if @probe_non_namespaced
|
57
52
|
|
58
53
|
self
|
59
54
|
end
|
60
55
|
|
61
56
|
def probe_queues
|
62
|
-
with_sidekiq
|
63
|
-
|
64
|
-
@metrics.add("sidekiq_queue_size", queue.size, name: queue.name)
|
65
|
-
@metrics.add("sidekiq_queue_latency_seconds", queue.latency, name: queue.name)
|
66
|
-
@metrics.add("sidekiq_queue_paused", queue.paused? ? 1 : 0, name: queue.name)
|
67
|
-
end
|
68
|
-
end
|
57
|
+
with_sidekiq { _probe_queues } if @probe_namespaced
|
58
|
+
with_sidekiq(false) { _probe_queues(false) } if @probe_non_namespaced
|
69
59
|
|
70
60
|
self
|
71
61
|
end
|
@@ -78,24 +68,8 @@ module GitLab
|
|
78
68
|
end
|
79
69
|
|
80
70
|
def probe_future_sets
|
81
|
-
|
82
|
-
with_sidekiq
|
83
|
-
Sidekiq.redis do |conn|
|
84
|
-
Sidekiq::Scheduled::SETS.each do |set|
|
85
|
-
# Default to 0; if all jobs are due in the future, there is no "negative" delay.
|
86
|
-
delay = 0
|
87
|
-
|
88
|
-
_job, timestamp = conn.zrangebyscore(set, "-inf", now.to_s, limit: [0, 1], withscores: true).first
|
89
|
-
delay = now - timestamp if timestamp
|
90
|
-
|
91
|
-
@metrics.add("sidekiq_#{set}_set_processing_delay_seconds", delay)
|
92
|
-
|
93
|
-
# zcount is O(log(N)) (prob. binary search), so is still quick even with large sets
|
94
|
-
@metrics.add("sidekiq_#{set}_set_backlog_count",
|
95
|
-
conn.zcount(set, "-inf", now.to_s))
|
96
|
-
end
|
97
|
-
end
|
98
|
-
end
|
71
|
+
with_sidekiq { _probe_future_sets } if @probe_namespaced
|
72
|
+
with_sidekiq(false) { _probe_future_sets(false) } if @probe_non_namespaced
|
99
73
|
end
|
100
74
|
|
101
75
|
# Count worker classes present in Sidekiq queues. This only looks at the
|
@@ -105,57 +79,22 @@ module GitLab
|
|
105
79
|
# will not have completely accurate statistics, but the probe performance
|
106
80
|
# will also not degrade as the queue gets larger.
|
107
81
|
def probe_jobs_limit
|
108
|
-
with_sidekiq
|
109
|
-
|
110
|
-
|
111
|
-
Sidekiq::Queue.all.each do |queue|
|
112
|
-
Sidekiq.redis do |conn|
|
113
|
-
conn.lrange("queue:#{queue.name}", 0, PROBE_JOBS_LIMIT).each do |job|
|
114
|
-
job_class = Sidekiq.load_json(job)["class"]
|
115
|
-
|
116
|
-
job_stats[job_class] += 1
|
117
|
-
end
|
118
|
-
end
|
119
|
-
end
|
120
|
-
|
121
|
-
job_stats.each do |class_name, count|
|
122
|
-
@metrics.add("sidekiq_enqueued_jobs", count, name: class_name)
|
123
|
-
end
|
124
|
-
end
|
82
|
+
with_sidekiq { _probe_jobs_limit } if @probe_namespaced
|
83
|
+
with_sidekiq(false) { _probe_jobs_limit(false) } if @probe_non_namespaced
|
125
84
|
|
126
85
|
self
|
127
86
|
end
|
128
87
|
|
129
88
|
def probe_workers
|
130
|
-
with_sidekiq
|
131
|
-
|
132
|
-
|
133
|
-
Sidekiq::Workers.new.map do |_pid, _tid, work|
|
134
|
-
job_klass = work["payload"]["class"]
|
135
|
-
|
136
|
-
worker_stats[job_klass] += 1
|
137
|
-
end
|
138
|
-
|
139
|
-
worker_stats.each do |class_name, count|
|
140
|
-
@metrics.add("sidekiq_running_jobs", count, name: class_name)
|
141
|
-
end
|
142
|
-
end
|
89
|
+
with_sidekiq { _probe_workers } if @probe_namespaced
|
90
|
+
with_sidekiq(false) { _probe_workers(false) } if @probe_non_namespaced
|
143
91
|
|
144
92
|
self
|
145
93
|
end
|
146
94
|
|
147
95
|
def probe_retries
|
148
|
-
with_sidekiq
|
149
|
-
|
150
|
-
|
151
|
-
Sidekiq::RetrySet.new.map do |job|
|
152
|
-
retry_stats[job.klass] += 1
|
153
|
-
end
|
154
|
-
|
155
|
-
retry_stats.each do |class_name, count|
|
156
|
-
@metrics.add("sidekiq_to_be_retried_jobs", count, name: class_name)
|
157
|
-
end
|
158
|
-
end
|
96
|
+
with_sidekiq { _probe_retries } if @probe_namespaced
|
97
|
+
with_sidekiq(false) { _probe_retries(false) } if @probe_non_namespaced
|
159
98
|
|
160
99
|
self
|
161
100
|
end
|
@@ -177,12 +116,106 @@ module GitLab
|
|
177
116
|
|
178
117
|
private
|
179
118
|
|
180
|
-
def
|
119
|
+
def _probe_workers(namespace = true)
|
120
|
+
labels = add_namespace_labels? ? { namespaced: namespace } : {}
|
121
|
+
worker_stats = Hash.new(0)
|
122
|
+
|
123
|
+
Sidekiq::Workers.new.map do |_pid, _tid, work|
|
124
|
+
job_klass = work["payload"]["class"]
|
125
|
+
|
126
|
+
worker_stats[job_klass] += 1
|
127
|
+
end
|
128
|
+
|
129
|
+
worker_stats.each do |class_name, count|
|
130
|
+
@metrics.add("sidekiq_running_jobs", count, **labels.merge({ name: class_name }))
|
131
|
+
end
|
132
|
+
end
|
133
|
+
|
134
|
+
def _probe_future_sets(namespace = true)
|
135
|
+
labels = add_namespace_labels? ? { namespaced: namespace } : {}
|
136
|
+
now = Time.now.to_f
|
137
|
+
Sidekiq.redis do |conn|
|
138
|
+
Sidekiq::Scheduled::SETS.each do |set|
|
139
|
+
# Default to 0; if all jobs are due in the future, there is no "negative" delay.
|
140
|
+
delay = 0
|
141
|
+
|
142
|
+
_job, timestamp = conn.zrangebyscore(set, "-inf", now.to_s, limit: [0, 1], withscores: true).first
|
143
|
+
delay = now - timestamp if timestamp
|
144
|
+
|
145
|
+
@metrics.add("sidekiq_#{set}_set_processing_delay_seconds", delay, **labels)
|
146
|
+
|
147
|
+
# zcount is O(log(N)) (prob. binary search), so is still quick even with large sets
|
148
|
+
@metrics.add("sidekiq_#{set}_set_backlog_count",
|
149
|
+
conn.zcount(set, "-inf", now.to_s), **labels)
|
150
|
+
end
|
151
|
+
end
|
152
|
+
end
|
153
|
+
|
154
|
+
def _probe_stats(namespace = true)
|
155
|
+
stats = Sidekiq::Stats.new
|
156
|
+
labels = add_namespace_labels? ? { namespaced: namespace } : {}
|
157
|
+
|
158
|
+
@metrics.add("sidekiq_jobs_processed_total", stats.processed, **labels)
|
159
|
+
@metrics.add("sidekiq_jobs_failed_total", stats.failed, **labels)
|
160
|
+
@metrics.add("sidekiq_jobs_enqueued_size", stats.enqueued, **labels)
|
161
|
+
@metrics.add("sidekiq_jobs_scheduled_size", stats.scheduled_size, **labels)
|
162
|
+
@metrics.add("sidekiq_jobs_retry_size", stats.retry_size, **labels)
|
163
|
+
@metrics.add("sidekiq_jobs_dead_size", stats.dead_size, **labels)
|
164
|
+
|
165
|
+
@metrics.add("sidekiq_default_queue_latency_seconds", stats.default_queue_latency, **labels)
|
166
|
+
@metrics.add("sidekiq_processes_size", stats.processes_size, **labels)
|
167
|
+
@metrics.add("sidekiq_workers_size", stats.workers_size, **labels)
|
168
|
+
end
|
169
|
+
|
170
|
+
def _probe_queues(namespace = true)
|
171
|
+
Sidekiq::Queue.all.each do |queue|
|
172
|
+
labels = { name: queue.name }
|
173
|
+
labels[:namespaced] = namespace if add_namespace_labels?
|
174
|
+
|
175
|
+
@metrics.add("sidekiq_queue_size", queue.size, **labels)
|
176
|
+
@metrics.add("sidekiq_queue_latency_seconds", queue.latency, **labels)
|
177
|
+
@metrics.add("sidekiq_queue_paused", queue.paused? ? 1 : 0, **labels)
|
178
|
+
end
|
179
|
+
end
|
180
|
+
|
181
|
+
def _probe_jobs_limit(namespace = true)
|
182
|
+
labels = add_namespace_labels? ? { namespaced: namespace } : {}
|
183
|
+
job_stats = Hash.new(0)
|
184
|
+
|
185
|
+
Sidekiq::Queue.all.each do |queue|
|
186
|
+
Sidekiq.redis do |conn|
|
187
|
+
conn.lrange("queue:#{queue.name}", 0, PROBE_JOBS_LIMIT).each do |job|
|
188
|
+
job_class = Sidekiq.load_json(job)["class"]
|
189
|
+
|
190
|
+
job_stats[job_class] += 1
|
191
|
+
end
|
192
|
+
end
|
193
|
+
end
|
194
|
+
|
195
|
+
job_stats.each do |class_name, count|
|
196
|
+
@metrics.add("sidekiq_enqueued_jobs", count, **labels.merge({ name: class_name }))
|
197
|
+
end
|
198
|
+
end
|
199
|
+
|
200
|
+
def _probe_retries(namespace = true)
|
201
|
+
labels = add_namespace_labels? ? { namespaced: namespace } : {}
|
202
|
+
retry_stats = Hash.new(0)
|
203
|
+
|
204
|
+
Sidekiq::RetrySet.new.map do |job|
|
205
|
+
retry_stats[job.klass] += 1
|
206
|
+
end
|
207
|
+
|
208
|
+
retry_stats.each do |class_name, count|
|
209
|
+
@metrics.add("sidekiq_to_be_retried_jobs", count, **labels.merge({ name: class_name }))
|
210
|
+
end
|
211
|
+
end
|
212
|
+
|
213
|
+
def with_sidekiq(namespaced = true)
|
181
214
|
# TODO: this is not concurrent safe as we change global context
|
182
215
|
# It means that we are unable to use many different sidekiq's
|
183
216
|
# which is not a problem as of now
|
184
217
|
Sidekiq.configure_client do |config|
|
185
|
-
config.redis = self.class.connection_pool[redis_options]
|
218
|
+
config.redis = self.class.connection_pool[redis_options(namespaced)]
|
186
219
|
end
|
187
220
|
|
188
221
|
return unless connected?
|
@@ -190,14 +223,14 @@ module GitLab
|
|
190
223
|
yield
|
191
224
|
end
|
192
225
|
|
193
|
-
def redis_options
|
226
|
+
def redis_options(namespaced = true)
|
194
227
|
options = {
|
195
228
|
url: @opts[:redis_url],
|
196
|
-
namespace: "resque:gitlab",
|
197
229
|
connect_timeout: 1,
|
198
230
|
reconnect_attempts: 0
|
199
231
|
}
|
200
232
|
|
233
|
+
options[:namespace] = "resque:gitlab" if namespaced
|
201
234
|
options[:id] = nil unless redis_enable_client?
|
202
235
|
options
|
203
236
|
end
|
@@ -218,6 +251,11 @@ module GitLab
|
|
218
251
|
@logger&.error "Error connecting to the Redis: #{e}"
|
219
252
|
@connected = false
|
220
253
|
end
|
254
|
+
|
255
|
+
def add_namespace_labels?
|
256
|
+
@probe_namespaced && @probe_non_namespaced
|
257
|
+
end
|
221
258
|
end
|
259
|
+
# rubocop:enable Metrics/ClassLength
|
222
260
|
end
|
223
261
|
end
|
@@ -0,0 +1,64 @@
|
|
1
|
+
require "spec_helper"
|
2
|
+
require "gitlab_exporter/sidekiq"
|
3
|
+
|
4
|
+
describe GitLab::Exporter::SidekiqProber do
|
5
|
+
let(:probe_list) { %w[probe_stats probe_queues probe_future_sets probe_jobs_limit probe_workers probe_retries] }
|
6
|
+
let(:opt) { {} }
|
7
|
+
let(:probe) { described_class.new(**opt) }
|
8
|
+
|
9
|
+
before do
|
10
|
+
# stub connection as we want check if the probe methods are called
|
11
|
+
# with the right namespace argument
|
12
|
+
allow(probe).to receive(:connected?).and_return(true)
|
13
|
+
end
|
14
|
+
|
15
|
+
context "when no namespace options are defined" do
|
16
|
+
it "defaults to probing non-namespaced keys only" do
|
17
|
+
probe_list.each do |probe_type|
|
18
|
+
expect(probe).not_to receive("_#{probe_type}".to_sym).with(false)
|
19
|
+
expect(probe).to receive("_#{probe_type}".to_sym).with(no_args)
|
20
|
+
|
21
|
+
probe.send(probe_type)
|
22
|
+
end
|
23
|
+
end
|
24
|
+
end
|
25
|
+
|
26
|
+
context "when probing both namespaces" do
|
27
|
+
let(:opt) { { probe_non_namespaced: true, probe_namespaced: true } }
|
28
|
+
|
29
|
+
it "probes both namespaces" do
|
30
|
+
probe_list.each do |probe_type|
|
31
|
+
expect(probe).to receive("_#{probe_type}".to_sym).with(false)
|
32
|
+
expect(probe).to receive("_#{probe_type}".to_sym).with(no_args)
|
33
|
+
|
34
|
+
probe.send(probe_type)
|
35
|
+
end
|
36
|
+
end
|
37
|
+
end
|
38
|
+
|
39
|
+
context "when probing non-namespaced only" do
|
40
|
+
let(:opt) { { probe_non_namespaced: true } }
|
41
|
+
|
42
|
+
it "probes non-namespaced only" do
|
43
|
+
probe_list.each do |probe_type|
|
44
|
+
expect(probe).to receive("_#{probe_type}".to_sym).with(false)
|
45
|
+
expect(probe).not_to receive("_#{probe_type}".to_sym).with(no_args)
|
46
|
+
|
47
|
+
probe.send(probe_type)
|
48
|
+
end
|
49
|
+
end
|
50
|
+
end
|
51
|
+
|
52
|
+
context "when probing namespace only" do
|
53
|
+
let(:opt) { { probe_namespaced: true } }
|
54
|
+
|
55
|
+
it "probes namespaced only" do
|
56
|
+
probe_list.each do |probe_type|
|
57
|
+
expect(probe).not_to receive("_#{probe_type}".to_sym).with(false)
|
58
|
+
expect(probe).to receive("_#{probe_type}".to_sym).with(no_args)
|
59
|
+
|
60
|
+
probe.send(probe_type)
|
61
|
+
end
|
62
|
+
end
|
63
|
+
end
|
64
|
+
end
|
metadata
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
--- !ruby/object:Gem::Specification
|
2
2
|
name: gitlab-exporter
|
3
3
|
version: !ruby/object:Gem::Version
|
4
|
-
version: 13.
|
4
|
+
version: 13.2.0
|
5
5
|
platform: ruby
|
6
6
|
authors:
|
7
7
|
- Pablo Carranza
|
@@ -235,6 +235,7 @@ files:
|
|
235
235
|
- spec/memstats_spec.rb
|
236
236
|
- spec/prometheus_metrics_spec.rb
|
237
237
|
- spec/ruby_spec.rb
|
238
|
+
- spec/sidekiq_spec.rb
|
238
239
|
- spec/spec_helper.rb
|
239
240
|
- spec/util_spec.rb
|
240
241
|
homepage: http://gitlab.com
|
@@ -273,5 +274,6 @@ test_files:
|
|
273
274
|
- spec/memstats_spec.rb
|
274
275
|
- spec/prometheus_metrics_spec.rb
|
275
276
|
- spec/ruby_spec.rb
|
277
|
+
- spec/sidekiq_spec.rb
|
276
278
|
- spec/spec_helper.rb
|
277
279
|
- spec/util_spec.rb
|