lex-llm-gateway 0.2.9 → 0.2.12

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: bce7fb746d14ff63db46cb4fc87613a5a0834c61e5e067dc4e5bf6122dbd251a
-  data.tar.gz: 2d7c0d14bef999fece93d633dd74c700d2594e76d1aae216a2afd3b51f13a534
+  metadata.gz: 23b44bc07cb28fc0e364b119984909f1d8becba720ea547b1255a8512188dec7
+  data.tar.gz: a4aae0ecf4e1b053405f74628af518156ae2ffc3d0678b4f920771a9e6a68ee6
 SHA512:
-  metadata.gz: 488014e8667b9782d18a56dee95250ce8869aff2f9d9e0a5254d77f3d6a5ffd9a1cea802c66766fa50da1e2fa0b5801fb6b1b17b460dfb39305ddd3eaa2aeb81
-  data.tar.gz: a65be2e6996f18b98cbdafb3f1e6145671a6eb4f37f63e7b6e500e272d8d16ba26e5414f32489cd5569c16c6881c17dab601833f1eae5edf2516c177077e54f3
+  metadata.gz: 1cf5d4fe5d3f0728dee2dfa04019f62c7e2fd799527df6a79b069f9aabf4ef7cff44674fa0405406e200705faaa03eff282c60ab9b0ca3db25a5f93fa7bb6c1f
+  data.tar.gz: c01a0edb556ccf0346b111b6187f05e4083dde3c52ecbc43707fbb720c40a73501a76a98c85e37389512fb2db0184d631ba98edbecab77b456cd1af1401b882d

data/CHANGELOG.md

@@ -1,5 +1,22 @@
 # Changelog
 
+## [0.2.12] - 2026-03-24
+
+### Fixed
+- fix `uninitialized constant Legion::Extensions::Llm::Gateway` when `ensure_namespace` pre-creates `Llm` as empty module before gem loads — unconditionally `remove_const(:Llm)` then reassign alias to `LLM`
+
+## [0.2.11] - 2026-03-23
+
+### Changed
+- Add `caller:` identity to all `Legion::LLM` call sites: 3 pipeline-path calls in `Runners::Inference` (`chat`, `embed`, `structured`) and 4 fleet worker calls in `Runners::FleetHandler` (`call_chat` ×2, `call_structured`, `call_embed`) with `extension: 'lex-llm-gateway'` and `operation: 'inference'`/`'fleet'`
+
+## [0.2.10] - 2026-03-23
+
+### Added
+- ProviderStats runner: health_report, provider_detail, circuit_summary for LLM provider observability
+- Reports circuit breaker state, routing adjustment, and health status per provider
+- 5 specs covering report generation, detail, circuit summary, and fallback behavior
+
 ## [0.2.9] - 2026-03-23
 
 ### Added

data/lib/legion/extensions/llm/gateway/runners/fleet_handler.rb

@@ -48,9 +48,11 @@ module Legion
             def call_chat(payload)
               messages = payload[:messages]
               if messages.is_a?(Array) && messages.size > 1
-                Legion::LLM.chat(model: payload[:model], messages: messages)
+                Legion::LLM.chat(model: payload[:model], messages: messages,
+                                 caller: { extension: 'lex-llm-gateway', operation: 'fleet' })
               else
-                Legion::LLM.chat(model: payload[:model], message: messages&.dig(0, :content))
+                Legion::LLM.chat(model: payload[:model], message: messages&.dig(0, :content),
+                                 caller: { extension: 'lex-llm-gateway', operation: 'fleet' })
               end
             end
 
@@ -58,13 +60,15 @@ module Legion
               Legion::LLM.structured(
                 model: payload[:model],
                 messages: payload[:messages],
-                schema: payload[:schema]
+                schema: payload[:schema],
+                caller: { extension: 'lex-llm-gateway', operation: 'fleet' }
               )
             end
 
             def call_embed(payload)
               text = payload[:text] || payload.dig(:messages, 0, :content)
-              Legion::LLM.embed(model: payload[:model], text: text)
+              Legion::LLM.embed(model: payload[:model], text: text,
+                                caller: { extension: 'lex-llm-gateway', operation: 'fleet' })
             end
 
             def build_response(correlation_id, response)

data/lib/legion/extensions/llm/gateway/runners/inference.rb

@@ -5,10 +5,16 @@ module Legion
     module LLM
       module Gateway
         module Runners
-          module Inference
+          module Inference # rubocop:disable Metrics/ModuleLength
             module_function
 
-            def chat(model: nil, provider: nil, **opts)
+            def chat(model: nil, provider: nil, **opts) # rubocop:disable Metrics/MethodLength
+              if pipeline_available?
+                log_deprecation(:chat)
+                return Legion::LLM.chat(model: model, provider: provider,
+                                        caller: { extension: 'lex-llm-gateway', operation: 'inference' }, **opts)
+              end
+
               start_ms = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond)
               response = dispatch_chat(model: model, provider: provider, **opts)
               elapsed_ms = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond) - start_ms
@@ -17,7 +23,13 @@ module Legion
               response
             end
 
-            def embed(text: nil, model: nil, provider: nil, **)
+            def embed(text: nil, model: nil, provider: nil, **) # rubocop:disable Metrics/MethodLength
+              if pipeline_available?
+                log_deprecation(:embed)
+                return Legion::LLM.embed(text, model: model, provider: provider,
+                                               caller: { extension: 'lex-llm-gateway', operation: 'inference' }, **)
+              end
+
               start_ms = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond)
               response = dispatch_embed(text: text, model: model, provider: provider, **)
               elapsed_ms = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond) - start_ms
@@ -26,7 +38,14 @@ module Legion
               response
             end
 
-            def structured(messages: nil, schema: nil, model: nil, provider: nil, **)
+            def structured(messages: nil, schema: nil, model: nil, provider: nil, **) # rubocop:disable Metrics/MethodLength
+              if pipeline_available?
+                log_deprecation(:structured)
+                return Legion::LLM.structured(messages: messages, schema: schema, model: model,
+                                              provider: provider,
+                                              caller: { extension: 'lex-llm-gateway', operation: 'inference' }, **)
+              end
+
               start_ms = ::Process.clock_gettime(::Process::CLOCK_MONOTONIC, :millisecond)
               response = dispatch_structured(messages: messages, schema: schema, model: model,
                                              provider: provider, **)
@@ -36,6 +55,21 @@ module Legion
               response
             end
 
+            def pipeline_available?
+              defined?(Legion::LLM::Pipeline::Executor) &&
+                defined?(Legion::LLM) &&
+                Legion::LLM.respond_to?(:pipeline_enabled?) &&
+                Legion::LLM.pipeline_enabled?
+            end
+
+            def log_deprecation(method)
+              return unless defined?(Legion::Logging)
+
+              Legion::Logging.warn(
+                "lex-llm-gateway is deprecated for #{method}, use Legion::LLM.#{method} directly"
+              )
+            end
+
             def dispatch_chat(message: nil, messages: nil, model: nil, provider: nil, **opts)
               tier = opts[:tier]
               Legion::Logging.debug "[Gateway::Inference] dispatch_chat tier=#{tier}" if defined?(Legion::Logging)

data/lib/legion/extensions/llm/gateway/runners/provider_stats.rb

@@ -0,0 +1,78 @@
+# frozen_string_literal: true
+
+module Legion
+  module Extensions
+    module LLM
+      module Gateway
+        module Runners
+          module ProviderStats
+            module_function
+
+            def health_report
+              return unavailable_result unless router_available?
+
+              tracker = health_tracker
+              providers = known_providers
+              providers.map { |name| provider_entry(name, tracker) }
+            end
+
+            def provider_detail(provider:)
+              return unavailable_result unless router_available?
+
+              provider_entry(provider, health_tracker)
+            end
+
+            def circuit_summary
+              return unavailable_result unless router_available?
+
+              tracker = health_tracker
+              providers = known_providers
+              {
+                total: providers.size,
+                closed: providers.count { |p| tracker.circuit_state(p) == :closed },
+                open: providers.count { |p| tracker.circuit_state(p) == :open },
+                half_open: providers.count { |p| tracker.circuit_state(p) == :half_open }
+              }
+            end
+
+            def router_available?
+              defined?(Legion::LLM::Router) &&
+                Legion::LLM::Router.respond_to?(:routing_enabled?) &&
+                Legion::LLM::Router.routing_enabled?
+            end
+
+            def health_tracker
+              Legion::LLM::Router.health_tracker
+            end
+
+            def known_providers
+              return [] unless defined?(Legion::Settings)
+
+              providers = Legion::Settings.dig(:llm, :providers)
+              return [] unless providers.is_a?(Hash)
+
+              providers.select { |_, cfg| cfg.is_a?(Hash) && cfg[:enabled] }
+                       .keys
+                       .map(&:to_sym)
+            rescue StandardError
+              []
+            end
+
+            def provider_entry(name, tracker)
+              {
+                provider: name.to_s,
+                circuit: tracker.circuit_state(name).to_s,
+                adjustment: tracker.adjustment(name),
+                healthy: tracker.circuit_state(name) != :open
+              }
+            end
+
+            def unavailable_result
+              { success: false, error: 'router_not_available' }
+            end
+          end
+        end
+      end
+    end
+  end
+end

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

@@ -4,7 +4,7 @@ module Legion
   module Extensions
     module LLM
       module Gateway
-        VERSION = '0.2.9'
+        VERSION = '0.2.12'
       end
     end
   end

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

@@ -9,6 +9,7 @@ module Legion
         extend Legion::Extensions::Core if Legion::Extensions.const_defined?(:Core)
       end
     end
-    Llm = LLM unless const_defined?(:Llm, false)
+    remove_const(:Llm) if const_defined?(:Llm, false)
+    Llm = LLM
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: lex-llm-gateway
 version: !ruby/object:Gem::Version
-  version: 0.2.9
+  version: 0.2.12
 platform: ruby
 authors:
 - Esity
@@ -191,6 +191,7 @@ files:
 - lib/legion/extensions/llm/gateway/runners/inference.rb
 - lib/legion/extensions/llm/gateway/runners/metering.rb
 - lib/legion/extensions/llm/gateway/runners/metering_writer.rb
+- lib/legion/extensions/llm/gateway/runners/provider_stats.rb
 - lib/legion/extensions/llm/gateway/runners/usage_reporter.rb
 - lib/legion/extensions/llm/gateway/transport/exchanges/inference.rb
 - lib/legion/extensions/llm/gateway/transport/exchanges/metering.rb