Module: Legion::LLM::Batch

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm/batch.rb

Class Method Summary collapse

Class Method Details

.enabled? ⇒ Boolean

Returns:

  • (Boolean) 


15
16
17
 
# File 'lib/legion/llm/batch.rb', line 15

def enabled?
  settings.fetch(:enabled, false) == true
end

.enqueue(messages: nil, model: nil, message: nil, callback: nil, provider: nil, priority: :normal, **opts) ⇒ Object 



19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
 
# File 'lib/legion/llm/batch.rb', line 19

def enqueue(messages: nil, model: nil, message: nil, callback: nil, provider: nil, priority: :normal, **opts)
  request_id = SecureRandom.uuid
  msgs = messages || (message ? [{ role: 'user', content: message }] : [])

  entry = {
    id:        request_id,
    messages:  msgs,
    model:     model,
    provider:  provider,
    callback:  callback,
    priority:  priority,
    opts:      opts,
    queued_at: Time.now.utc
  }

  @mutex.synchronize { queue << entry }
  ensure_flush_timer
  log.debug "Legion::LLM::Batch enqueued #{request_id} (queue size: #{queue_size})"
  request_id
end

.flush(max_size: nil, max_wait: nil) ⇒ Object 



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
 
# File 'lib/legion/llm/batch.rb', line 40

def flush(max_size: nil, max_wait: nil)
  effective_max  = max_size || settings.fetch(:max_batch_size, 100)
  effective_wait = max_wait || settings.fetch(:window_seconds, 300)
  cutoff = Time.now.utc - effective_wait

  to_flush = @mutex.synchronize do
    ready = queue.select { |e| e[:queued_at] <= cutoff }
                 .sort_by { |e| priority_rank(e[:priority]) }
                 .first(effective_max)
    ready.each { |e| queue.delete(e) }
    ready
  end

  return [] if to_flush.empty?

  log.debug "Legion::LLM::Batch flushing #{to_flush.size} request(s)"

  groups = to_flush.group_by { |e| [e[:provider], e[:model]] }
  results = []

  groups.each do |(provider, model), entries|
    entries.each do |entry|
      result = submit_single(entry, provider: provider, model: model)
      entry[:callback]&.call(result)
      results << { id: entry[:id], status: result[:status], result: result }
    end
  end

  results
end

.queue_size ⇒ Object 



71
72
73
 
# File 'lib/legion/llm/batch.rb', line 71

def queue_size
  @mutex.synchronize { queue.size }
end

.reset! ⇒ Object 



88
89
90
91
 
# File 'lib/legion/llm/batch.rb', line 88

def reset!
  @mutex.synchronize { @queue = [] }
  stop_flush_timer
end

.status ⇒ Object 



75
76
77
78
79
80
81
82
83
84
85
86
 
# File 'lib/legion/llm/batch.rb', line 75

def status
  entries = @mutex.synchronize { queue.dup }
  oldest = entries.min_by { |e| e[:queued_at] }
  {
    enabled:        enabled?,
    queue_size:     entries.size,
    max_batch_size: settings.fetch(:max_batch_size, 100),
    window_seconds: settings.fetch(:window_seconds, 300),
    oldest_queued:  oldest ? oldest[:queued_at].iso8601 : nil,
    by_priority:    entries.group_by { |e| e[:priority] }.transform_values(&:size)
  }
end

.stop_flush_timer ⇒ Object 



93
94
95
96
 
# File 'lib/legion/llm/batch.rb', line 93

def stop_flush_timer
  @flush_timer&.shutdown if @flush_timer.respond_to?(:shutdown)
  @flush_timer = nil
end