Class: RocketJob::Worker

Inherits:
Object
  • Object
show all
Includes:
SemanticLogger::Loggable
Defined in:
lib/rocket_job/worker.rb

Overview

Worker

A worker runs on a single operating system thread Is usually started under a Rocket Job server process.

Direct Known Subclasses

ThreadWorker

Defined Under Namespace

Classes: Shutdown

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(id: 0, server_name: "inline:0") ⇒ Worker

Returns a new instance of Worker.



19
20
21
22
23
24
25
 
# File 'lib/rocket_job/worker.rb', line 19

def initialize(id: 0, server_name: "inline:0")
  @id             = id
  @server_name    = server_name
  @name           = "#{server_name}:#{id}"
  @re_check_start = Time.now
  @current_filter = Config.filter || {}
end

Instance Attribute Details

#current_filter ⇒ Object

Returns the value of attribute current_filter.



9
10
11
 
# File 'lib/rocket_job/worker.rb', line 9

def current_filter
  @current_filter
end

#id ⇒ Object

Returns the value of attribute id.



9
10
11
 
# File 'lib/rocket_job/worker.rb', line 9

def id
  @id
end

#name ⇒ Object (readonly)

Returns the value of attribute name.



10
11
12
 
# File 'lib/rocket_job/worker.rb', line 10

def name
  @name
end

#server_name ⇒ Object (readonly)

Returns the value of attribute server_name.



10
11
12
 
# File 'lib/rocket_job/worker.rb', line 10

def server_name
  @server_name
end

Instance Method Details

#add_to_current_filter(filter) ⇒ Object

Add the supplied filter to the current filter.



228
229
230
231
232
233
234
235
236
237
238
 
# File 'lib/rocket_job/worker.rb', line 228

def add_to_current_filter(filter)
  filter.each_pair do |k, v|
    current_filter[k] =
      if (previous = current_filter[k])
        v.is_a?(Array) ? previous + v : v
      else
        v
      end
  end
  current_filter
end

#alive? ⇒ Boolean

Returns:

  • (Boolean) 


27
28
29
 
# File 'lib/rocket_job/worker.rb', line 27

def alive?
  true
end

#backtrace ⇒ Object 



31
32
33
 
# File 'lib/rocket_job/worker.rb', line 31

def backtrace
  Thread.current.backtrace
end

#find_and_assign_job ⇒ Object

Finds the next job to work on in priority based order and assigns it to this worker.

Applies the current filter to exclude filtered jobs.

Returns nil if no jobs are available for processing.

Notes:

  • An already running batch job is joined with a read-only query, since concurrency for a batch job is coordinated per-slice (see Sliced::Input#next_slice), not on the job document. Writing worker_name/state to the shared job document on every poll turns it into a write-contention hotspot: with many workers MongoDB serializes the updates and retries the write conflicts (server log id 46404), which is what makes a large batch job slow down as workers are added.
  • Only a queued job is claimed with a write, transitioning it to running. That write is naturally distributed: each queued job is claimed once.


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
 
# File 'lib/rocket_job/worker.rb', line 195

def find_and_assign_job
  SemanticLogger.silence(:info) do
    scheduled = RocketJob::Job.where(run_at: nil).or(:run_at.lte => Time.now)
    working   = RocketJob::Job.queued.or(state: "running", sub_state: "processing")
    query     = RocketJob::Job.and(working, scheduled)
    query     = query.and(current_filter) unless current_filter.blank?
    query     = query.sort(priority: 1, _id: 1)

    # Retry only when a queued job was claimed by another worker first.
    loop do
      job = query.first
      return nil unless job

      # Already running batch job: join it without writing to the job document.
      return job if job.running?

      # Queued job: claim it atomically, guarding on it still being queued.
      # find_one_and_update returns the pre-image (state: queued) so that the
      # caller still runs throttles and `start!`.
      # Clear any stale throttle reason from a previous throttled poll so a job
      # that is now allowed to run no longer reports as throttled in Mission Control.
      update = {
        "$set"   => {"worker_name" => name, "state" => "running"},
        "$unset" => {"throttled_by" => "", "throttled_at" => ""}
      }
      claimed = RocketJob::Job.queued.where(id: job.id).
                find_one_and_update(update, bypass_document_validation: true)
      return claimed if claimed
    end
  end
end

#join(*_args) ⇒ Object 



35
36
37
 
# File 'lib/rocket_job/worker.rb', line 35

def join(*_args)
  true
end

#kill ⇒ Object 



39
40
41
 
# File 'lib/rocket_job/worker.rb', line 39

def kill
  true
end

#next_available_job ⇒ Object

Returns [RocketJob::Job] the next job available for processing. Returns [nil] if no job is available for processing.

Notes:

  • Destroys expired jobs
  • Runs job throttles and skips the job if it is throttled.
    • Adding that filter to the current filter to exclude from subsequent polling.


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
 
# File 'lib/rocket_job/worker.rb', line 130

def next_available_job
  until shutdown?
    job = find_and_assign_job
    return unless job

    if job.expired?
      job.fail_on_exception! do
        job.worker_name = name
        job.destroy
        logger.info("Destroyed expired job.")
      end
      next
    end

    # Batch Job that is already started?
    # Batch has its own throttles for slices.
    return job if job.running?

    # Should this job be throttled?
    next if job.fail_on_exception! { throttled_job?(job) }
    # Job failed during throttle execution?
    next if job.failed?

    # Start this job!
    job.fail_on_exception! { job.start!(name) }
    return job if job.running?
  end
end

#random_wait_interval ⇒ Object

Returns [Float] a randomized poll interval in seconds up to the maximum configured poll interval.



241
242
243
 
# File 'lib/rocket_job/worker.rb', line 241

def random_wait_interval
  rand(Config.max_poll_seconds * 1000) / 1000
end

#release_active_record_connections ⇒ Object

Return the database connections for this thread back to the connection pool.



102
103
104
105
106
107
108
109
110
111
 
# File 'lib/rocket_job/worker.rb', line 102

def release_active_record_connections
  return unless defined?(ActiveRecord::Base)

  if ActiveRecord::Base.respond_to?(:connection_handler)
    # Rails 7.2+
    ActiveRecord::Base.connection_handler.clear_active_connections!(:all)
  else
    ActiveRecord::Base.clear_active_connections!
  end
end

#reset_filter_if_expired ⇒ Object

Resets the current job filter if the relevant time interval has passed



114
115
116
117
118
119
120
121
 
# File 'lib/rocket_job/worker.rb', line 114

def reset_filter_if_expired
  # Only clear out the current_filter after every `re_check_seconds`
  time = Time.now
  return unless (time - @re_check_start) > Config.re_check_seconds

  @re_check_start     = time
  self.current_filter = Config.filter || {}
end

#run ⇒ Object

Process jobs until it shuts down

Params worker_id [Integer] The number of this worker for logging purposes



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
 
# File 'lib/rocket_job/worker.rb', line 61

def run
  Thread.current.name = format("rocketjob %03i", id)
  logger.info "Started"

  until shutdown?
    reset_filter_if_expired
    job = next_available_job

    if job
      # Returns true when this job has no more work immediately available,
      # for example a batch job that has run out of slices, or was throttled.
      no_immediate_work = job.rocket_job_work(self, false)

      # Return the database connections for this thread back to the connection pool.
      release_active_record_connections

      sleep_seconds =
        if no_immediate_work
          # Stagger workers so that they don't all poll at the same time.
          random_wait_interval
        else
          # Work was performed and more queued work is likely available,
          # so poll again immediately rather than waiting a full interval.
          0
        end
    else
      # No work was found, so wait the full poll interval before checking again.
      sleep_seconds = Config.max_poll_seconds
    end

    wait_for_shutdown?(sleep_seconds)
  end

  logger.info "Stopping"
rescue Exception => e
  logger.fatal("Unhandled exception in job processing thread", e)
ensure
  release_active_record_connections
end

#shutdown! ⇒ Object 



47
48
49
 
# File 'lib/rocket_job/worker.rb', line 47

def shutdown!
  true
end

#shutdown? ⇒ Boolean

Returns:

  • (Boolean) 


43
44
45
 
# File 'lib/rocket_job/worker.rb', line 43

def shutdown?
  false
end

#throttled_job?(job) ⇒ Boolean

Whether the supplied job has been throttled and should be ignored.

Returns:

  • (Boolean) 


160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
 
# File 'lib/rocket_job/worker.rb', line 160

def throttled_job?(job)
  # Evaluate job throttles, if any.
  throttle = job.rocket_job_throttles.matching_throttle(job)
  return false unless throttle

  add_to_current_filter(throttle.extract_filter(job))
  # Restore retrieved job so that other workers can process it later, recording
  # why it was throttled so it is visible in Mission Control. This reuses the
  # write that requeues the job, so it adds no extra database round trip.
  job.set(
    worker_name:  nil,
    state:        :queued,
    throttled_by: throttle.extract_description(job),
    throttled_at: Time.now
  )
  true
end

#wait_for_shutdown?(_timeout = nil) ⇒ Boolean

Returns [true|false] whether the shutdown indicator was set

Returns:

  • (Boolean) 


52
53
54
 
# File 'lib/rocket_job/worker.rb', line 52

def wait_for_shutdown?(_timeout = nil)
  false
end