Module: Legion::LLM::Fleet::Dispatcher

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

Class Method Summary collapse

Class Method Details

.build_routing_key(provider:, request_type:, model:, provider_instance: nil, context_window: nil, boundary: nil, eligibility_fingerprint: nil, routing_style: nil) ⇒ Object 



78
79
80
81
82
83
84
85
86
87
88
89
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 78

def build_routing_key(provider:, request_type:, model:, provider_instance: nil, context_window: nil, boundary: nil,
                      eligibility_fingerprint: nil, routing_style: nil)
  style = routing_style || default_routing_style
  return Lane.offering_key(instance_id: provider_instance || provider, model: model, operation: request_type) if style.to_s == 'offering_lane'

  if style.to_s == 'shared_lane'
    return Lane.routing_key(operation: request_type, model: model, context_window: context_window,
                            boundary: boundary, eligibility_fingerprint: eligibility_fingerprint)
  end

  "llm.request.#{provider}.#{request_type}.#{sanitize_model(model)}"
end

.context_window_from(options) ⇒ Object 



100
101
102
103
104
105
106
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 100

def context_window_from(options)
  limits = fetch_option(options, :limits) || {}
  fetch_option(options, :context_window) ||
    fetch_option(options, :max_context_size) ||
    fetch_option(options, :max_input_tokens) ||
    fetch_option(limits, :context_window)
end

.default_routing_style ⇒ Object 



91
92
93
94
95
96
97
98
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 91

def default_routing_style
  Legion::LLM::Settings.value(:routing, :tiers, :fleet, :routing_style) ||
    Legion::LLM::Settings.value(:routing, :fleet, :routing_style) ||
    :shared_lane
rescue StandardError => e
  handle_exception(e, level: :warn, operation: 'llm.fleet.dispatcher.default_routing_style')
  :shared_lane
end

.dispatch(model: nil, messages: nil, request: nil, message_context: {}, routing_key: nil, reply_to: nil, **opts) ⇒ Object

Backwards-compatible shim: supports old (model:, messages:) and new (request:, message_context:) callers



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
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 14

def dispatch(model: nil, messages: nil, request: nil, message_context: {}, routing_key: nil, reply_to: nil, **opts)
  log.debug "[llm][fleet][dispatcher] action=dispatch.enter model=#{model} routing_key=#{routing_key} fleet_available=#{fleet_available?}"
  return error_result('fleet_unavailable', message_context: message_context) unless fleet_available?

  # Old calling convention: build minimal params from model/messages
  if request.nil? && (model || messages)
    provider = opts[:provider] || 'ollama'
    request_type = opts[:request_type] || 'chat'
    routing_key ||= build_routing_key(provider: provider, request_type: request_type, model: model,
                                      provider_instance: opts[:provider_instance],
                                      context_window: context_window_from(opts),
                                      boundary: opts[:network_boundary],
                                      eligibility_fingerprint: opts[:eligibility_fingerprint],
                                      routing_style: opts[:routing_style])
    reply_to ||= ReplyDispatcher.agent_queue_name
    correlation_id = next_correlation_id
    future = register_response(correlation_id)
    timeout = resolve_timeout(request_type: request_type, override: opts[:timeout])
    publish_opts = opts.except(:timeout).merge(ttl: effective_ttl(opts, timeout))
    publish_result = publish_request(
      correlation_id: correlation_id,
      routing_key: routing_key, reply_to: reply_to,
      provider: provider, model: model, request_type: request_type,
      messages: messages, message_context: message_context, **publish_opts
    )
    return publish_error_result(publish_result, correlation_id, message_context: message_context) unless publish_accepted?(publish_result)

    return wait_for_response(correlation_id, timeout: timeout, message_context: message_context, future: future)
  end

  # New calling convention
  request_opts =
    if request.respond_to?(:to_h)
      request.to_h.transform_keys(&:to_sym)
    else
      {}
    end
  request_opts = request_opts.merge(opts)

  provider = request_opts[:provider] || 'ollama'
  request_type = request_opts[:request_type] || 'chat'
  model = request_opts[:model]
  routing_key ||= build_routing_key(provider: provider, request_type: request_type, model: model,
                                    provider_instance: request_opts[:provider_instance],
                                    context_window: context_window_from(request_opts),
                                    boundary: request_opts[:network_boundary],
                                    eligibility_fingerprint: request_opts[:eligibility_fingerprint],
                                    routing_style: request_opts[:routing_style] || opts[:routing_style])
  reply_to ||= ReplyDispatcher.agent_queue_name
  correlation_id = next_correlation_id
  future = register_response(correlation_id)
  timeout = resolve_timeout(request_type: request_type, override: request_opts[:timeout] || opts[:timeout])
  request_opts[:ttl] = effective_ttl(request_opts, timeout)
  publish_result = publish_request(
    correlation_id: correlation_id,
    routing_key: routing_key, reply_to: reply_to,
    provider: provider, model: model, request_type: request_type,
    message_context: message_context, **request_opts.except(:provider, :model, :request_type, :timeout)
  )
  return publish_error_result(publish_result, correlation_id, message_context: message_context) unless publish_accepted?(publish_result)

  wait_for_response(correlation_id, timeout: timeout, message_context: message_context, future: future)
end

.effective_ttl(options, timeout) ⇒ Object 



108
109
110
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 108

def effective_ttl(options, timeout)
  fetch_option(options, :ttl) || fetch_option(options, :expiration_seconds) || timeout
end

.error_result(reason, message_context: {}) ⇒ Object 



224
225
226
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 224

def error_result(reason, message_context: {})
  { success: false, error: reason, message_context: message_context }
end

.fetch_option(hash, key) ⇒ Object 



112
113
114
115
116
117
118
119
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 112

def fetch_option(hash, key)
  return nil unless hash.respond_to?(:key?)

  string_key = key.to_s
  return hash[string_key] if hash.key?(string_key)

  hash[key] if hash.key?(key)
end

.fleet_available? ⇒ Boolean

Returns:

  • (Boolean) 


133
134
135
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 133

def fleet_available?
  transport_ready? && fleet_enabled?
end

.fleet_enabled? ⇒ Boolean

Returns:

  • (Boolean) 


141
142
143
144
145
146
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 141

def fleet_enabled?
  routing = Legion::LLM::Settings.value(:routing, default: {})
  return true unless routing.is_a?(Hash)

  fetch_option(routing, :use_fleet) != false
end

.nested_fetch(hash, *keys) ⇒ Object 



121
122
123
124
125
126
127
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 121

def nested_fetch(hash, *keys)
  keys.reduce(hash) do |current, key|
    return nil unless current.respond_to?(:key?)

    fetch_option(current, key)
  end
end

.next_correlation_id ⇒ Object 



161
162
163
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 161

def next_correlation_id
  "req_#{SecureRandom.uuid}"
end

.publish_accepted?(publish_result) ⇒ Boolean

Returns:

  • (Boolean) 


181
182
183
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 181

def publish_accepted?(publish_result)
  publish_result.is_a?(Hash) ? publish_result[:accepted] == true : true
end

.publish_error_result(publish_result, correlation_id, message_context: {}) ⇒ Object 



185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 185

def publish_error_result(publish_result, correlation_id, message_context: {})
  ReplyDispatcher.deregister(correlation_id)
  status = publish_result.is_a?(Hash) ? publish_result[:status]&.to_sym : :failed
  error = case status
          when :unroutable
            'no_fleet_queue'
          when :nacked
            'fleet_backpressure'
          when :confirm_timeout
            'fleet_publish_timeout'
          else
            'fleet_publish_failed'
          end
  {
    success:         false,
    error:           error,
    publish_status:  status,
    correlation_id:  correlation_id,
    message_context: message_context
  }
end

.publish_request(correlation_id: next_correlation_id, **opts) ⇒ Object 



169
170
171
172
173
174
175
176
177
178
179
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 169

def publish_request(correlation_id: next_correlation_id, **opts)
  opts[:fleet_correlation_id] = correlation_id
  log.debug("[llm][fleet][dispatcher] action=publish_request correlation_id=#{correlation_id} routing_key=#{opts[:routing_key]}")

  if defined?(Legion::LLM::Transport::Messages::FleetRequest)
    Legion::LLM::Transport::Messages::FleetRequest.new(**opts).publish
  else
    log.debug('[llm][fleet][dispatcher] action=skip_publish reason=transport_not_loaded')
    { accepted: false, status: :failed, error: 'transport_not_loaded' }
  end
end

.register_response(correlation_id, expected = {}) ⇒ Object 



165
166
167
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 165

def register_response(correlation_id, expected = {})
  ReplyDispatcher.register(correlation_id, expected: expected)
end

.resolve_timeout(request_type: :default, override: nil) ⇒ Object 



148
149
150
151
152
153
154
155
156
157
158
159
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 148

def resolve_timeout(request_type: :default, override: nil)
  return override if override

  fleet = Legion::LLM::Settings.value(:routing, :tiers, :fleet) ||
          Legion::LLM::Settings.value(:routing, :fleet) ||
          {}
  timeouts = fetch_option(fleet, :timeouts) || {}
  fetch_option(timeouts, request_type.to_sym) || fetch_option(fleet, :timeout_seconds) || 30
rescue StandardError => e
  handle_exception(e, level: :warn, operation: 'llm.fleet.dispatcher.resolve_timeout')
  30
end

.sanitize_model(model) ⇒ Object 



129
130
131
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 129

def sanitize_model(model)
  model.to_s.gsub(':', '.')
end

.timeout_result(correlation_id, timeout, message_context: {}) ⇒ Object 



219
220
221
222
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 219

def timeout_result(correlation_id, timeout, message_context: {})
  { success: false, error: 'fleet_timeout', correlation_id: correlation_id,
    timeout: timeout, message_context: message_context }
end

.transport_ready? ⇒ Boolean

Returns:

  • (Boolean) 


137
138
139
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 137

def transport_ready?
  Legion::LLM::Settings.transport_connected?
end

.wait_for_response(correlation_id, timeout:, message_context: {}, future: nil) ⇒ Object 



207
208
209
210
211
212
213
214
215
216
217
 
# File 'lib/legion/llm/fleet/dispatcher.rb', line 207

def wait_for_response(correlation_id, timeout:, message_context: {}, future: nil)
  log.debug "[llm][fleet][dispatcher] action=wait_for_response correlation_id=#{correlation_id} timeout=#{timeout}"
  future ||= ReplyDispatcher.register(correlation_id)
  result = future.value!(timeout)
  result || timeout_result(correlation_id, timeout, message_context: message_context)
rescue Concurrent::CancelledOperationError => e
  handle_exception(e, level: :debug, handled: true, operation: 'llm.fleet.dispatcher.wait_cancelled')
  timeout_result(correlation_id, timeout, message_context: message_context)
ensure
  ReplyDispatcher.deregister(correlation_id)
end