1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
|
# frozen_string_literal: true
module Ci
# This class responsible for assigning
# proper pending build to runner on runner API request
class RegisterJobService
attr_reader :runner, :metrics
TEMPORARY_LOCK_TIMEOUT = 3.seconds
Result = Struct.new(:build, :build_json, :valid?)
##
# The queue depth limit number has been determined by observing 95
# percentile of effective queue depth on gitlab.com. This is only likely to
# affect 5% of the worst case scenarios.
MAX_QUEUE_DEPTH = 45
def initialize(runner)
@runner = runner
@metrics = ::Gitlab::Ci::Queue::Metrics.new(runner)
end
def execute(params = {})
db_all_caught_up =
::Ci::Runner.sticking.all_caught_up?(:runner, runner.id)
@metrics.increment_queue_operation(:queue_attempt)
result = @metrics.observe_queue_time(:process, @runner.runner_type) do
process_queue(params)
end
# Since we execute this query against replica it might lead to false-positive
# We might receive the positive response: "hi, we don't have any more builds for you".
# This might not be true. If our DB replica is not up-to date with when runner event was generated
# we might still have some CI builds to be picked. Instead we should say to runner:
# "Hi, we don't have any more builds now, but not everything is right anyway, so try again".
# Runner will retry, but again, against replica, and again will check if replication lag did catch-up.
if !db_all_caught_up && !result.build
metrics.increment_queue_operation(:queue_replication_lag)
::Ci::RegisterJobService::Result.new(nil, false) # rubocop:disable Cop/AvoidReturnFromBlocks
else
result
end
end
private
def process_queue(params)
valid = true
depth = 0
each_build(params) do |build|
depth += 1
@metrics.increment_queue_operation(:queue_iteration)
if depth > max_queue_depth
@metrics.increment_queue_operation(:queue_depth_limit)
valid = false
break
end
# We read builds from replicas
# It is likely that some other concurrent connection is processing
# a given build at a given moment. To avoid an expensive compute
# we perform an exclusive lease on Redis to acquire a build temporarily
unless acquire_temporary_lock(build.id)
@metrics.increment_queue_operation(:build_temporary_locked)
# We failed to acquire lock
# - our queue is not complete as some resources are locked temporarily
# - we need to re-process it again to ensure that all builds are handled
valid = false
next
end
result = process_build(build, params)
next unless result
if result.valid?
@metrics.register_success(result.build)
@metrics.observe_queue_depth(:found, depth)
return result # rubocop:disable Cop/AvoidReturnFromBlocks
else
# The usage of valid: is described in
# handling of ActiveRecord::StaleObjectError
valid = false
end
end
@metrics.increment_queue_operation(:queue_conflict) unless valid
@metrics.observe_queue_depth(:conflict, depth) unless valid
@metrics.observe_queue_depth(:not_found, depth) if valid
@metrics.register_failure
Result.new(nil, nil, valid)
end
# rubocop: disable CodeReuse/ActiveRecord
def each_build(params, &blk)
queue = ::Ci::Queue::BuildQueueService.new(runner)
builds = begin
if runner.instance_type?
queue.builds_for_shared_runner
elsif runner.group_type?
queue.builds_for_group_runner
else
queue.builds_for_project_runner
end
end
if runner.ref_protected?
builds = queue.builds_for_protected_runner(builds)
end
# pick builds that does not have other tags than runner's one
builds = queue.builds_matching_tag_ids(builds, runner.tags.ids)
# pick builds that have at least one tag
unless runner.run_untagged?
builds = queue.builds_with_any_tags(builds)
end
# pick builds that older than specified age
if params.key?(:job_age)
builds = queue.builds_queued_before(builds, params[:job_age].seconds.ago)
end
build_ids = retrieve_queue(-> { queue.execute(builds) })
@metrics.observe_queue_size(-> { build_ids.size }, @runner.runner_type)
build_ids.each { |build_id| yield Ci::Build.find(build_id) }
end
# rubocop: enable CodeReuse/ActiveRecord
def retrieve_queue(queue_query_proc)
##
# We want to reset a load balancing session to discard the side
# effects of writes that could have happened prior to this moment.
#
::Gitlab::Database::LoadBalancing::Session.clear_session
@metrics.observe_queue_time(:retrieve, @runner.runner_type) do
queue_query_proc.call
end
end
def process_build(build, params)
unless build.pending?
@metrics.increment_queue_operation(:build_not_pending)
##
# If this build can not be picked because we had stale data in
# `ci_pending_builds` table, we need to respond with 409 to retry
# this operation.
#
if ::Ci::UpdateBuildQueueService.new.remove!(build)
return Result.new(nil, nil, false)
end
return
end
if runner.matches_build?(build)
@metrics.increment_queue_operation(:build_can_pick)
else
@metrics.increment_queue_operation(:build_not_pick)
return
end
# In case when 2 runners try to assign the same build, second runner will be declined
# with StateMachines::InvalidTransition or StaleObjectError when doing run! or save method.
if assign_runner!(build, params)
present_build!(build)
end
rescue ActiveRecord::StaleObjectError
# We are looping to find another build that is not conflicting
# It also indicates that this build can be picked and passed to runner.
# If we don't do it, basically a bunch of runners would be competing for a build
# and thus we will generate a lot of 409. This will increase
# the number of generated requests, also will reduce significantly
# how many builds can be picked by runner in a unit of time.
# In case we hit the concurrency-access lock,
# we still have to return 409 in the end,
# to make sure that this is properly handled by runner.
@metrics.increment_queue_operation(:build_conflict_lock)
Result.new(nil, nil, false)
rescue StateMachines::InvalidTransition
@metrics.increment_queue_operation(:build_conflict_transition)
Result.new(nil, nil, false)
rescue StandardError => ex
@metrics.increment_queue_operation(:build_conflict_exception)
# If an error (e.g. GRPC::DeadlineExceeded) occurred constructing
# the result, consider this as a failure to be retried.
scheduler_failure!(build)
track_exception_for_build(ex, build)
# skip, and move to next one
nil
end
def max_queue_depth
MAX_QUEUE_DEPTH
end
# Force variables evaluation to occur now
def present_build!(build)
# We need to use the presenter here because Gitaly calls in the presenter
# may fail, and we need to ensure the response has been generated.
presented_build = ::Ci::BuildRunnerPresenter.new(build) # rubocop:disable CodeReuse/Presenter
build_json = ::API::Entities::Ci::JobRequest::Response.new(presented_build).to_json
Result.new(build, build_json, true)
end
def assign_runner!(build, params)
build.runner_id = runner.id
build.runner_session_attributes = params[:session] if params[:session].present?
failure_reason, _ = pre_assign_runner_checks.find { |_, check| check.call(build, params) }
if failure_reason
@metrics.increment_queue_operation(:runner_pre_assign_checks_failed)
build.drop!(failure_reason)
else
@metrics.increment_queue_operation(:runner_pre_assign_checks_success)
build.run!
end
!failure_reason
end
def acquire_temporary_lock(build_id)
return true unless Feature.enabled?(:ci_register_job_temporary_lock, runner)
key = "build/register/#{build_id}"
Gitlab::ExclusiveLease
.new(key, timeout: TEMPORARY_LOCK_TIMEOUT.to_i)
.try_obtain
end
def scheduler_failure!(build)
Gitlab::OptimisticLocking.retry_lock(build, 3, name: 'register_job_scheduler_failure') do |subject|
subject.drop!(:scheduler_failure)
end
rescue StandardError => ex
build.doom!
# This requires extra exception, otherwise we would loose information
# why we cannot perform `scheduler_failure`
track_exception_for_build(ex, build)
end
def track_exception_for_build(ex, build)
Gitlab::ErrorTracking.track_exception(ex,
build_id: build.id,
build_name: build.name,
build_stage: build.stage,
pipeline_id: build.pipeline_id,
project_id: build.project_id
)
end
def pre_assign_runner_checks
{
missing_dependency_failure: -> (build, _) { !build.has_valid_build_dependencies? },
runner_unsupported: -> (build, params) { !build.supported_runner?(params.dig(:info, :features)) },
archived_failure: -> (build, _) { build.archived? },
project_deleted: -> (build, _) { build.project.pending_delete? },
builds_disabled: -> (build, _) { !build.project.builds_enabled? },
user_blocked: -> (build, _) { build.user&.blocked? }
}
end
end
end
Ci::RegisterJobService.prepend_mod_with('Ci::RegisterJobService')
|