lex-llm-vllm 0.2.7 → 0.2.9

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 60cd8aa03dc672912cdd86dadd1dd869bcb0cea55187c29a06d6808850cf8b58
-  data.tar.gz: 26ec677b3b731fa5ab4fa024a4a046285d8ff7550bdb8ca3de92f35c1f71733c
+  metadata.gz: c681beae79a3636380cbb8f75f3b0deb92722ee8dcfe150569944d4cd678ecd4
+  data.tar.gz: a072817f69752bde450cb67776b9b67021c680b5c744a9c05fddd6048821b871
 SHA512:
-  metadata.gz: 0b790758768552288cbcb083c8b1b7a1892d6262223633ec36e50a014b2ebc9f75bb43bc4cb451aadfbbebe82dea33cbdff86129d2a90d4873b87ed5ec288d8d
-  data.tar.gz: 0e969eba8171a0dfb830137193173129da7b051a9d84902ee9275fc5bcf8602a3c634a2b26c464e7e8b7dc17216c8d971f899c20144dcf535cdffb798743de19
+  metadata.gz: 9e4fdb96b3e7084371aa29f058072d2eb094c5872f0a10a4974c9bf2c16a6527d5e321563a7e24e0a9ee9ea52471ecf1c264a4e38e3c206930202739760d4135
+  data.tar.gz: f09e2f1c922a9466493281223371b57ff8b765850d15b9f783819551d9dc9aa1d8a36e5b40d6b9059d6dcf31998e2c6ca610a87c23c68ca70102e2e13fb5f193

data/CHANGELOG.md

@@ -1,5 +1,14 @@
 # Changelog
 
+## 0.2.9 - 2026-05-12
+
+- Route fleet actor load failures through `Legion::Logging::Helper` instead of direct warnings.
+- Add debug logging around vLLM instance discovery, fleet worker dispatch, offering construction, payload rendering, and management endpoints.
+
+## 0.2.8 - 2026-05-07
+
+- Read vLLM thinking defaults from the active provider instance config so per-instance `enable_thinking` settings affect chat payloads.
+
 ## 0.2.7 - 2026-05-07
 
 - Fix merge order in `discover_instances` so a user-supplied `tier:` in instance config is no longer clobbered by the `:direct` default.

data/lib/legion/extensions/llm/vllm/actors/fleet_worker.rb

@@ -3,7 +3,11 @@
 begin
   require 'legion/extensions/actors/subscription'
 rescue LoadError => e
-  warn(e.message) if $VERBOSE
+  require 'legion/extensions/llm/vllm'
+  unless defined?(Legion::Extensions::Actors::Subscription)
+    Legion::Extensions::Llm::Vllm.handle_exception(e, level: :warn, handled: false,
+                                                      operation: 'vllm.fleet_worker.load_actor_runtime')
+  end
 end
 
 unless defined?(Legion::Extensions::Actors::Subscription)
@@ -12,6 +16,7 @@ end
 
 require 'legion/extensions/llm/vllm'
 require 'legion/extensions/llm/fleet/provider_responder'
+require 'legion/logging'
 
 module Legion
   module Extensions
@@ -20,6 +25,8 @@ module Legion
         module Actor
           # Subscription actor for vLLM fleet request consumption.
           class FleetWorker < Legion::Extensions::Actors::Subscription
+            include Legion::Logging::Helper
+
             def runner_class
               'Legion::Extensions::Llm::Vllm::Runners::FleetWorker'
             end
@@ -33,7 +40,9 @@ module Legion
             end
 
             def enabled?
-              Legion::Extensions::Llm::Fleet::ProviderResponder.enabled_for?(Vllm.discover_instances)
+              Legion::Extensions::Llm::Fleet::ProviderResponder.enabled_for?(Vllm.discover_instances).tap do |enabled|
+                log.debug { "vLLM fleet worker enabled=#{enabled}" }
+              end
             end
           end
         end

data/lib/legion/extensions/llm/vllm/provider.rb

@@ -1,6 +1,7 @@
 # frozen_string_literal: true
 
 require 'legion/extensions/llm'
+require 'legion/logging'
 require 'uri'
 
 module Legion
@@ -94,7 +95,9 @@ module Legion
                      else
                        Array(@cached_models)
                      end
-            models.map { |model_info| offering_from_model(model_info) }
+            models.map { |model_info| offering_from_model(model_info) }.tap do |offerings|
+              log.debug { "built #{offerings.size} vLLM offering(s) live=#{live}" }
+            end
           rescue StandardError => e
             handle_exception(e, level: :warn, handled: true, operation: 'vllm.discover_offerings')
             []
@@ -106,18 +109,25 @@ module Legion
           end
 
           def reset_prefix_cache(reset_running_requests: nil, reset_external: nil)
+            log.debug do
+              "resetting vLLM prefix cache reset_running_requests=#{reset_running_requests.inspect} " \
+                "reset_external=#{reset_external.inspect}"
+            end
             connection.post(with_query(reset_prefix_cache_url, reset_running_requests:, reset_external:), {}).body
           end
 
           def reset_mm_cache
+            log.debug { 'resetting vLLM multimodal cache' }
             connection.post(reset_mm_cache_url, {}).body
           end
 
           def sleep(level: 1)
+            log.debug { "putting vLLM worker to sleep level=#{level.inspect}" }
             connection.post(with_query(sleep_url, level:), {}).body
           end
 
           def wake_up(tags: nil)
+            log.debug { "waking vLLM worker tags=#{Array(tags).inspect}" }
             query = Array(tags).map { |tag| ['tags', tag] }
             connection.post(with_query(wake_up_url, query), {}).body
           end
@@ -150,6 +160,10 @@ module Legion
             payload = super
             payload.delete(:reasoning_effort)
             payload[:chat_template_kwargs] = { enable_thinking: true } if thinking_enabled?(thinking)
+            log.debug do
+              "rendered vLLM payload model=#{model.respond_to?(:id) ? model.id : model} stream=#{stream} " \
+                "tools=#{tools.respond_to?(:size) ? tools.size : 0} thinking=#{payload.key?(:chat_template_kwargs)}"
+            end
             payload
           end
 
@@ -162,18 +176,25 @@ module Legion
           end
 
           def vllm_thinking_setting
+            instance_thinking_enabled? || global_thinking_enabled?
+          rescue StandardError => e
+            handle_exception(e, level: :debug, handled: true, operation: 'vllm.thinking_setting')
+            false
+          end
+
+          def instance_thinking_enabled?
+            return config.enable_thinking if config.respond_to?(:enable_thinking)
+
+            config.respond_to?(:[]) && config[:enable_thinking] == true
+          end
+
+          def global_thinking_enabled?
             return false unless defined?(Legion::Settings)
 
             vllm = Legion::Settings.dig(:llm, :providers, :vllm)
             return false unless vllm.is_a?(Hash)
 
-            vllm[:enable_thinking] == true ||
-              vllm['enable_thinking'] == true ||
-              vllm.dig(:instances, :default, :enable_thinking) == true ||
-              vllm.dig('instances', 'default', 'enable_thinking') == true
-          rescue StandardError => e
-            handle_exception(e, level: :debug, handled: true, operation: 'vllm.thinking_setting')
-            false
+            vllm[:enable_thinking] == true || vllm['enable_thinking'] == true
           end
 
           def parse_list_models_response(response, provider, capabilities)

data/lib/legion/extensions/llm/vllm/runners/fleet_worker.rb

@@ -2,6 +2,7 @@
 
 require 'legion/extensions/llm/fleet/provider_responder'
 require 'legion/extensions/llm/vllm'
+require 'legion/logging'
 
 module Legion
   module Extensions
@@ -10,9 +11,17 @@ module Legion
         module Runners
           # Runner entrypoint for vLLM fleet request execution.
           module FleetWorker
+            include Legion::Logging::Helper
+            extend Legion::Logging::Helper
+
             module_function
 
             def handle_fleet_request(payload, delivery: nil, properties: nil)
+              log.debug do
+                "handling vLLM fleet request request_id=#{payload_field(payload, :request_id).inspect} " \
+                  "provider_instance=#{payload_field(payload, :provider_instance).inspect} " \
+                  "operation=#{payload_field(payload, :operation).inspect}"
+              end
               Legion::Extensions::Llm::Fleet::ProviderResponder.call(
                 payload: payload,
                 provider_family: Vllm::PROVIDER_FAMILY,
@@ -22,6 +31,16 @@ module Legion
                 properties: properties
               )
             end
+
+            def payload_field(payload, key)
+              return unless payload.respond_to?(:[])
+
+              payload[key] || payload[key.to_s]
+            rescue StandardError => e
+              handle_exception(e, level: :debug, handled: true, operation: 'vllm.fleet_worker.payload_field',
+                                  field: key)
+              nil
+            end
           end
         end
       end

data/lib/legion/extensions/llm/vllm/version.rb

@@ -4,7 +4,7 @@ module Legion
   module Extensions
     module Llm
       module Vllm
-        VERSION = '0.2.7'
+        VERSION = '0.2.9'
       end
     end
   end

data/lib/legion/extensions/llm/vllm.rb

@@ -3,6 +3,7 @@
 require 'legion/extensions/llm'
 require 'legion/extensions/llm/vllm/provider'
 require 'legion/extensions/llm/vllm/version'
+require 'legion/logging'
 
 module Legion
   module Extensions
@@ -65,6 +66,7 @@ module Legion
             end
           end
 
+          log.debug { "discovered #{instances.size} vLLM instance(s): #{instances.keys.join(', ')}" }
           instances
         end
 
@@ -92,7 +94,8 @@ module Legion
           require 'uri'
           host = URI.parse(url.to_s).host.to_s.downcase
           %w[localhost 127.0.0.1 ::1].include?(host) ? :local : :direct
-        rescue URI::InvalidURIError
+        rescue URI::InvalidURIError => e
+          handle_exception(e, level: :debug, handled: true, operation: 'vllm.infer_tier_from_endpoint')
           :direct
         end
       end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-llm-vllm
 version: !ruby/object:Gem::Version
-  version: 0.2.7
+  version: 0.2.9
 platform: ruby
 authors:
 - LegionIO