Class: Chantier::ThreadPool

Inherits:
Object
  • Object
show all
Defined in:
lib/thread_pool.rb

Overview

Allows you to spin off a pool of Threads that is not larger than X. You can then enqueue tasks to be executed within that pool. When all slots are full the caller will be blocked until a slot becomes available.

manager = ThreadPool.new(slots = 4)
jobs_hose.each_job do | job |
  # this call will block until a slot becomes available
  manager.fork_task do # this block runs in a subprocess
    Churner.new(job).churn
  end
  manager.still_running? # => most likely "true"
end

manager.block_until_complete! #=> Will block until all the subprocesses have terminated

If you have a finite Enumerable at hand you can also launch it into the ThreadPool, like so:

manager = ThreadPool.new(slots = 4)

manager.map_fork(job_tickets) do | job_ticket |
  # this block will run in a forked subprocess
  Churner.new(job).churn
  ...
end

Can be rewritten using Threads if operation on JVM/Rubinius will be feasible.

Constant Summary collapse

SCHEDULER_SLEEP_SECONDS =

The manager uses loops in a few places. By doing a little sleep() in those loops we can yield process control back to the OS which brings the CPU usage of the managing process to small numbers. If you just do a loop {} MRI will saturate a whole core and not let go off of it until the loop returns.

(1.0 / 500)

Instance Method Summary collapse

Constructor Details

#initialize(num_threads, max_failures: nil) ⇒ ThreadPool

Initializes a new ProcessPool with the given number of workers. If max_failures is given the fork_task method will raise an exception if more than N threads spawned have raised during execution.



40
41
42
43
44
45
46
47
48
49
50
51
52
 
# File 'lib/thread_pool.rb', line 40

def initialize(num_threads, max_failures: nil)
  raise "Need at least 1 slot, given #{num_threads.to_i}" unless num_threads.to_i > 0
  @threads = [nil] * num_threads.to_i
  @semaphore = Mutex.new
  
  # Failure counters
  @failure_count = 0
  @max_failures = max_failures && max_failures.to_i
  
  # Information on the last exception that happened
  @aborted = false
  @last_representative_exception = nil
end

Instance Method Details

#block_until_complete! ⇒ Object

Analogous to Process.wait or wait_all - will block until all of the process slots have been freed.



109
110
111
112
113
114
115
 
# File 'lib/thread_pool.rb', line 109

def block_until_complete!
  @threads.map do |e| 
    if e.respond_to?(:join) && e.alive?
      e.join
    end
  end
end

#fork_task(&blk) ⇒ Object

Run the given block in a thread. This method will block the thread it is called from until a slot in the thread table becomes free.



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
 
# File 'lib/thread_pool.rb', line 74

def fork_task(&blk)
  if @last_representative_exception
    raise "Reached error limit of #{@max_failures} (last error was #{@last_representative_exception.inspect})"
  end
  
  destination_slot_idx = nil
  
  # Try to find a slot in the process table where this job can go
  catch :_found do
    loop do
      @semaphore.synchronize do
        if destination_slot_idx = @threads.index(nil)
          @threads[destination_slot_idx] = true # occupy it
          throw :_found
        end
      end
      sleep SCHEDULER_SLEEP_SECONDS # Breathing room
    end
  end
  
  # No need to lock this because we already reserved that slot
  @threads[destination_slot_idx] = Thread.new do
    # Run the given block
    run_block_with_exception_protection(&blk)
    # ...and remove that process from the process table
    @semaphore.synchronize { @threads[destination_slot_idx] = nil }
  end
end

#map_fork(arguments_per_job, &blk) ⇒ Object

Distributes the elements in the given Enumerable to parallel workers, N workers at a time. The method will return once all the workers for all the elements of the Enumerable have terminated.

pool = ThreadPool.new(5)
pool.map_fork(array_of_urls) do | single_url |
  Faraday.get(single_url).response ...
  ...
  ...
end



64
65
66
67
68
69
 
# File 'lib/thread_pool.rb', line 64

def map_fork(arguments_per_job, &blk)
  arguments_per_job.each do | single_block_argument |
    fork_task { yield(single_block_argument) }
  end
  block_until_complete!
end

#still_running? ⇒ Boolean

Tells whether some processes are still churning

Returns:

  • (Boolean) 


104
105
106
 
# File 'lib/thread_pool.rb', line 104

def still_running?
  @threads.any?{|e| e && e.respond_to?(:alive?) && e.alive? }
end