Module: Legion::LLM::API::Namespaces::OpenAI::Batches

Extended by:
Legion::Logging::Helper
Defined in:
lib/legion/llm/api/namespaces/openai/batches.rb

Constant Summary collapse

BATCH_STORE =

rubocop:disable Style/MutableConstant

{}
BATCH_MUTEX = 
Mutex.new
BATCH_POOL_MUTEX = 
Mutex.new

Class Method Summary collapse

Class Method Details

.batch_pool ⇒ Object 



21
22
23
24
25
26
27
28
29
30
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 21

def self.batch_pool
  return @batch_pool if @batch_pool

  BATCH_POOL_MUTEX.synchronize do
    @batch_pool ||= begin
      pool_size = Legion::LLM::Settings.value(:api, :batch_pool_size, default: 4)
      Concurrent::FixedThreadPool.new(pool_size, fallback_policy: :abort)
    end
  end
end

.dispatch_batch_request(batch_id, req, caller_context: {}) ⇒ Object 



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
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 245

def self.dispatch_batch_request(batch_id, req, caller_context: {})
  custom_id = req[:custom_id] || req['custom_id'] || SecureRandom.uuid
  body      = req[:body] || req['body'] || {}
  messages  = body[:messages] || body['messages'] || []
  model     = body[:model]    || body['model']    || 'default'

  effective_caller = caller_context.empty? ? { source: 'openai_batch', path: "/v1/batches/#{batch_id}" } : caller_context

  inference_request = Legion::LLM::Inference::Request.build(
    messages: messages,
    routing:  { model: model },
    caller:   effective_caller
  )

  executor = Legion::LLM::Inference::Executor.new(inference_request)
  response = executor.call

  log.debug("[llm][api][openai][batches] batch_id=#{batch_id} custom_id=#{custom_id} status=success")
  {
    custom_id: custom_id,
    response:  {
      status_code: 200,
      body:        { choices: [{ message: response&.message }] }
    }
  }
rescue StandardError => e
  handle_exception(e, level: :warn, handled: true, operation: "llm.api.openai.batches.request.#{custom_id}")
  {
    custom_id: custom_id,
    error:     { code: 'server_error', message: e.message }
  }
end

.files_dir ⇒ Object 



278
279
280
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 278

def self.files_dir
  ::File.join(::Dir.home, '.legionio', 'data', 'files')
end

.load_batch_requests(file_id) ⇒ Object 



234
235
236
237
238
239
240
241
242
243
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 234

def self.load_batch_requests(file_id)
  file_path = ::File.join(Batches.files_dir, file_id)
  return [] unless ::File.exist?(file_path)

  ::File.readlines(file_path).filter_map do |line|
    Legion::JSON.load(line.strip)
  rescue StandardError
    nil
  end
end

.process_batch(batch_id) ⇒ Object

Class-level batch processor (called from thread pool)



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
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 179

def self.process_batch(batch_id)
  entry = BATCH_MUTEX.synchronize { BATCH_STORE[batch_id]&.dup }
  return unless entry

  extend Legion::Logging::Helper

  BATCH_MUTEX.synchronize do
    BATCH_STORE[batch_id][:status]         = :in_progress
    BATCH_STORE[batch_id][:in_progress_at] = Time.now
  end

  file_id  = entry[:input_file_id]
  requests = load_batch_requests(file_id)

  BATCH_MUTEX.synchronize do
    BATCH_STORE[batch_id][:request_counts][:total] = requests.size
  end

  caller_ctx = BATCH_MUTEX.synchronize { BATCH_STORE[batch_id]&.dig(:caller_context) } || {}

  results = requests.map do |req|
    cancelled = BATCH_MUTEX.synchronize { BATCH_STORE[batch_id]&.dig(:status) == :cancelling }
    break [] if cancelled

    dispatch_batch_request(batch_id, req, caller_context: caller_ctx)
  end.flatten.compact

  cancelled = BATCH_MUTEX.synchronize { BATCH_STORE[batch_id]&.dig(:status) == :cancelling }

  BATCH_MUTEX.synchronize do
    next unless BATCH_STORE[batch_id]

    if cancelled
      BATCH_STORE[batch_id][:status]       = :cancelled
      BATCH_STORE[batch_id][:cancelled_at] = Time.now
    else
      completed_count = results.count { |r| r[:response] }
      failed_count    = results.count { |r| r[:error] }
      BATCH_STORE[batch_id][:status]       = :completed
      BATCH_STORE[batch_id][:completed_at] = Time.now
      BATCH_STORE[batch_id][:results]      = results
      BATCH_STORE[batch_id][:request_counts][:completed] = completed_count
      BATCH_STORE[batch_id][:request_counts][:failed]    = failed_count
    end
  end
rescue StandardError => e
  handle_exception(e, level: :error, handled: false, operation: "llm.api.openai.batches.process.#{batch_id}")
  BATCH_MUTEX.synchronize do
    next unless BATCH_STORE[batch_id]

    BATCH_STORE[batch_id][:status]    = :failed
    BATCH_STORE[batch_id][:failed_at] = Time.now
  end
end

.registered(app) ⇒ Object

rubocop:disable Metrics/AbcSize,Metrics/MethodLength



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
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 32

def self.registered(app) # rubocop:disable Metrics/AbcSize,Metrics/MethodLength
  log.debug('[llm][api][namespaces][openai][batches] registering routes')

  # POST /v1/batches
  app.post '/v1/batches' do # rubocop:disable Metrics/BlockLength
    require_llm!
    body = parse_request_body
    validate_required!(body, :input_file_id, :endpoint)

    batch_id = "batch_#{SecureRandom.hex(16)}"
    endpoint = body[:endpoint].to_s
    file_id  = body[:input_file_id].to_s
    window   = body[:completion_window] || '24h'
    metadata = body[:metadata] || {}

    caller_context = build_server_caller(source: 'openai_batch', path: request.path, env: env)

    entry = {
      id:                batch_id,
      object:            'batch',
      endpoint:          endpoint,
      errors:            nil,
      input_file_id:     file_id,
      completion_window: window,
      status:            :validating,
      output_file_id:    nil,
      error_file_id:     nil,
      created_at:        Time.now,
      in_progress_at:    nil,
      expires_at:        Time.now + 86_400,
      finalizing_at:     nil,
      completed_at:      nil,
      failed_at:         nil,
      expired_at:        nil,
      cancelling_at:     nil,
      cancelled_at:      nil,
      request_counts:    { total: 0, completed: 0, failed: 0 },
      metadata:          metadata,
      results:           [],
      caller_context:    caller_context
    }

    BATCH_MUTEX.synchronize { BATCH_STORE[batch_id] = entry }
    log.debug("[llm][api][openai][batches] action=create batch_id=#{batch_id} endpoint=#{endpoint} file_id=#{file_id}")

    begin
      Batches.batch_pool.post { Batches.process_batch(batch_id) }
    rescue Concurrent::RejectedExecutionError
      BATCH_MUTEX.synchronize { BATCH_STORE.delete(batch_id) }
      log.warn("[llm][api][openai][batches] action=queue_full batch_id=#{batch_id}")
      halt 503, { 'Content-Type' => 'application/json' },
           Legion::JSON.dump({ error: { message: 'Batch queue is full. Retry after existing batches complete.',
                                        type: 'server_error', code: 'batch_queue_full' } })
    end

    content_type :json
    status 200
    Legion::JSON.dump(Batches.serialize_batch(entry))
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.batches.create')
    openai_error(e.message, type: 'server_error', code: 'internal_error', status_code: 500)
  end

  # GET /v1/batches (list)
  app.get '/v1/batches' do
    require_llm!
    limit = params[:limit]&.to_i || 20
    after = params[:after]

    entries = BATCH_MUTEX.synchronize { BATCH_STORE.values.dup }
    entries.sort_by! { |b| -b[:created_at].to_i }

    if after
      idx     = entries.index { |b| b[:id] == after }
      entries = entries[(idx + 1)..] if idx
    end

    entries = entries.first(limit)
    log.debug("[llm][api][openai][batches] action=list count=#{entries.size}")

    content_type :json
    status 200
    Legion::JSON.dump({
                        object:   'list',
                        data:     entries.map { |b| Batches.serialize_batch(b) },
                        has_more: false,
                        first_id: entries.first&.dig(:id),
                        last_id:  entries.last&.dig(:id)
                      })
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.batches.list')
    openai_error(e.message, type: 'server_error', code: 'internal_error', status_code: 500)
  end

  # GET /v1/batches/:id
  app.get '/v1/batches/:id' do
    require_llm!
    batch_id = params[:id]
    entry    = BATCH_MUTEX.synchronize { BATCH_STORE[batch_id] }

    unless entry
      halt 404, { 'Content-Type' => 'application/json' },
           Legion::JSON.dump({ error: { message: "Batch '#{batch_id}' not found.",
                                        type: 'invalid_request_error', code: 'batch_not_found' } })
    end

    log.debug("[llm][api][openai][batches] action=get batch_id=#{batch_id} status=#{entry[:status]}")
    content_type :json
    status 200
    Legion::JSON.dump(Batches.serialize_batch(entry))
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.batches.get')
    openai_error(e.message, type: 'server_error', code: 'internal_error', status_code: 500)
  end

  # POST /v1/batches/:id/cancel
  app.post '/v1/batches/:id/cancel' do
    require_llm!
    batch_id = params[:id]
    entry    = BATCH_MUTEX.synchronize { BATCH_STORE[batch_id] }

    unless entry
      halt 404, { 'Content-Type' => 'application/json' },
           Legion::JSON.dump({ error: { message: "Batch '#{batch_id}' not found.",
                                        type: 'invalid_request_error', code: 'batch_not_found' } })
    end

    BATCH_MUTEX.synchronize do
      BATCH_STORE[batch_id][:status]        = :cancelling
      BATCH_STORE[batch_id][:cancelling_at] = Time.now
    end

    log.debug("[llm][api][openai][batches] action=cancel batch_id=#{batch_id}")
    content_type :json
    status 200
    Legion::JSON.dump(Batches.serialize_batch(BATCH_MUTEX.synchronize { BATCH_STORE[batch_id] }))
  rescue StandardError => e
    handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.batches.cancel')
    openai_error(e.message, type: 'server_error', code: 'internal_error', status_code: 500)
  end

  log.debug('[llm][api][namespaces][openai][batches] routes registered')
rescue StandardError => e
  handle_exception(e, level: :error, handled: false, operation: 'llm.api.openai.batches.register')
end

.serialize_batch(entry) ⇒ Object 



282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
 
# File 'lib/legion/llm/api/namespaces/openai/batches.rb', line 282

def self.serialize_batch(entry)
  {
    id:                entry[:id],
    object:            'batch',
    endpoint:          entry[:endpoint],
    errors:            entry[:errors],
    input_file_id:     entry[:input_file_id],
    completion_window: entry[:completion_window],
    status:            entry[:status].to_s,
    output_file_id:    entry[:output_file_id],
    error_file_id:     entry[:error_file_id],
    created_at:        entry[:created_at]&.to_i,
    in_progress_at:    entry[:in_progress_at]&.to_i,
    expires_at:        entry[:expires_at]&.to_i,
    finalizing_at:     entry[:finalizing_at]&.to_i,
    completed_at:      entry[:completed_at]&.to_i,
    failed_at:         entry[:failed_at]&.to_i,
    expired_at:        entry[:expired_at]&.to_i,
    cancelling_at:     entry[:cancelling_at]&.to_i,
    cancelled_at:      entry[:cancelled_at]&.to_i,
    request_counts:    entry[:request_counts],
    metadata:          entry[:metadata]
  }.compact
end